A MAJOR LEAP FORWARD

IN PHENOTYPING

'HWHUPLQHWKHIXQFWLRQDOLPSDFWRI JHQHVRQPHWDEROLVPLQDQKRXU

XFp Cell Energy Phenotype Test Kit

5HTXHVWDFREE poster and energy phenotype analysis

RI\RXUFHOOVDWZZZVHDKRUVHELRFRPSKHQR

Mitochondrial Respiration

Using as few as 15,000 live cells and an XFp Analyzer,

WKH;)&HOO(QHUJ\3KHQRW\SH7HVWLGHQWLHV\RXUFHOOV
metabolic phenotype as well as their metabolic potential
WKHXWLOL]DWLRQRIPLWRFKRQGULDOUHVSLUDWLRQDQGJO\FRO\VLV

Energetic

Aerobic

Metab

olic P

otenti

al
Stressed
Phenotype

Baseline
Phenotype

Quiescent

Glycolytic
Glycolysis

Focused on antibodies for 40 years.

Not on advertising.

Bethyl Laboratories, Inc. has been dedicated to supporting scientific discovery through
its qualified antibody products and custom antibody services since its founding in 1972.
Every antibody that Bethyl sells has been manufactured to exacting standards at its sole
location in Montgomery, Texas, and has been validated in-house by Bethyls team of
scientists. Antibodies are tested across a range of applications including western blot,
immunoprecipitation, immunohistochemistry, immunocytochemistry, ChIP, proximity
ligation assay and ELISA.
Currently, Bethyls portfolio consists of over 7,150 catalog products; offering close to
5,700 primary antibodies targeting over 2,700 proteins and 1,450 secondary antibodies
raised againstimmunoglobulins from over 25 species. Trial sizes are available for over
4,000 antibodies targeting more than 2,350 protein targets. They are conveniently
priced at $50 and serve as an opportunity to discover for yourself why our antibodies
arereally good.
For really good antibodies, visit bethyl.com/trialsize

Terms & Conditions: $50 pricing for US customers only; international customers please
contact your distributor for details. Trial sizes (catalog # ending in -T) are not available
for all antibodies and existing promotions or discounts do not apply.
2015 Bethyl Laboratories, Inc. All rights reserved.

SAY GOOD-BYE
TO THE DARKROOM

Introducing the ZOE Fluorescent Cell Imager.

No darkroom, no training, no overwhelming user interface.
Combining brightfield capabilities with multichannel fluorescence, this cell
imager is both affordable and easy to use your perfect solution for routine
cell culture and imaging applications.

Learn more at bio-rad.com/info/newzoe

We make really good antibodies.

Not really good ads.

For really good antibodies, visit bethyl.com/trialsize

2015 Bethyl Laboratories, Inc. All rights reserved.

Foreword
We are so pleased to present the Best of Cell Reports 2015. In year 4 we saw some major
changes, including moving to weekly publication. From tales of very long-lived whales
to an analysis of what might have improved last years influenza vaccine, we published
top-notch content from across the life sciences. That made it all the more difficult to
put together this issue, and there were some truly tough choices to make. With the final
content in this supplement, we try to represent the breadth of biology covered by Cell
Reports and have based our choices on reader downloads, altmetric scores, and citation
information. We would urge you (once you have perused the Best of of course!) to browse
our table of contents to explore the full range of Open Access papers we publish in your
field and beyond.
We are really looking forward to 2016. In the last year, Cell Reports has been globetrotting.
We organized the Ubiquitous Ubiquitin Signaling Lablinks in Cambridge, UK and the RNA
in the Nervous System meeting in New York, USA. In November 2015, we cohosted a
major symposium in Singapore on human genomics and organized a day-long cancer and
metabolism meeting in Philadelphia. We plan to be out and about in 2016 and will hopefully
visit your neighborhood, too!
Last, but certainly not least, we want to thank the researchers who are the authors,
reviewers, readers, and editorial and advisory board members that make Cell Reports
possible. With an expanding editorial and production team internally and editorial board
advisors externally, we are so excited about 2016. But for now, kick back, grab a coffee,
tea, or beverage, and enjoy the following selection of papers.
Finally, we are grateful for the generosity of our sponsors, who helped to make this reprint
collection possible.

For information for the Best of Series, please contact:

Jonathan Christison
Program Director, Best of Cell Press
e: [email protected]
p: 617-397-2893
t: @cellPressBiz

The MojoSort Magnetic Cell Separation System is designed for the

separation of target populations using positive or negative selection.
MojoSort nanoparticles deliver excellent purity and yield at an
unmatched, affordable price. Magnetically sorted cells can be used
for multiple downstream applications.
MojoSort advantages:

Small and large test size formats to meet research needs

Robust performance

Preserved cell functionality after sorting

Excellent price

Add some Mojo to your experiment and explore the possibilities!

Before Isolation

After MojoSort Isolation

After Competitor Isolation

A suspension of single cells from pooled C57BL/6 mouse spleen and lymph nodes was prepared to isolate CD4+ T cells using
the MojoSort Mouse CD4 T Cell Isolation Kit. Cells were stained with PE anti-mouse CD4 (clone RM4-4), APC anti-mouse CD3
(145-2C11), and 7-AAD. Grateful Dead cells were excluded from the analysis.

To learn more, visit: biolegend.com/mojosort

Toll-Free Tel: (US & Canada): 1.877.BIOLEGEND (246.5343)
Tel: 858.768.5800

biolegend.com
08-0052-23

Best of 2015
Reports
Leukemia-Associated Somatic Mutations Drive
Distinct Patterns of Age-Related Clonal Hemopoiesis

Thomas McKerrell, Naomi Park, Thaidy Moreno, Carolyn S.

Grove, Hannes Ponstingl, Jonathan Stephens, Understanding
Society Scientic Group, Charles Crawley, Jenny Craig,
Mike A. Scott, Clare Hodkinson, Joanna Baxter, Roland
Rad, Duncan R. Forsyth, Michael A. Quail, Eleftheria Zeggini,
Willem Ouwehand, Ignacio Varela, and George S. Vassiliou

Accelerating Novel Candidate Gene Discovery in

Neurogenetic Disorders via Whole-Exome Sequencing
of Prescreened Multiplex Consanguineous Families

Anas M. Alazami, Nisha Patel, Hanan E. Shamseldin, Shamsa

Anazi, Mohammed S. Al-Dosari, Fatema Alzahrani, Hadia
Hijazi, Muneera Alshammari, Mohammed A. Aldahmesh,
Mustafa A. Salih, Eissa Faqeih, Amal Alhashem, Fahad A.
Bashiri, Mohammed Al-Owain, Amal Y. Kentab, Sameera
Sogaty, Saeed Al Tala, Mohamad-Hani Temsah, Maha
Tulbah, Rasha F. Aljelaify, Saad A. Alshahwan, Mohammed
Zain Seidahmed, Adnan A. Alhadid, Hesham Aldhalaan,
Fatema AlQallaf, Wesam Kurdi, Majid Alfadhel, Zainab Babay,
Mohammad Alsogheer, Namik Kaya, Zuhair N. Al-Hassnan,
Ghada M.H. Abdel-Salam, Nouriya Al-Sannaa, Fuad Al
Mutairi, Heba Y. El Khashab, Saeed Bohlega, Xiaofei Jia,
Henry C. Nguyen, Rakad Hammami, Nouran Adly, Jawahir
Y. Mohamed, Firdous Abdulwahab, Niema Ibrahim, Ewa A.
Naim, Banan Al-Younes, Brian F. Meyer, Mais Hashem, Ranad
Shaheen, Yong Xiong, Mohamed Abouelhoda, Abdulrahman
A. Aldeeri, Dorota M. Monies, and Fowzan S. Alkuraya

Analysis of Intron Sequences Reveals Hallmarks of

Circular RNA Biogenesis in Animals

Andranik Ivanov, Sebastian Memczak, Emanuel Wyler,

Francesca Torti, Hagit T. Porath, Marta R. Orejuela, Michael
Piechotta, Erez Y. Levanon, Markus Landthaler, Christoph
Dieterich, and Nikolaus Rajewsky

ThermoMouse: An In Vivo Model to Identify Modulators

of UCP1 Expression in Brown Adipose Tissue

Andrea Galmozzi, Si B. Sonne, Svetlana Altshuler-Keylin,

Yutaka Hasegawa, Kosaku Shinoda, Ineke H.N. Luijten, Jae
Won Chang, Louis Z. Sharp, Benjamin F. Cravatt, Enrique
Saez, and Shingo Kajimura

TRIM28 Represses Transcription of Endogenous

Retroviruses in Neural Progenitor Cells

Liana Fasching, Adamandia Kapopoulou, Rohit Sachdeva,

Rebecca Petri, Marie E. Jnsson, Christian Mnne, Priscilla
Turelli, Patric Jern, Florence Cammas, Didier Trono, and
Johan Jakobsson

(continued)

Articles
Comparative Hi-C Reveals that CTCF Underlies
Evolution of Chromosomal Domain Architecture

Matteo Vietri Rudan, Christopher Barrington, Stephen

Henderson, Christina Ernst, Duncan T. Odom, Amos Tanay,
and Suzana Hadjur

Heterogeneities in Nanog Expression Drive Stable

Commitment to Pluripotency in the Mouse Blastocyst

Panagiotis Xenopoulos, Minjung Kang, Alberto Puliato,

Stefano Di Talia, and Anna-Katerina Hadjantonakis

Direct Activation of STING in the Tumor

Microenvironment Leads to Potent and Systemic
Tumor Regression and Immunity

Leticia Corrales, Laura Hix Glickman, Sarah M. McWhirter,

David B. Kanne, Kelsey E. Sivick, George E. Katibah,
Seng-Ryong Woo, Edward Lemmens, Tamara Banda,
Justin J. Leong, Ken Metchette, Thomas W. Dubensky, Jr.,
and Thomas F. Gajewski

Manipulation of the Quorum Sensing Signal AI-2

Affects the Antibiotic-Treated Gut Microbiota

Jessica Ann Thompson, Rita Almeida Oliveira, Ana Djukovic,

Carles Ubeda, and Karina Bivar Xavier

Resources
Mapping Social Behavior-Induced Brain Activation
at Cellular Resolution in the Mouse

Yongsoo Kim, Kannan Umadevi Venkataraju, Kith Pradhan,

Carolin Mende, Julian Taranda, Srinivas C. Turaga, Ignacio
Arganda-Carreras, Lydia Ng, Michael J. Hawrylycz, Kathleen
S. Rockland, H. Sebastian Seung, and Pavel Osten

Insights into the Evolution of Longevity from the

Bowhead Whale Genome

Michael Keane, Jeremy Semeiks, Andrew E. Webb, Yang

I. Li, Vctor Quesada, Thomas Craig, Lone Bruhn Madsen,
Sipko van Dam, David Brawand, Patrcia I. Marques, Pawel
Michalak, Lin Kang, Jong Bhak, Hyung-Soon Yim, Nick V.
Grishin, Nynne Hjort Nielsen, Mads Peter Heide-Jrgensen,
Elias M. Oziolor, Cole W. Matson, George M. Church, Gary
W. Stuart, John C. Patton, J. Craig George, Robert Suydam,
Knud Larsen, Carlos Lpez-Otn, Mary J. OConnell, John W.
Bickham, Bo Thomsen, and Joo Pedro de Magalhes

IN VIVO
IMAGING
PROBES
Figure 1

Detect apoptosis & caspase activity in cancer,

embryonic development & ischemic conditions.
In vivo imaging probes bring greater insight to your
preclinical research. CAS-MAP imaging probes
allow you to distinguish between apoptotic and
healthy cells in vivo.
In vivo imaging probes are injected IV. The probes
are cell-permeant and bind to active caspases while
unbound probe is removed via the circulatory system. Visualize using live animal imaging or harvest
tissue for ex vivo analysis by microscopy or FACS.

Translate from cell to animal

using in vivo imaging probes.

Download application note:

vergentbio.com/invivo
844.803.0346

Figure 2
Figure 2
Figure 1: Apoptotic cells are labeled green
using CAS-MAP probes in an in vivo study
of postnatal development. Figure 2: Apoptotic COLO205 tumor cells are labeled in
response to treatment.

Whats in Your Sample?

Find Your Answers!

rndsystems.com/
Immunoassays

Data You Can Trust

International Corporation

MHC Class I & II Tetramers MHC Monomers CD1d Tetramers

MBL has been providing tetramer reagents for over
a decade and has recently extended its portfolio
through the acquisition of the Beckman Coulter
iTAg MHC Tetramer product line. MBL International
is your source for quality reagents for detection
QHCPVKIGPURGEKE6EGNNUCPF0-6EGNNU
  +PETGCUGFURGEKEKV[QHRTQRTKGVCT[DOWVCVKQP
  9KFGUGNGEVKQPQHCNNGNGU*WOCP/QWUG
4JGUWU/CECSWG%JKEMGP
  +FGPVKECVKQPQH#IURGEKE%&+CPF%&+ T cells
  #
 UUC[GZKDKNKV[WUKPIDKQVKP[NCVGF/*%OQPQOGTU
  &
 GVGEVKQPQH0-6EGNNUWUKPI%&FVGVTCOGTU

Selective detection of
CPVKIGPURGEKE6EGNNU

HLA-A*02:01 CMV pp65 Tetramer

%/8PGICVKXGUCORNG

%/8RQUKVKXGUCORNG

5RGEKE6%4

6GVTCOGT2'

DOWVCVKQP

6GVTCOGT2'

%&

*.#RGRVKFG%QORNGZ
%&(+6%

%QPVCEV/$.+PVGTPCVKQPCNVQFC[VQNGCTPOQTGCDQWV
QWTRTQFWEVUCPFJQYYGECPJGNR[QWFKUEQXGTYC[U
VQOCMG[QWTUEKGPEGOQTGGHEKGPVCPFGHHGEVKXG

%&(+6%

mblintl.com 800-200-5459 781-939-6963

life. science. discovery.

MC-RUO-020

For Research Use Only. Not for use in diagnostic procedures.

2015 MBL International Corporation. All rights reserved.

SIMPLE. EFFECTIVE.
AFFORDABLE.

CELL-BASED
VIABILITY

REPORTER

TOXICITY

METABOLISM

PROLIFERATION
OXYGEN CONSUMPTION

PROTEIN SYNTHESIS
PHAGOCYTOSIS & AUTOPHAGY

BIOCHEMICALS ASSAY KITS ANTIBODIES

PROTEINS RESEARCH SERVICES
Visit www.caymanchem.com for more information

The BMG LABTECH All Stars

Innovative, high-performance microplate readers for all assay needs

CLARIOstar

PHERAstar FSX

Omega series

SPECTROstar Nano

The most sensitive

monochromator-based
microplate reader.
Equipped with revolutionary
LVF monochromatorsTM,
it is the perfect reader for
exible assay development.

The new gold standard for

High Throughput Screening.
The PHERAstar FSX is the
new reference multi-mode
plate reader, combining
highest sensitivity with the
fastest read times.

Upgradeable single to
multi-mode lter-based
microplate readers.
The Omega series offers
a combination of exibility
and performance for any
life science application.

Microplate reader with

ultra-fast detection of
UV/Vis absorbance.
Spectra from 220 - 1000 nm
are detected in less than
one second per sample in
microplates and cuvettes.

Find all our microplate readers on www.bmglabtech.com

EXQUISITE
SENSITIVITY.

HARNESS THE POWER OF ADVANCED

SENSITIVITY AND RESOLUTION.
The CytoFLEX Flow Cytometer has the dynamic range to see dim and
bright populations in the same sample. And with 15 parameters,
including 13 channels for fluorescent detection, heterogeneous
populations can be defined with exquisite clarity.
Learn more at goo.gl/Ew8KYw

2015 Beckman Coulter Life Sciences. All rights reserved. CytoFLEX is for research use only. Class 1 Laser Product. Not for use in diagnostic
purposes. CytoFLEX and CytExpert are trademarks of Xitogen Technologies (Suzhou), Inc., a Beckman Coulter company. Beckman Coulter,
the stylized logo and fast track to success are trademarks of Beckman Coulter, Inc. Beckman Coulter and the stylized logo are registered in
the USPTO. All other trademarks are the property of their respective owners.

Are you
still injecting?
Focus on your research instead, and let
ALZET Osmotic Pumps do the dosing for you.
ALZET pumps are a superior alternative to repetitive
injections and other dosing methods that require
frequent animal handling. These fully implantable
pumps provide continuous and precise administration,
for up to 6 weeks with a single pump, to unrestrained
lab animals as small as mice. ALZET pumps are
economical and easy to use by research personnel.
Connection to a catheter enables direct delivery to
vessels, cerebral ventricles, and other target sites.
Learn more at alzet.com.
Now available: iPRECIO Pumps
Programmable Rellable
Implantable
Small size for mice and rats
Learn more at www.alzet.com/iprecio

2015 DURECT Corporation. ALZET is a trademark of DURECT Corporation.

From the moment when

the pieces fell into place,
and you saw the big picture,
through late nights reviewing literature,
designing procedures,
testing hypotheses
nothing fell through the cracks,
because too much was at stake.
The work you did then
advanced your field,
and your career.
We are proud to have been there,
protecting your work, and you,
with the very best in air containment,
contamination control, and
controlled environment technology.
Now, as your work gives rise
to new discoveries

We cant wait to see

what you do next.
bakerco.com
Biological Safety Cabinets Clean Benches Fume Hoods CO2 Incubators Hypoxia & Anaerobic Workstations

HUMANIZED
TM

NSG MICE

SOPHISTICATED. INNOVATIVE. VERSATILE.

JAX humanized NSGTM is the platform of
choice for studying immuno-oncology,
viral host-pathogen interactions, or the
development of novel drug therapies.

START YOUR
STUDY TODAY!
Expect delivery of your order
within 7-10 business days.

1-800-422-6423 (US, Canada & Puerto Rico)

1-207-288-5845 (from any location)
jax.org/humanized-mice

VISIT SLAS2016.ORG
FOR FULL PROGRAM
INFORMATION.

SAN DIEGO

CONVENTION CENTER

JANUARY

23-27, 2016
SLAS2016.ORG

Recycling plastic
is fantastic!
Rainin TerraRack
reduces tip rack waste.
Request your free
sample today!
mt.com/rainin-tr

Because the future is in your hands

TALK TO HENRIK

Henrik is our R&D Manager

MADE IN SWEDEN

PrecisA Monoclonals are precise, accurate and targeted. The stringent production process and
characterization procedure ensures premium performance in approved applications, with dened
specicity, multiplexing opportunities, secured continuity and stable supply.
If youre in need of an antibody to target a protein we dont list in our online catalog you might be
interested in our antibody co-development program. Learn more today at atlasantibodies.com/talktohenrikj

Cell Reports

Report
Leukemia-Associated Somatic Mutations Drive
Distinct Patterns of Age-Related
Clonal Hemopoiesis
Thomas McKerrell,1,13 Naomi Park,2,13 Thaidy Moreno,3 Carolyn S. Grove,1 Hannes Ponstingl,1 Jonathan Stephens,4,5
Understanding Society Scientic Group,6 Charles Crawley,7 Jenny Craig,7 Mike A. Scott,7 Clare Hodkinson,4,8
Joanna Baxter,4,8 Roland Rad,9,10 Duncan R. Forsyth,11 Michael A. Quail,2 Eleftheria Zeggini,12 Willem Ouwehand,4,5,12
Ignacio Varela,3 and George S. Vassiliou1,4,7,*
1Haematological

Cancer Genetics, Wellcome Trust Sanger Institute, Cambridge CB10 1SA, UK

Research Group, Wellcome Trust Sanger Institute, Cambridge CB10 1SA, UK
3Instituto de Biomedicina y Biotecnologa de Cantabria (CSIC-UC-Sodercan), Departamento de Biologa Molecular, Universidad de
Cantabria, 39011 Santander, Spain
4Department of Haematology, Cambridge Biomedical Campus, University of Cambridge, Cambridge CB2 0XY, UK
5NHS Blood and Transplant, Cambridge Biomedical Campus, Cambridge CB2 0PT, UK
6Institute for Social and Economic Research, University of Essex, Colchester CO4 3SQ, UK
7Department of Haematology, Cambridge University Hospitals NHS Trust, Cambridge CB2 0QQ, UK
8Cambridge Blood and Stem Cell Biobank, Department of Haematology, University of Cambridge, Cambridge CB2 0XY, UK
9Department of Medicine II, Klinikum Rechts der Isar, Technische Universita
t Munchen, 81675 Munchen, Germany
10German Cancer Consortium (DKTK), German Cancer Research Center (DKFZ), 69120 Heidelberg, Germany
11Department of Medicine for the Elderly, Cambridge University Hospitals NHS Trust, Cambridge CB2 0QQ, UK
12Human Genetics, Wellcome Trust Sanger Institute, Cambridge CB10 1SA, UK
13Co-rst author
*Correspondence: [email protected]
http://dx.doi.org/10.1016/j.celrep.2015.02.005
This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/3.0/).
2Sequencing

SUMMARY

Clonal hemopoiesis driven by leukemia-associated

gene mutations can occur without evidence of a blood
disorder. To investigate this phenomenon, we interrogated 15 mutation hot spots in blood DNA from 4,219
individuals using ultra-deep sequencing. Using only
the hot spots studied, we identied clonal hemopoiesis in 0.8% of individuals under 60, rising to 19.5% of
those R90 years, thus predicting that clonal hemopoiesis is much more prevalent than previously realized.
DNMT3A-R882 mutations were most common and,
although their prevalence increased with age, were
found in individuals as young as 25 years. By contrast,
mutations affecting spliceosome genes SF3B1 and
SRSF2, closely associated with the myelodysplastic
syndromes, were identied only in those aged >70
years, with several individuals harboring more than
one such mutation. This indicates that spliceosome
gene mutations drive clonal expansion under selection
pressures particular to the aging hemopoietic system
and explains the high incidence of clonal disorders
associated with these mutations in advanced old age.
INTRODUCTION
Cancers develop through the combined action of multiple mutations that are acquired over time (Nowell, 1976). This paradigm is

well established in hematological malignancies, whose clonal

history can be traced back for several years or even decades
(Ford et al., 1998; Kyle et al., 2002). It is also clear from studies
of paired diagnostic-relapsed leukemia samples that recurrent
disease can harbor some, but not always all, mutations present
at diagnosis, providing evidence for the presence of a clone of
ancestral pre-leukemic stem cells that escape therapy and
give rise to relapse through the acquisition of new mutations
(Ding et al., 2012; Kronke et al., 2013). Studies of such phenomena have dened a hierarchical structure among particular leukemia mutations, with some, such as those affecting the gene
DNMT3A, displaying the characteristics of leukemia-initiating lesions and driving the expansion of hemopoietic cell clones prior
to the onset of leukemia (Ding et al., 2012; Shlush et al., 2014).
These observations suggest that individuals without overt features of a hematological disorder may harbor hemopoietic cell
clones carrying leukemia-associated mutations. In fact, such
mutations, ranging from large chromosomal changes (Jacobs
et al., 2012; Laurie et al., 2012) to nucleotide substitutions (Busque et al., 2012), have been found to drive clonal hemopoiesis in
some individuals. Recent reanalyses of large exome-sequencing
data sets of blood DNA showed that clonal hemopoiesis is more
common than previously realized and increases with age to
affect up to 11% of those over 80 and 18.4% of those over
90 years (Genovese et al., 2014; Jaiswal et al., 2014; Xie et al.,
2014). The presence of such clones was associated with an
increased risk of developing hematological or other cancers
and a higher all-cause mortality, probably due to an increased
risk of cardiovascular disease (Genovese et al., 2014; Jaiswal
et al., 2014).
Cell Reports 10, 12391245, March 3, 2015 2015 The Authors 1239

Table 1. Mutation Hot Spots Interrogated in This Study

Gene

Target Codon

DNMT3A

R882

JAK2

V617

NPM1

L287

SRSF2

P95

SF3B1

K666

SF3B1

K700

IDH1

R132

IDH2

R140

IDH2

R172

KRAS

G12

NRAS

G12

NRAS

Q61

KIT

D816

FLT3

D835

FLT3

N676

Also see Table S1 for detailed information about numbers of samples

screened for each mutation.

The important ndings of these studies were based on analysis of exome-sequencing data sets that were generated for
the study of constitutional genomes, thus trading genomewide coverage for reduced sensitivity for detecting small subclonal events. We used the different approach of targeted resequencing of selected leukemia-associated mutation hot spots
in blood DNA from more than 4,000 individuals unselected for
blood disorders. In addition to increasing the sensitivity for detecting subclonal mutations, this approach enabled us to prospectively select and study a large number of elderly individuals.
Our results show that clonal hemopoiesis is signicantly more
common than anticipated, give new insights into the distinct
age-distribution and biological behavior of clonal hemopoiesis
driven by different mutations, and help explain the increased
incidence of myelodysplastic syndromes (MDSs) with advancing
age.

We obtained adequate coverage (R1,000 reads at all studied

hot spots) from 4,067 blood DNA samples and identied mutation-bearing clones in 105 of these. Of note, not all hot spots
were studied in all samples and the derived incidence of mutations in our population as a whole was 3.24% (Table S1). However, the incidence rose signicantly with age from 0.2% in the
1729 to 19.5% in the 9098 years age group (Figure 1A). We
found one or more samples with mutations at 9 of the 15 hot
spot codons studied, with VAFs varying widely within and between mutation groups (Table 2).
The most-common mutations were those affecting DNMT3A
R882, whose incidence rose with age from 0.2% (1/489) in the
1725 to a peak of 3.1% (11/355) in the 8089 age group. A
similar pattern was observed with JAK2 V617F mutations (Figure 1A). By contrast, spliceosome gene mutations at SRSF2
P95, SF3B1 K666, and SF3B1 K700 were exclusively observed
in people aged over 70 years, rising sharply from 1.8% in those
aged 7079 to 8.3% in the 9098 years age group. Among all
samples, we identied only six individuals with more than one
mutation; signicantly, ve of them had two independent spliceosome gene mutations of different VAFs (Figure 1B). Unfortunately, in each of three cases with two mutations at the same or
nearby positions, neighboring SNPs were not informative and
the variants could not be phased (see Supplemental Experimental Procedures).Occasional mutations in the genes IDH1,
IDH2, NRAS, and KRAS were also seen. Except for three samples with IDH1/2 mutations, hemoglobin concentrations did not
differ signicantly between individuals with and without hot
spot mutations (Figure S3A). For samples with full blood count
results available, JAK2 V617F mutant cases had a higher platelet
count (albeit within the normal range) than no mutation cases,
whereas other results did not differ (Figure S3B). No hot spot mutations were found in the few cord blood (n = 18) and post-transplantation (n = 32) samples studied.
Finally, despite using a very sensitive method and a mutationcalling script written specically for this purpose, no samples
with NPM1 mutations of VAF R 0.008 were identied. In fact,
variant reads reporting a canonical NPM1 mutation (mutation A;
TCTG duplication) were detected in only 1 of 4,067 samples at a
VAF of 0.0012 (4/3,466 reads).

RESULTS
DISCUSSION
To investigate the incidence, target genes, and age distribution
of age-related clonal hemopoiesis (ARCH), we performed targeted re-sequencing for hot spot mutations at 15 gene loci recurrently mutated in myeloid malignancies (Table 1) using blood
DNA from 3,067 blood donors aged 1770 (Wellcome Trust
Case Control Consortium [WTCCC]) and 1,152 unselected
individuals aged 6098 years (United Kingdom Household
Longitudinal Study [UKHLS]; see Figure S1 for detailed age distributions). To do this, we developed and validated a robust
methodology, employing barcoded multiplex PCR of mutational
hot spots followed by next-generation sequencing (MiSeq) and
bioinformatic analysis, to extract read counts and allelic fractions
for reference and non-reference nucleotides. This reliably detected mutation-associated circulating blood cell clones with a
variant allele fraction (VAF) R 0.008 (0.8%; see Supplemental
Experimental Procedures and Figure S2).
1240 Cell Reports 10, 12391245, March 3, 2015 2015 The Authors

Hematological malignancies develop through the serial acquisition of somatic mutations in a process that can take many years
or even decades (Ford et al., 1998; Kyle et al., 2002). Also, it is
clear that the presence of hemopoietic cells carrying leukemiaassociated mutations is only followed by the onset of hematological malignancies in a minority of cases (Busque et al., 2012;
Genovese et al., 2014; Jacobs et al., 2012; Jaiswal et al., 2014;
Laurie et al., 2012; Xie et al., 2014). In order to understand the
incidence and clonal dynamics of pre-leukemic clonal hemopoiesis, we interrogated 15 leukemia-associated mutation hot spots
using a highly sensitive methodology able to detect small clones
with mutations.
We show that clonal hemopoiesis is rare in the young but becomes common with advancing age. In particular, we observed
that ARCH driven by the mutations studied here doubled in

Figure 1. Prevalence and Age Distribution

of Hot Spot Mutations Driving Clonal Hemopoiesis
(A) Prevalence of mutations driving clonal hemopoiesis by age.
(B) Samples with more than one mutation, variant
allele fraction (VAF) of each mutation present, and
age of participant.
Also see Figure S1 for age distribution of all
participants.

frequency in successive decades after the age of 50, rising from

1.5% in those aged 5059 to 19.5% in those aged 9098 (Figure 1). Of note, 61 of 112 clones identied had a VAF % 3%
(Table 2), and it is likely that most of these would not have
been detected by conventional exome sequencing, which gives
lower than 10-fold average coverage compared to the current
study (see Table S2 for comparison to such studies), with
some recurrently mutated regions giving particularly low coverage (Genovese et al., 2014). Notably, our study did not search
for non-hot-spot mutations associated with ARCH such as those
affecting genes TET2 and ASXL1 or DNMT3A codons other than
R882 (Genovese et al., 2014; Jaiswal et al., 2014; Xie et al., 2014).
Assuming that the incidence of small clones is similar for such
mutations as for the hot spot mutations we studied here, the
mean projected true incidence of ARCH driven by leukemiaassociated mutations in those older than 90 years is greater
than 70% (Figure S4). This makes clonal hemopoiesis an almost
inevitable consequence of advanced aging.
Another signicant nding of our study is the disparate age
distribution of ARCH associated with different mutation types.
In particular, we found that, although DNMT3A R882 and JAK2
V617F mutations become more common with age, they were
also found in younger individuals. This is in keeping with the
increasing cumulative likelihood of their stochastic acquisition
with the passage of time. In contrast, spliceosome gene mutations were found exclusively in those aged 70 years or older,
replicating the sharp rise beyond this age in the incidence of
MDSs driven by these mutations and the fact that, among unselected MDS patients, those with spliceosome mutations are
signicantly older than those without (Haferlach et al., 2014;
Lin et al., 2014; Papaemmanuil et al., 2013; Wu et al., 2012).

Exome-sequencing studies describe a

much-lower rate of spliceosome mutations (Genovese et al., 2014; Jaiswal
et al., 2014; Xie et al., 2014), but this is
again likely to reect their lower sensitivity
for detecting small clones, which was a
particular limitation at spliceosome mutation hot spots as these were captured/
sequenced at lower-than-average depths
(Table S2). In our study, 19/33 SF3B1- or
SRSF2-associated clones had a VAF %
5%, with 13 of these at VAFs % 3% (Table 2), the majority of which would not
have been detected by low-coverage
sequencing. The identication of ARCH
driven by spliceosome gene mutations is in keeping with the
fact that these are founding mutations in the clonal evolution of
MDS and related hematological malignancies (Cazzola et al.,
2013; Haferlach et al., 2014; Papaemmanuil et al., 2013).
We propose that the exclusive identication of spliceosome
gene mutations in those aged R70 years can be explained by
differences in the prevailing pressures on clonal selection at
different ages, which can in turn explain how different gene mutations can generate detectable clonal expansions at different
ages (Figure 2). The alternatives are that spliceosome mutations
are associated with slower rates of clonal expansion or that they
are detected later because they contribute less to circulating leukocytes. Both of these scenarios are less plausible, given the
complete absence of such mutations even at low VAFs in
younger age groups. For any somatic mutation imparting a clonal
advantage to a stem/progenitor cell and leading to the generation of a steadily expanding clone, one would expect such a
clone to be detectable at a smaller size at earlier and a larger
size at later time points, as is the case for DNMT3A R882 and
JAK2 V617 mutations. Instead, clones (of any size) driven by
mutant SRSF2 and SF3B1 were observed exclusively in individuals aged 70 years or older, suggesting that these only begin to
expand later in life. Furthermore, considerable support for the
presence of a different selection milieu comes from the observation that ve of six patients with multiple mutations harbored two
independent spliceosome gene mutations, indicative of convergent evolution, i.e., evolution to overcome a shared selective
pressure or to exploit a shared environment (Greaves and Maley,
2012; Rossi et al., 2008).
It is tempting to consider the nature of age-related changes in
normal hemopoiesis that make it permissive to the outgrowth of
Cell Reports 10, 12391245, March 3, 2015 2015 The Authors 1241

Table 2. Amino Acid Consequences and VAFs of the 112 Clonal Mutations Identied in This Study
Mutation
Hot Spot
DNMT3A R882

Codon
p.R882H

VAF (%)
4.14

Age

Mutation
Hot Spot

25

Codon

VAF (%)

Age

p.R882H

32.02

81

Mutation
Hot Spot

Codon

VAF (%)

Age

IDH1 R132

p.R132H

42.13

84

p.R132C

0.92

92

p.R140Q

6.67

76

p.R882C

2.33

35

p.R882H

1.14

81

p.R882H

3.80

42

p.R882H

3.06

81

IDH2 R140
SRSF2 P95

p.R882H

4.00

42

p.R882H

2.17

81

p.R882H

1.25

43

p.R882H

1.13

82

p.P95R

4.46

70

p.P95L

3.35

72

p.R882H

19.00

48

p.R882H

1.46

82

p.P95H

0.86

73

p.R882H

1.18

49

p.R882C

2.62

82

p.P95H

0.84

77

p.R882S

1.74

49

p.R882C

6.15

89

p.P95L

0.97

79y

p.R882H

9.87

50

p.R882C

2.00

94

p.P95L

0.85

80yy

p.R882H

0.83

51

p.V617F

1.56

34

p.P95H

6.67

80yy

p.R882C

1.10

51

p.V617F

4.91

42

p.P95L

0.96

81

p.R882C

12.50

52

p.V617F

7.72

45

p.P95H

6.40

82

JAK2V617F

p.R882C

1.28

53

p.V617F

0.85

62

p.P95L

2.74

85

p.R882C

2.47

54

p.V617F

25.44

64

p.P95R

7.52

87

p.R882H

1.95

55

p.V617F

7.41

65

p.P95L

5.84

88**

p.R882C

30.22

55

p.V617F

1.03

67

p.P95H

10.48

88**

p.R882C

1.22

56

p.V617F

0.88

71

p.P95R

2.71

88

p.R882H

0.91

58

p.V617F

3.75

71

p.P95R

17.05

90z

p.R882H

4.17

60

p.V617F

1.16

75

p.K700E

1.04

76

SF3B1 K700

p.R882H

5.90

60

p.V617F

2.30

77

p.K700E

6.63

81

p.R882H

9.60

60

p.V617F

1.92

78

p.K700E

0.79

82

p.R882H

2.73

60

p.V617F

2.26

80*

p.K700E

12.59

p.R882C

9.33

60

p.V617F

4.25

80

p.K700E

8.77

83
83zz

p.R882H

7.03

61

p.V617F

1.92

80

p.K700E

1.02

84

p.R882C

1.21

61

p.V617F

3.71

80

p.K700E

0.85

90z

p.R882H

0.86

63

p.V617F

15.48

81

p.R882H

2.54

64

p.V617F

1.21

82

SF3B1 K666

p.K700E

1.37

90

p.K666N

1.33

70

p.R882H

3.19

67

p.V617F

1.62

85

p.K666N

5.01

79

p.R882H

2.74

70

p.V617F

0.83

85

p.K666N

13.36

79y

p.R882H

4.27

74

p.V617F

1.98

86

p.K666N

15.43

80*

p.R882H

0.85

74

p.V617F

25.94

88

p.K666N

4.60

81

p.R882H

0.85

75

p.V617F

10.88

88**

p.K666E

1.09

83zz

p.R882C

1.12

77

p.V617F

2.94

90

p.K666N

35.11

86

p.R882C

1.15

78

p.V617F

1.23

90

p.K666N

19.70

86

p.R882H

1.26

79

p.R882H

16.66

80

p.R882C

4.28

80

p.R882C

3.66

80

KRAS G12
NRAS G12

p.G12 R

0.94

55

p.K666N

16.55

86

p.G12S

2.78

78

p.K666E

3.34

95

p.G12S

1.50

61

p.G12D

0.96

62

Mutations identied in the same sample are highlighted with the same symbol (*, **, y, yy, z, and zz).

clones driven by spliceosome mutations. HSCs do not operate in

isolation; instead, their normal survival and behavior are closely
dependent on interactions with the hemopoietic microenvironment (Calvi et al., 2003; Rossi et al., 2008; Zhang et al., 2003).
Therefore, both cell-intrinsic and microenvironmental factors inuence hemopoietic aging (Rossi et al., 2008; Woolthuis et al.,
2011). For example, there is good evidence for age-related
changes in cell-intrinsic properties of HSCs in both mice (Cham1242 Cell Reports 10, 12391245, March 3, 2015 2015 The Authors

bers et al., 2007; Rossi et al., 2005) and humans (Rube et al.,
2011; Taraldsrud et al., 2009), and it is also clear that aging
has a profound effect on the hemopoietic niche, reducing its
ability to sustain polyclonal hemopoiesis, favoring oligo- or
monoclonality instead (Vas et al., 2012). These and many other
observations provide strong evidence that changes in the hemopoietic system subject HSCs to changing pressures during
normal aging, driving clonal selection (Rossi et al., 2008).

Figure 2. Proposed Kinetics of Hemopoietic Clones Driven by

Different Gene Mutations
Mutations such as DNMT3A R882H/C or JAK2 V617F drive a slow but inexorable clonal expansion, leading to the outgrowth of a detectable clone after a
certain latency. By contrast, mutations affecting spliceosome genes, such as
SF3B1 and SRSF2, and acquired at the same age for the purposes of this
model give no proliferative advantage initially but do so later in the context of
an aging hemopoietic compartment. Their effects may operate by prolonging
stem cell survival and repopulating tness beyond that of normal stem cells or
by exploiting cell-extrinsic changes in the aging microenvironment.

A striking example of such selection was described in a 115year-old woman whose peripheral white blood cells were shown
to be primarily the offspring of only two related HSC clones,
whose cargo of approximately 450 somatic mutations did not
include known leukemogenic mutations (Holstege et al., 2014).
In the absence of somatic driver mutations, it is probable that
such selection is driven by well-demonstrated epigenetic differences between individual HSCs (Fraga et al., 2005) or by stochastic events. Furthermore, clonal hemopoiesis in the absence
of a known leukemia-driver mutation was also well documented
recently (Genovese et al., 2014), and whereas unknown or undetected drivers may be responsible for many cases of this phenomenon, it is also highly plausible that a stochastic process
of clonal selection or loss may operate in others. Our study provides evidence that spliceosome gene mutations offer a means
to exploit age-related changes in hemopoiesis to drive clonal hemopoiesis in advanced old age, an observation that blurs the
boundary between driver and passenger mutations. Such
a context dependency is not a surprising attribute for the effects
of spliceosome mutations, which have not, so far, been shown to
impart a primary proliferative advantage to normal hemopoietic
stem and progenitor cells (Matsunawa et al., 2014; Visconte
et al., 2012).
A nal important nding of our study was the almost complete
absence of canonical NPM1 mutations in our collection of more
than 4,000 people, despite the use of a highly sensitive assay for
their detection, designed specically for this study. Among more
than 10 million mapped reads covering this mutation hot spot,
we identied only four reads in a single sample reporting a canonical mutation (mutation A; TCTG duplication). Given their frequency in myeloid leukemia (Cancer Genome Atlas Research
Network, 2013) and the fact that they are not late mutations
(Kronke et al., 2013; Shlush et al., 2014), this observation frames
NPM1 mutations as gatekeepers of leukemogenesis, i.e., their

acquisition appears to be closely associated with the development of frank leukemia. In this light, the frequent co-occurrence
of DNMT3A and NPM1 mutations suggests that the former
behave as rafts that enable NPM1 mutant clones to be
founded and expanded, thus facilitating onward evolution toward acute myeloid leukemia.
We used a highly sensitive method to search for evidence of
clonal hemopoiesis driven by 15 recurrent leukemogenic mutations in more than 4,000 individuals. Our results demonstrate
that the incidence of clonal hemopoiesis is much higher than
suggested by exome-sequencing studies, that spliceosome
gene mutations drive clonal outgrowth primarily in the context
of an aging hemopoietic compartment, and that NPM1 mutations do not drive ARCH, indicating that their acquisition is
closely associated with frank leukemia.
EXPERIMENTAL PROCEDURES
Patient Samples
Samples were obtained with written informed consent and in accordance with
the Declaration of Helsinki and appropriate ethics committee approvals from
all participants (approval reference numbers 10/H0604/02, 07/MRE05/44,
and 05/Q0106/74). Maternal consent was obtained for the use of cord blood
samples. Samples were obtained from 3,067 blood donors aged 1770 years
(WTCCC; UK Blood Services 1 [UKBS1] and UKBS2 common controls),
1,152 unselected individuals aged 6098 years (UKHLS; https://www.
understandingsociety.ac.uk/), 32 patients that had undergone a hemopoietic
stem cell transplant (12 autologous and 20 allogeneic; Tables S3 and S4)
1 month to 14 years previously, and 18 cord blood samples. Age distribution
of the WTCCC and UKHLS cohorts/samples is shown in Figure S1. Hemoglobin concentrations were available for a total of 3,587 of the 4,067 samples from
which adequate sequencing data were obtained for analysis, including 102 of
105 samples with mutations. Full blood count results were available for 2,952
WTCCC samples. The average blood donation frequency for WTCCC donors
was 1.6 donations of one unit per year. Details of donations by individual participants were not available.
Targeted Sequencing
Genomic DNA was used to simultaneously amplify several gene loci using
multiplex PCR, in order to capture and analyze 15 mutational hot spots enriched for, but not exclusive to, targets of mutations thought to arise early in
leukemogenesis (Table 1). We used three multiplex primer combinations
(Plex1-3), guided by our ndings, to capture the targeted mutational hot spots
(Table S1). Primers were designed using the Hi-Plex PCR-MPS (massively
parallel sequencing) strategy (Nguyen-Dumont et al., 2013), except for JAK2
V617 and Plex2 primers, which were designed using MPRIMER (Shen
et al., 2010). These and additional primer sequences used in each Plex and details of PCR- and DNA-sequencing protocols are detailed in Supplemental
Experimental Procedures. Methodological validation experiments are shown
in Figure S2.
Bioinformatic Analysis
Sequencing data were aligned to the human reference genome (hg19) using
BWA. Subsequently, the SAMTOOLS pileup command was used to generate
pileup les from the generated bam les (version 0.1.8; http://samtools.
sourceforge.net; Li et al., 2009). A exible in-house Perl script generated by
our group, MIDAS (Conte et al., 2013), was modied in order to interrogate
only the hot spot nucleotide positions of interest (those with reported mutations in the COSMIC database; Forbes et al., 2015) on the pileup le, considering only those reads with a sequence quality higher than 25 and a mapping
quality higher than 15. For each sample, the numbers of reads reporting the
reference and variant alleles at each position were extracted. VAFs were
derived by dividing the number of reads reporting the most-frequent variant
nucleotide to the total. In order to detect NPM1 mutations with high sensitivity,

Cell Reports 10, 12391245, March 3, 2015 2015 The Authors 1243

we wrote a bespoke Perl script described in Supplemental Experimental

Procedures.
Statistical Analyses and Mutation-Calling Threshold
We chose a threshold VAF of R0.008 (0.8%) to call clones with a heterozygous mutation representing R1.6% of blood leukocytes. From validation experiments and data analysis (see Supplemental Experimental Procedures
and Figure S2D), we determined that the maximum false-positive error rate
for calling a mutation (VAF R 0.008) due to variant allele counts that are solely
due to PCR-MiSeq error was negligible (p < 105). For comparisons of blood
cell counts and hemoglobin concentrations, we used non-paired t tests. For
summary statistics of read coverage (Table S2) and for the purposes of deriving
an estimate of the overall incidence of clonal hemopoiesis (Figure S4), we used
published tables of all mutations reported by three recent studies that employed whole-exome-sequencing analyses to identify individuals with clonal
hemopoiesis (Genovese et al., 2014; Jaiswal et al., 2014; Xie et al., 2014).
ACCESSION NUMBERS
The European Genome-Phenome Archive (EGA) accession number for the
sequencing data reported in this paper is EGAS00001000814.

Received: December 14, 2014

Revised: January 19, 2015
Accepted: January 29, 2015
Published: February 26, 2015
REFERENCES
Busque, L., Patel, J.P., Figueroa, M.E., Vasanthakumar, A., Provost, S., Hamilou, Z., Mollica, L., Li, J., Viale, A., Heguy, A., et al. (2012). Recurrent somatic
TET2 mutations in normal elderly individuals with clonal hematopoiesis. Nat.
Genet. 44, 11791181.
Calvi, L.M., Adams, G.B., Weibrecht, K.W., Weber, J.M., Olson, D.P., Knight,
M.C., Martin, R.P., Schipani, E., Divieti, P., Bringhurst, F.R., et al. (2003).
Osteoblastic cells regulate the haematopoietic stem cell niche. Nature 425,
841846.
Cancer Genome Atlas Research Network (2013). Genomic and epigenomic
landscapes of adult de novo acute myeloid leukemia. N. Engl. J. Med. 368,
20592074.
Cazzola, M., Della Porta, M.G., and Malcovati, L. (2013). The genetic basis of
myelodysplasia and its clinical relevance. Blood 122, 40214034.

SUPPLEMENTAL INFORMATION

Chambers, S.M., Shaw, C.A., Gatza, C., Fisk, C.J., Donehower, L.A., and
Goodell, M.A. (2007). Aging hematopoietic stem cells decline in function and
exhibit epigenetic dysregulation. PLoS Biol. 5, e201.

Supplemental Information includes Supplemental Experimental Procedures,

four gures, and four tables and can be found with this article online at
http://dx.doi.org/10.1016/j.celrep.2015.02.005.

Conte, N., Varela, I., Grove, C., Manes, N., Yusa, K., Moreno, T., Segonds-Pichon, A., Bench, A., Gudgin, E., Herman, B., et al. (2013). Detailed molecular
characterisation of acute myeloid leukaemia with a normal karyotype using targeted DNA capture. Leukemia 27, 18201825.

AUTHOR CONTRIBUTIONS
G.S.V. conceived and designed the study. G.S.V. and T. McKerrell supervised
the study, analyzed data, and wrote the manuscript. N.P. and T. McKerrell
performed experimental procedures. I.V. and T. Moreno wrote scripts and performed bioinformatics analysis. H.P., T. McKerrell, and G.S.V. performed statistical analyses. E.Z., C.S.G., M.A.Q., and R.R. contributed to study strategy
and to technical and analytical aspects. U.S.S.G., E.Z., W.O, J.C., C.C., J.B.,
J.S., C.H., M.A.S., and D.R.F. contributed to sample acquisition and subject
recruitment.
ACKNOWLEDGMENTS
This project was funded by a Wellcome Trust Clinician Scientist Fellowship
(100678/Z/12/Z; to T. McKerrell) and by the Wellcome Trust Sanger Institute
(grant number WT098051). G.S.V. is funded by a Wellcome Trust Senior
Fellowship in Clinical Science (WT095663MA), and work in his laboratory is
also funded by Leukaemia Lymphoma Research and the Kay Kendal
Leukaemia Fund. I.V. is funded by Spanish Ministerio de Economa y Competitividad subprograma Ramon y Cajal. C.S.G. is funded by a Leukaemia Lymphoma Research Clinical Research Training Fellowship. We thank Servicio
Santander Supercomputacion for their support. We acknowledge use of
DNA from The UK Blood Services Collection of Common Controls (UKBS
collection), funded by the Wellcome Trust grant 076113/C/04/Z, by the Juvenile Diabetes Research Foundation grant WT061858, and by the National Institute of Health Research of England. The collection was established as part of
the Wellcome Trust Case-Control Consortium. We also gratefully acknowledge use of blood DNA samples and data from participants of the UK Household Longitudinal Study (https://www.understandingsociety.ac.uk/), collected
by NatCen and the Institute for Social and Economic Research, University of
Essex, and funded by the Economic and Social Research Council, UK. We
thank the Cambridge Blood and Stem Cell Biobank and the Cancer Molecular
Diagnosis Laboratory, Cambridge Biomedical Research Centre (National Institute for Health Research, UK) for help with sample collection and processing.
Finally, we thank Nathalie Smerdon, Richard Rance, Lucy Hildyard, Ben Softly,
and Britt Killian for help with sample management, DNA sequencing, and data
processing. G.S.V. is a consultant for KYMAB and receives an educational
grant from Celgene.

1244 Cell Reports 10, 12391245, March 3, 2015 2015 The Authors

Ding, L., Ley, T.J., Larson, D.E., Miller, C.A., Koboldt, D.C., Welch, J.S.,
Ritchey, J.K., Young, M.A., Lamprecht, T., McLellan, M.D., et al. (2012). Clonal
evolution in relapsed acute myeloid leukaemia revealed by whole-genome
sequencing. Nature 481, 506510.
Forbes, S.A., Beare, D., Gunasekaran, P., Leung, K., Bindal, N., Boutselakis,
H., Ding, M., Bamford, S., Cole, C., Ward, S., et al. (2015). COSMIC: exploring
the worlds knowledge of somatic mutations in human cancer. Nucleic Acids
Res. 43, D805D811.
Ford, A.M., Bennett, C.A., Price, C.M., Bruin, M.C., Van Wering, E.R., and
Greaves, M. (1998). Fetal origins of the TEL-AML1 fusion gene in identical
twins with leukemia. Proc. Natl. Acad. Sci. USA 95, 45844588.
Fraga, M.F., Ballestar, E., Paz, M.F., Ropero, S., Setien, F., Ballestar, M.L.,
Heine-Suner, D., Cigudosa, J.C., Urioste, M., Benitez, J., et al. (2005). Epigenetic differences arise during the lifetime of monozygotic twins. Proc. Natl.
Acad. Sci. USA 102, 1060410609.
Genovese, G., Kahler, A.K., Handsaker, R.E., Lindberg, J., Rose, S.A., Bakhoum, S.F., Chambert, K., Mick, E., Neale, B.M., Fromer, M., et al. (2014).
Clonal hematopoiesis and blood-cancer risk inferred from blood DNA
sequence. N. Engl. J. Med. 371, 24772487.
Greaves, M., and Maley, C.C. (2012). Clonal evolution in cancer. Nature 481,
306313.
Haferlach, T., Nagata, Y., Grossmann, V., Okuno, Y., Bacher, U., Nagae, G.,
Schnittger, S., Sanada, M., Kon, A., Alpermann, T., et al. (2014). Landscape
of genetic lesions in 944 patients with myelodysplastic syndromes. Leukemia
28, 241247.
Holstege, H., Pfeiffer, W., Sie, D., Hulsman, M., Nicholas, T.J., Lee, C.C., Ross,
T., Lin, J., Miller, M.A., Ylstra, B., et al. (2014). Somatic mutations found in the
healthy blood compartment of a 115-yr-old woman demonstrate oligoclonal
hematopoiesis. Genome Res. 24, 733742.
Jacobs, K.B., Yeager, M., Zhou, W., Wacholder, S., Wang, Z., Rodriguez-Santiago, B., Hutchinson, A., Deng, X., Liu, C., Horner, M.J., et al. (2012). Detectable clonal mosaicism and its relationship to aging and cancer. Nat. Genet. 44,
651658.
Jaiswal, S., Fontanillas, P., Flannick, J., Manning, A., Grauman, P.V., Mar,
B.G., Lindsley, R.C., Mermel, C.H., Burtt, N., Chavez, A., et al. (2014). Agerelated clonal hematopoiesis associated with adverse outcomes. N. Engl. J.
Med. 371, 24882498.

Kronke, J., Bullinger, L., Teleanu, V., Tschurtz, F., Gaidzik, V.I., Kuhn, M.W.,
Rucker, F.G., Holzmann, K., Paschka, P., Kapp-Schworer, S., et al. (2013).
Clonal evolution in relapsed NPM1-mutated acute myeloid leukemia. Blood
122, 100108.
Kyle, R.A., Therneau, T.M., Rajkumar, S.V., Offord, J.R., Larson, D.R., Plevak,
M.F., and Melton, L.J., 3rd. (2002). A long-term study of prognosis in monoclonal gammopathy of undetermined signicance. N. Engl. J. Med. 346,
564569.
Laurie, C.C., Laurie, C.A., Rice, K., Doheny, K.F., Zelnick, L.R., McHugh, C.P.,
Ling, H., Hetrick, K.N., Pugh, E.W., Amos, C., et al. (2012). Detectable clonal
mosaicism from birth to old age and its relationship to cancer. Nat. Genet.
44, 642650.
Li, H., Handsaker, B., Wysoker, A., Fennell, T., Ruan, J., Homer, N., Marth, G.,
Abecasis, G., and Durbin, R.; 1000 Genome Project Data Processing Subgroup (2009). The Sequence Alignment/Map format and SAMtools. Bioinformatics 25, 20782079.
Lin, C.C., Hou, H.A., Chou, W.C., Kuo, Y.Y., Wu, S.J., Liu, C.Y., Chen, C.Y.,
Tseng, M.H., Huang, C.F., Lee, F.Y., et al. (2014). SF3B1 mutations in patients
with myelodysplastic syndromes: the mutation is stable during disease evolution. Am. J. Hematol. 89, E109E115.
Matsunawa, M., Yamamoto, R., Sanada, M., Sato-Otsubo, A., Shiozawa, Y.,
Yoshida, K., Otsu, M., Shiraishi, Y., Miyano, S., Isono, K., et al. (2014). Haploinsufciency of Sf3b1 leads to compromised stem cell function but not to myelodysplasia. Leukemia 28, 18441850.
Nguyen-Dumont, T., Pope, B.J., Hammet, F., Southey, M.C., and Park, D.J.
(2013). A high-plex PCR approach for massively parallel sequencing. Biotechniques 55, 6974.
Nowell, P.C. (1976). The clonal evolution of tumor cell populations. Science
194, 2328.

Rossi, D.J., Jamieson, C.H., and Weissman, I.L. (2008). Stems cells and the
pathways to aging and cancer. Cell 132, 681696.
Rube, C.E., Fricke, A., Widmann, T.A., Furst, T., Madry, H., Pfreundschuh, M.,
and Rube, C. (2011). Accumulation of DNA damage in hematopoietic stem and
progenitor cells during human aging. PLoS ONE 6, e17487.
Shen, Z., Qu, W., Wang, W., Lu, Y., Wu, Y., Li, Z., Hang, X., Wang, X., Zhao, D.,
and Zhang, C. (2010). MPprimer: a program for reliable multiplex PCR primer
design. BMC Bioinformatics 11, 143.
Shlush, L.I., Zandi, S., Mitchell, A., Chen, W.C., Brandwein, J.M., Gupta, V.,
Kennedy, J.A., Schimmer, A.D., Schuh, A.C., Yee, K.W., et al.; HALT Pan-Leukemia Gene Panel Consortium (2014). Identication of pre-leukaemic haematopoietic stem cells in acute leukaemia. Nature 506, 328333.
Taraldsrud, E., Grgaard, H.K., Solheim, S., Lunde, K., Flisand, Y., Arnesen,
H., Seljeot, I., and Egeland, T. (2009). Age and stress related phenotypical
changes in bone marrow CD34+ cells. Scand. J. Clin. Lab. Invest. 69, 7984.
Vas, V., Senger, K., Dorr, K., Niebel, A., and Geiger, H. (2012). Aging of the
microenvironment inuences clonality in hematopoiesis. PLoS ONE 7, e42080.
Visconte, V., Rogers, H.J., Singh, J., Barnard, J., Bupathi, M., Traina, F.,
McMahon, J., Makishima, H., Szpurka, H., Jankowska, A., et al. (2012).
SF3B1 haploinsufciency leads to formation of ring sideroblasts in myelodysplastic syndromes. Blood 120, 31733186.
Woolthuis, C.M., de Haan, G., and Huls, G. (2011). Aging of hematopoietic
stem cells: Intrinsic changes or micro-environmental effects? Curr. Opin. Immunol. 23, 512517.
Wu, S.J., Kuo, Y.Y., Hou, H.A., Li, L.Y., Tseng, M.H., Huang, C.F., Lee, F.Y.,
Liu, M.C., Liu, C.W., Lin, C.T., et al. (2012). The clinical implication of SRSF2
mutation in patients with myelodysplastic syndrome and its stability during disease evolution. Blood 120, 31063111.

Papaemmanuil, E., Gerstung, M., Malcovati, L., Tauro, S., Gundem, G., Van
Loo, P., Yoon, C.J., Ellis, P., Wedge, D.C., Pellagatti, A., et al.; Chronic Myeloid
Disorders Working Group of the International Cancer Genome Consortium
(2013). Clinical and biological implications of driver mutations in myelodysplastic syndromes. Blood 122, 36163627, quiz 3699.

Xie, M., Lu, C., Wang, J., McLellan, M.D., Johnson, K.J., Wendl, M.C.,
McMichael, J.F., Schmidt, H.K., Yellapantula, V., Miller, C.A., et al. (2014).
Age-related mutations associated with clonal hematopoietic expansion and
malignancies. Nat. Med. 20, 14721478.

Rossi, D.J., Bryder, D., Zahn, J.M., Ahlenius, H., Sonu, R., Wagers, A.J., and
Weissman, I.L. (2005). Cell intrinsic alterations underlie hematopoietic stem
cell aging. Proc. Natl. Acad. Sci. USA 102, 91949199.

Zhang, J., Niu, C., Ye, L., Huang, H., He, X., Tong, W.G., Ross, J., Haug, J.,
Johnson, T., Feng, J.Q., et al. (2003). Identication of the haematopoietic
stem cell niche and control of the niche size. Nature 425, 836841.

Cell Reports 10, 12391245, March 3, 2015 2015 The Authors 1245

Cell Reports

Report
Accelerating Novel Candidate Gene Discovery in
Neurogenetic Disorders via Whole-Exome Sequencing
of Prescreened Multiplex Consanguineous Families
Anas M. Alazami,1,23 Nisha Patel,1,23 Hanan E. Shamseldin,1,23 Shamsa Anazi,1 Mohammed S. Al-Dosari,2
Fatema Alzahrani,1 Hadia Hijazi,1 Muneera Alshammari,3 Mohammed A. Aldahmesh,1 Mustafa A. Salih,3 Eissa Faqeih,4
Amal Alhashem,5,6 Fahad A. Bashiri,3 Mohammed Al-Owain,5,7 Amal Y. Kentab,3 Sameera Sogaty,8 Saeed Al Tala,9
Mohamad-Hani Temsah,3 Maha Tulbah,10 Rasha F. Aljelaify,11 Saad A. Alshahwan,6 Mohammed Zain Seidahmed,12
Adnan A. Alhadid,3 Hesham Aldhalaan,13 Fatema AlQallaf,13 Wesam Kurdi,10 Majid Alfadhel,14 Zainab Babay,15
Mohammad Alsogheer,16 Namik Kaya,1 Zuhair N. Al-Hassnan,5,7 Ghada M.H. Abdel-Salam,17 Nouriya Al-Sannaa,18
Fuad Al Mutairi,14 Heba Y. El Khashab,3,19 Saeed Bohlega,13 Xiaofei Jia,20 Henry C. Nguyen,20 Rakad Hammami,1
Nouran Adly,1 Jawahir Y. Mohamed,1 Firdous Abdulwahab,1 Niema Ibrahim,1 Ewa A. Naim,1,21 Banan Al-Younes,1,21
Brian F. Meyer,1,21 Mais Hashem,1 Ranad Shaheen,1 Yong Xiong,20 Mohamed Abouelhoda,1,21
Abdulrahman A. Aldeeri,1,22 Dorota M. Monies,1,21 and Fowzan S. Alkuraya1,5,21,*
1Department

of Genetics, King Faisal Specialist Hospital and Research Center, Riyadh 11211, Saudi Arabia
of Pharmacognosy, College of Pharmacy, King Saud University, Riyadh 11451, Saudi Arabia
3Department of Pediatrics, King Khalid University Hospital and College of Medicine, King Saud University, Riyadh 11451, Saudi Arabia
4Department of Pediatrics, King Fahad Medical City, Riyadh 11525, Saudi Arabia
5Department of Anatomy and Cell Biology, College of Medicine, Alfaisal University, Riyadh 11533, Saudi Arabia
6Department of Pediatrics, Prince Sultan Military Medical City, Riyadh 11159, Saudi Arabia
7Department of Medical Genetics, King Faisal Specialist Hospital and Research Center, Riyadh 11211, Saudi Arabia
8Department of Pediatrics, King Fahad General Hospital, Jeddah 23325, Saudi Arabia
9Department of Pediatrics, Armed Forces Hospital, Khamis Mushayt 62413, Saudi Arabia
10Department of Obstetrics & Gynecology, King Faisal Specialist Hospital, Riyadh 11211, Saudi Arabia
11Center of Excellence for Genomics, King Abdulaziz City for Science and Technology, Riyadh 11442, Saudi Arabia
12Department of Pediatrics, Security Forces Hospital, Riyadh 12625, Saudi Arabia
13Department of Neurosciences, King Faisal Specialist Hospital and Research Center, Riyadh 11211, Saudi Arabia
14Division of Genetics, Department of Pediatrics, King Saud bin Abdulaziz University for Health Sciences, King Abdulaziz Medical City, Riyadh
14611, Saudi Arabia
15Department of Obstetrics and Gynecology, College of Medicine, King Saud University, Riyadh 11451, Saudi Arabia
16Department of Psychiatry, College of Medicine, King Saud University, Riyadh 11451, Saudi Arabia
17Department of Clinical Genetics, Human Genetics and Genome Research Division, National Research Centre, Cairo 12345, Egypt
18Department of Pediatrics, Johns Hopkins Aramco Healthcare, Dhahran 34465, Saudi Arabia
19Department of Pediatrics, Childrens Hospital, Ain Shams University, Cairo 01234, Egypt
20Department of Molecular Biophysics and Biochemistry, Yale University, New Haven, CT 06520, USA
21Saudi Human Genome Program, King Abdulaziz City for Science and Technology, Riyadh 11442, Saudi Arabia
22Department of Internal Medicine, College of Medicine, King Saud University, Riyadh 11451, Saudi Arabia
23Co-rst author
*Correspondence: [email protected]
http://dx.doi.org/10.1016/j.celrep.2014.12.015
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/).
2Department

SUMMARY

Our knowledge of disease genes in neurological disorders is incomplete. With the aim of closing this gap,
we performed whole-exome sequencing on 143
multiplex consanguineous families in whom known
disease genes had been excluded by autozygosity
mapping and candidate gene analysis. This prescreening step led to the identication of 69 recessive genes not previously associated with disease,
of which 33 are here described (SPDL1, TUBA3E,
INO80, NID1, TSEN15, DMBX1, CLHC1, C12orf4,
WDR93, ST7, MATN4, SEC24D, PCDHB4, PTPN23,
TAF6, TBCK, FAM177A1, KIAA1109, MTSS1L,
148 Cell Reports 10, 148161, January 13, 2015 2015 The Authors

XIRP1, KCTD3, CHAF1B, ARV1, ISCA2, PTRH2,

GEMIN4, MYOCD, PDPR, DPH1, NUP107, TMEM92,
EPB41L4A, and FAM120AOS). We also encountered
instances in which the phenotype departed signicantly from the established clinical presentation of
a known disease gene. Overall, a likely causal mutation was identied in >73% of our cases. This study
contributes to the global effort toward a full compendium of disease genes affecting brain function.
INTRODUCTION
Neurogenetic disorders represent the largest category of Mendelian diseases in humans. They encompass a wide array of

clinical presentations that range from the common e.g., intellectual disability (>1%) to the very rare, e.g., neurodegeneration
with brain iron accumulation (one to three per 106) (Kalman
et al., 2012; Maulik et al., 2011). The highly prevalent involvement
of the nervous system in many Mendelian disorders coincides
with the observation that >80% of all human genes are expressed at some stage of brain development (Hawrylycz et al.,
2012) and suggests that the brain is one of the most vulnerable
organs to genetic perturbation. In fact high-resolution microarray
analysis of the human genome reveals that intellectual disability
is the common phenotypic denominator of genomic disorders
that involve losses or gains of genes (Coe et al., 2012).
Variances in clinical presentation are a major obstacle in
establishing a working molecular classication of neurological
disease, because even where the clinical presentation is highly
specic, genetic heterogeneity is the rule. In the setting of
autosomal recessive neurogenetic disorders where parents are
related, a homozygosity scan can serve as a guide to the underlying genetic cause even when the phenotype is atypical (Alkuraya, 2010). Another major challenge in assigning a molecular
classication is that many neurological disease genes have not
been identied yet.
Novel disease gene discovery in this eld has been tremendously abetted by next-generation sequencing, a tool with the
capacity to, theoretically, unravel the genetic cause of all neurological diseases. This full theoretical potential has not yet been
reached unfortunately, although the technology continues to
evolve. For example, two large studies on the genetics of intellectual disability using whole-exome sequencing (WES) provided
a yield of 16%55%, and even though the collective sample size
was >150, only seven novel genes were identied (de Ligt et al.,
2012; Rauch et al., 2012). In these studies, samples could not be
enriched for novel gene discovery, and simultaneously the anticipated mutations were heterozygous, detection of which poses a
challenge for the currently available sequencing technology
(especially regarding insertions/deletions) (Harismendy et al.,
2009). The presence of these two obstacles likely hindered the
authors ability to obtain a higher yield. Thus, alternative/complementary approaches are required to facilitate the discovery of
novel neurogenetic disease genes. In this study, we show that
the analysis of the entire set of autozygous intervals per individual (the autozygome) in multiplex consanguineous patients, as a
prescreen, can markedly increase the yield of WES to identify
candidate genes not previously associated with disease. Even
when known genes were identied using this approach, the
phenotype was often sufciently different to explain why the
gene had been missed by the autozygosity lter. The 33 candidate disease genes we showcase in this study will augment
the global hunt for the genetics of brain development and
function.
RESULTS
Clinical Report
In total, 143 multiplex families met our inclusion criteria (a neurogenetic diagnosis, positive family history, consanguineous
parents, and no candidates identied by autozygosity mapping).
Intellectual disability was the most common clinical feature.

Other phenotypes that were also enriched included global

developmental delay, autism, epilepsy, primary microcephaly,
ataxia, and neurodegeneration. Table S1 summarizes the clinical
features of the entire cohort.
Autozygosity Mapping Is a Powerful Enrichment Tool for
Novel Candidate Gene Discovery
As a prescreening step for each case, the regions of homozygosity (ROH), the telltale sign of shared ancestral haplotypes, were
interrogated for disease genes that matched the patients
phenotype. These genes were prioritized for Sanger sequencing.
If negative, or when no compelling disease genes were evident in
the ROH, patient DNA was subjected to WES under the assumption that this will reveal a novel disease gene. This assumption
fails, however, to account for certain scenarios. When disease
genes within the ROH are examined for a likely candidate, it is
possible that the clinical picture may sufciently deviate from
the classical phenotype ascribed to a particular gene such that
the gene falls outside our consideration. Specically, we list in
Table 1 cases in which there was sufcient discrepancy between
the classical and observed phenotypes that the respective
genes were missed in the autozygosity mapping stage. Some
phenotypes can even be considered unique rather than an
expansion of a known phenotype.
A second reason why WES may reveal a known disease gene
is that the causal mutation may lie within an ROH but fail to be
detected, which is one of the known pitfalls of autozygosity
mapping (Alkuraya, 2012). Figure S1 captures the homozygosity
pattern of a number of patients for whom the candidate ROH was
missed, because it did not meet our ROH size cutoff, or it appeared to be shared by an unaffected member of the family, or
the overlap of the gene with the ROH was not clearly discernible.
One case (11DG0165) evaded detection due to the presence of
a deep intronic mutation, which was later uncovered using RTPCR. A third scenario for why WES may expose a known disease
gene is that the gene was simply overlooked when searching for
plausible candidates within an ROH. This occurred in 13DG1803
where TUSC3, a known intellectual disability gene, was not
noticed during the prescreening stage. Further scenarios include
incomplete clinical information at the time of analysis (as
occurred with 09DG-00774 and 12DG0926) and poor description in the literature (discussed below). We emphasize, however,
that cases of failed autozygosity map prescreening are the
exception rather than the rule because they represent only
7.7% (11/143) of all cases and have reduced the hypothetical
yield of WES in revealing candidate genes not associated with
disease, or known genes with unique phenotypes, by only 2.2%.
WES Is a Powerful Novel Candidate Gene Discovery Tool
in Neurogenetic Disorders
Consistent with our prescreening enrichment step, WES revealed 69 genes that were not known to the authors at the
time of exome capture. Of these, 36 have since been published
by either us or others (detailed in Table S1), leaving 33 candidate
genes that are here reported. The phenotypes associated with
these are described in Table 2 and include intellectual disability,
autism, progressive cerebellar atrophy, primary microcephaly,
brain atrophy and other malformations, and myopathy. Of these
Cell Reports 10, 148161, January 13, 2015 2015 The Authors 149

150 Cell Reports 10, 148161, January 13, 2015 2015 The Authors

SPG20

CYP27A1

NPC2

ARFGEF2 NM_006420:c.656_657insC:p.P219fs

PNKP

RYR1

FBN2

12DG0096

12DG1571

10DG0672

11DG1951

11DG1510

12DG0975

12DG0104

NM_147127.4:c.3870_3893 dup:p.K1293_K1300 dup

EVC2

DDHD2

WDR81

ZNF526

10DG1721

13DG0583

13DG0010

12DG0685

NM_002972:c.1327G > A:p.D443N

SBF1

AP4M1

09DG0301

11DG1767

NM_004722:c.C952T:p.R318*

NM_020247:c.1744 dup:p.S582Kfs*148

09DG00930 ADCK3

NM_133444:c.479A > C:p.K160T

NM_001163809:c.845G > A:p.G282E

NM_015214.2:c.1249_1891 del:p.A417Mfs*23 (large scale deletion)

NM_000059.3:c.9152 delC:p.P3051Hfs*11

08DG00385 BRCA2

NM_001999:c.1064G > A:p.G355D

NM_000540:c.6617C > T:p.T2206M

NM_007254:c.1250_1251insAACGGGTCGCCATCGAC:p.R418Tfs*55

NM_006432:c.88G > A: p.V30M

NM_000784:c.1342C > T:p.R448C

NM_001142294:c.1450_1451insA:p.T484fs

NM_000405.4:c.164C > T:p.P55L

GM2A

09DG0057

Mutation

Gene

ID

spastic paraplegia, severe ID,

poor speech development

Charcot-Marie-Tooth disease
type 4B3

cerebellar ataxia, seizure and

cerebellar atrophy

nonsyndromic intellectual
disability

cerebellar ataxia, mental

retardation, and dysequilibrium
syndrome-2

spastic paraplegia

Ellis-van-Creveld syndrome

Fanconi anemia

congenital contractural
arachnodactyly

minicore myopathy

progressive microcephaly,
infantile-onset seizures,
and developmental delay

periventricular heterotopia
with microcephaly

Nieman-Pick disease

cerebrotendinous
xanthomatosis

spastic paraplegia

GM2-gangliosidosis,
AB variant

Published Phenotype

this study

this study

this study

this study

Shaheen
et al. (2013)

Shaheen
et al. (2014)

this study

this study

this study

this study

this study

this study

this study

this study

Reference

microcephaly, speech delay, spasticity;

brain MRI: hypomyelination, hypoplastic
corpus callosum, and brain atrophy.

this study

Charcot-Marie-Tooth with microcephaly, Alazami

ophthalmoplegia and syndactyly
et al. (2014)

isolated cerebellar hypoplasia

intellectual disability, Noonan-like

facies, and pulmonary stenosis

neonatal death due to severe brain

malformation (hydranencephaly and
severe cerebellar hypoplasia)

isolated cerebellar atrophy

Meckel-Gruber syndrome

primordial dwarsm

fetal akinesia with brain ischemia and

neonatal death

ptosis, no motor delay, and sacral

agenesis

primary microcephaly, global

developmental delay, no seizures

global developmental delay, epilepsy,

and hydrocephalus

microcephaly, static encephalopathy,

no organomegaly, and normal retina

severe choreoathetosis, no cataract,

and normal cholestanol

speech and motor delay, tremor,

microcephaly, and strabismus

progressive neurodegeneration with

onset at 8 years, no organomegaly
and normal retina

Observed Phenotype

Table 1. Cases with Mutations in Known Disease Genes following WES, Where the Patient Phenotype Diverged from the Established Literature

mutations, ten are truncating or located at splice sites. For the

missense changes, we determined pathogenicity based on the
in silico prediction of at least two established algorithms, as
well as 3D modeling of wild-type and mutant residues whenever
structure information on homologous proteins was available
(Figure S2). Our minimum threshold for assigning candidacy to
a gene was that the variant had to be the only one to survive
the stringent pipeline illustrated in Figure 2. A total of 37 cases
remain unsolved, some because more than one variant survived our lters. Table S1 lists these, and it is important to note
that some of these variants may indeed represent bona de
novel disease genes.
In addition, WES revealed what appears to be phenotypes that
have not been described for the respective genes in 16 cases
(Table 1). One striking example is case 10DG0672 in which
we identied a previously reported homozygous mutation in
NPC2, a known gene for Nieman-Pick disease. Neither the index
nor his sibling had the typical presentation of progressive neurodegeneration or hepatosplenomegaly, thus making the diagnosis of Nieman-Pick nearly impossible on clinical grounds.
Finally, WES also revealed mutations in known genes for cases
with classical phenotypes, where the gene was missed either
due to a pitfall in autozygosity mapping or incomplete phenotyping (nine cases, discussed above) or because the phenotype
was not well described in the literature as in two instances.
The rst such instance is MGAT2-related dysmorphism, which
had only been documented by a single photograph (CormierDaire et al., 2000). The second instance involves CTSD, which
was described as causing congenital neuronal ceroid lipofuscinosis when, in fact, the detailed description of the case was
severe microlissencephaly and hyperekplexia, which is identical
to the phenotype of our case 09DG00288 (Fritchie et al., 2009).
Overall, the yield of WES in our cohort was 105 out of 143
(73.4%).
Although we used autozygosity mapping coupled with WES,
an alternative strategy is to simply use the mapping information
provided by WES analysis. Although most autozygous regions
can indeed be inferred by this second strategy, we had previously shown that the use of high-throughput genotyping results
in cleaner ROH data with sharper overall resolution (Carr et al.,
2013).
3D Modeling of Missense Variants to Support
Pathogenicity
3D modeling was performed for selected gene products using
the web-based homology modeling engine Phyre2 (Kelley and
Sternberg, 2009). Four of the models were built with very high
condence and suggested a pathogenic nature for the identied
mutations. In the case of TUBA3E, the model was built based on
the structure of tubulin alpha-3E (PDB ID: 3EDL)(Tan et al., 2008)
that shares 97% sequence identity with the TUBA3E gene product (Figure S2A). The R215C mutation changes the charge distribution on the surface of the protein. Although not participating
in microtubule formation (Tan et al., 2008), this site might be
involved in the binding of microtubule to other proteins or cofactors. The structure of the gene product of TSEN15, human tRNA
splicing endonuclease, has been solved by nuclear magnetic
resonance (Song and Markley, 2007) (PDB ID: 2GW6). The

conserved W76 is embedded inside the protein and is critical

in protein folding (Figure S2B). Mutation of W76G would leave
a void in the core of the protein and very likely lead to misfolded
proteins. PTRH2 encodes peptidyl-tRNA hydrolase 2. The crystal structure of a domain of PTRH2 has been solved (PDB ID:
1QSR). The observed mutation at Q85 is located in the middle
of a helix and forms a hydrogen bond with the side chain of
T157 (Figure S2C). Mutation of Q85P would destabilize PTRH2
as it would disrupt the hydrogen bond. Furthermore, the mutation to a proline may cause a kink in the helix and distort the overall fold of the structure. Iron-sulfur cluster assembly 2, or ISCA2,
was also modeled with the structure of its homolog of IscA from
Thermosynechococcus elongates (PDB ID: 1X0G), which is a
a2b2 heterotetramer. ISCA2 has modest sequence identity
(27% to PDB ID: 1X0G) to the a chain of the IscA. However,
Phyre2 predicted the same structural fold with 100% condence. The protein is involved in the maturation of iron-sulfur
proteins. According to the model, the G77S mutation occurs in
a loop that is directly involved in iron-sulfur cluster binding by
providing a chelating cysteine, C79 (Figure S2D). We speculate
that such a mutation might affect the stability or exibility of
the loop and therefore interfere with its efciency of binding to
the iron-sulfur cluster.
DISCUSSION
Several attempts have been made in the recent past to
accelerate the discovery of novel neurological disease genes.
One of the earliest attempts was the high-throughput Sanger
sequencing of all coding exons on the X chromosome in a large
cohort of >200 families with suspected X-linked intellectual
disability (Tarpey et al., 2009). In addition to the laborious nature
of this approach, the yield was somewhat modest (three novel
genes) partly because enrichment for novel gene discovery
was not feasible, and also partly due to a large proportion of
X-linked disease genes having already been established (de
Brouwer et al., 2007). High-resolution molecular karyotyping is
a powerful tool to identify a large number of DNA gains and losses that are associated with various neurological phenotypes,
but the yield is typically <15%, and it rarely identies single
genes due to the nature of the assay (Miller et al., 2010). Morrow
et al. used autozygosity mapping in nearly 90 consanguineous
families with autism, followed by Sanger sequencing of candidate genes within the linked ROH, to identify ve novel autism
genes (Morrow et al., 2008). The lower yield of that study likely
originates from the use of conventional sequencing methods,
coupled with the potentially non-Mendelian behavior of autism
genes.
The advent of next-generation sequencing has revolutionized
the search for Mendelian neurocognitive genes. Rauch et al. and
de Ligt et al. studied >150 cases of intellectual disability using a
trio-exome design, and, although that approach is compatible
with identifying recessive disease genes, they only identied heterozygous de novo mutations, including seven novel genes (de
Ligt et al., 2012; Rauch et al., 2012). The known bias of current
WES against heterozygous mutations, especially insertions
and deletions, as well as the inability of the investigators to enrich
their cohort for novel genes are likely explanations for the lower
Cell Reports 10, 148161, January 13, 2015 2015 The Authors 151

152 Cell Reports 10, 148161, January 13, 2015 2015 The Authors

Phenotype

primary microcephaly
and neonatal death

microlissencephaly
and global
developmental delay

primary microcephaly
and global
developmental delay

hydrocephalus,
muscle weakness
and global
developmental delay

primary microcephaly
and global
developmental delay

ID

12DG1528

11DG0443

10DG1705

08DG00041

13DG0167

TSEN15

NID1

INO80

TUBA3E

CCDC99
(SPDL1)

Gene

NM_001127394:
c.226T > G:p.W76G

NM_002508.2:
c.3385+1G > A

INO80:NM_017553:
c.1501T > C:p.S501P,
INO80:NM_017553:
c.3737G > A:p.R1246Q

NM_207312:c.643C >
T:p.R215C

NM_017785:c.1724_
1747 del:p.S575_T582
del

Mutation

missense

splicing
(in-frame
insertion
conrmed
by RTPCR)

missense

missense

in-frame
deletion

Mutation
Type

Mice decient of NID1 exhibit

neurologic decits including
seizure-like symptoms and loss
of muscle control in the hind
legs and show altered basement
membrane morphology in
selected locations including
brain capillaries and the lens
capsule. Additionally, a variant
of unknown signicance has
been reported in a family with
brain malformations. Only surviving
variant and segregates in family.

INO80 has been shown as necessary

for DNA damage repair. Abnormal
DNA damage repair underlies multiple
forms of microcephaly, e.g., PHC1
and PNKP. Supported by a single
linkage peak. Only surviving variant
and segregates in family.

Mutations in several tubulins have

been linked to lissencephaly and
other cortical malformations (TUBA1
A, TUBA8, TUBB2B, TUBB3, TUBB5,
and TUBG1). Only surviving variant
and segregates in family.

CCDC99 has been demonstrated

to control poleward movement of
chromosomes along the mitotic
spindles. Many primary
microcephaly genes encode
proteins that are involved in
mitotic spindle regulation,
e.g., ASPM and WDR62. Only
surviving variant and segregates
in family.

Reference

PMID:
18711368

PMID:
12480912
and 23674478

PMID:
21947284,
19829069,
24029917,
and 20118933

PMID:
24860126

PMID:
20427577
and 24875059

(Continued on next page)

tRNA splicing
Mutations in other family members,
endonuclease 15 e.g., TSEN54, have been linked to
pontocerebellar hypoplasia. Only
surviving variant and segregates
in family.

Nidogen 1

INO80 complex
subunit E

tubulin, alpha 3e

spindle
apparatus
coiled-coil
protein 1

Gene Description Supporting Evidence

Table 2. Cases with Mutations in Candidate Genes, along with Available Evidence from the Literature to Support Candidacy

Cell Reports 10, 148161, January 13, 2015 2015 The Authors 153

Phenotype

global developmental
delay, epilepsy and
poor weight gain

myopathy

global developmental
delay

autism spectrum
disorder

global developmental
delay and brain
atrophy

holoprosencephaly

intellectual disability
and epilepsy

primary microcephaly
and global
developmental delay

brain atrophy and

global developmental
delay

12DG0929

09DG0405

09DG00102

11DG1513

11DG2479

12DG1901

12DG2051

10DG1069

08DG-00322

Continued

ID

Table 2.

PTPN23

PCDHB4

SEC24D

MATN4

ST7

WDR93

C12orf4

C2orf63
(CLHC1)

DMBX1

Gene

NM_015466:c.3995G >
T:p.R1332L

NM_018938.2:c.915
del:p.K305Nfs*12

NM_014822:c.697G >
C:p.G233R

NM_030590:c.515G >
C:p.G172A

NM_021908:c.489T >
G:p.Y163X

NM_020212:c.280T >
C:p.Y94H

NM_020374:exon6:
c.637_638insAAAC:
p.K213fs

NM_001135598:
c.779G > A:p.R260Q

NM_147192:c.367C >
T:p.R123W

Mutation

missense

frameshift

missense

missense

stopgain

missense

frameshift

missense

missense

Mutation
Type

protein tyrosine
phosphatase,
nonreceptor
type 23

protocadherin
beta 4

SEC24 family,
member D
(S. cerevisiae)

Matrilin 4

suppression of
tumorigenicity 7

WD repeat
domain 93

chromosome 12
open reading
frame 4

clathrin heavychain linker

domain
containing 1

Reference

PMID:
19378249

PMID:
22495309
and 22765916

PMID:
23596517

PMID:
11549321

PMID:
10889047

PMID:
23664116

PMID:
15314164
and 17873059

(Continued on next page)

Mouse model displays early

embryonic lethality. Gene is
highly expressed in developing
brain. Only surviving variant and
segregates in family.

PDCHB4 has been associated

with autism. Other protocadherin
members have been linked to
epilepsy, cognitive impairment
as well as autistic features. Only
surviving variant and segregates
in family.

Mouse model displays early

embryonic lethality, but the
mouse model of its paralog
SEC24B displays abnormal
neural tube development. Only
surviving variant and segregates
in family.

Gene is highly expressed in

developing brain. Only surviving
variant and segregates in family.

Disruption of ST7 has been

reported in one ASD patient
with a translocation (t(7;13)
(q31.3;q21)). Only surviving
variant and segregates in family.

Only surviving variant and

segregates in family.

Only surviving variant and

segregates in family.

Mutation in BICD2 which interacts

with CLHC1 causes spinal muscular
atrophy. Only surviving variant and
segregates in family.

Mouse model displays hyperactivity

and hypophagia. Gene is highly
expressed in developing brain. Only
surviving variant and segregates in
family.

Gene Description Supporting Evidence

diencephalon/
mesencephalon
homeobox 1

154 Cell Reports 10, 148161, January 13, 2015 2015 The Authors

Phenotype

global developmental
delay and
dysmorphism

global developmental
delay, epilepsy,
dysmorphism,
hypotonia, and VSD

macrocephaly, ID,
dolichocephaly,
and mild obesity

Dandy-Walker
malformation,
hydrocephalus,
exed deformity,
club feet,
micrognathia,
and pleural effusion

neurodegeneration
and brain iron
accumulation

11DG0932

10DG1670

13DG1472

13DG1900

10DG0264

Continued

ID

Table 2.

TAF6 RNA
polymerase II,
TATA box binding
protein (TBP)associated factor

Independently identied by another

group (B. Yuan, D. Pehlivan,
E. Karaca, N.P., W.-L. Charng,
T. Gambin, C. Gonzaga-Jauregui,
V.R. Sutton, G. Yesil, S.T. Bozdogan,
T. Tos, A. Koparir, E. Koparir,
C.R. Beck, S. Gu, H. Aslan,
O.O. Yuregir, K. Al Rubeaan,
D. Alnaqeb, M.J.A., Y. Bayram,
M.M. Atik, H. Aydin, B. Geckinli,
M. Seven, H. Ulucan, E. Fenercioglu,
M. Ozen, S. Jhangiani, D.M. Muzny,
E. Boerwinkle, Baylor-Hopkins Center
for Mendelian Genomics, B. Tuysuz,
F.S.A., R.A. Gibbs, and J.R. Lupski,
unpublished data). Only surviving
variant and segregates in family.

Gene Description Supporting Evidence

NM_138383:c.1790C >
T:p.T597M

MTSS1L

missense

NM_015312.3:c.1557T > stopgain

A:p.Y519*

KIAA1109

frameshift

metastasis
suppressor
1-like

KIAA1109

family with
sequence
similarity 177,
member A1

PMID:
17956977

PMID:
19640479

PMID:
23977024,
19963289

Reference

(Continued on next page)

This mutation affects a poorly

characterized isoform of VAC14
deciency of which in mouse
results in severe neurodegeneration.
Only surviving variant and
segregates in family.

Deletion in Drosophila results

in lethality, and the rare
homozygous ies that reach
adulthood exhibit severe
neurological signs: seizures
and inability to walk or stand
for prolonged period. Only
surviving variant and segregates
in family.

Only surviving variant and

segregates in family.

splicing
TBC1 domain
TBCK knockdown signicantly
(frameshift) containing kinase suppresses mTOR signaling,
which plays a critical role in
multiple neurological disorders.
Only surviving variant and
segregates in family.

missense

Mutation
Type

NM_001079519:c.297_
298insA:p.L99fs

NM_033115:
c.1708+1G > A

NM_005641.3:c.212T >
C:p.I71T

Mutation

FAM177A1

TBCK

TAF6

Gene

Cell Reports 10, 148161, January 13, 2015 2015 The Authors 155

severe psychomotor
retardation, seizure,
and cerebellar
hypoplasia

global developmental
delay and ADHD

neurodegenerative
disease

13DG2274

13DG0832

08DGRC00077

neurodegeneration
with marked white
matter changes with
high lactate peak in
the brain, consistent
with mitochondrial
encephalopathy

primary microcephaly

13DG1935

14DG0152

Phenotype

Continued

ID

Table 2.

ISCA2

ARV1

NM_194279:c.229G >
A:p.G77S

NM_022786:c.565G >
A:p.G189R

NM_005441:c.496A >
G:p.I166V

NM_016121:c.1036_
1073 del:p.P346Tfs*4

KCTD3

CHAF1B

NM_194293:c.4495G >
A:p.E1499K

Mutation

XIRP1

Gene

missense

missense

missense

frameshift

missense

Mutation
Type

iron-sulfur
cluster
assembly 2
homolog

ARV1 homolog
(S. cerevisiae)

chromatin
assembly
factor 1,
subunit B

potassium
channel
tetramerization
domain
containing 3

Reference

PMID:
12145310

PMID:
22177093

PMID:
19623214

PMID:
22366181

(Continued on next page)

PMID:
Maps to the only shared haplotype
22323289
in the extended family. ISCA2 is a
member of mitochondrial iron-sulfur
cluster (ISC) assembly machinery.
Depletion of ISCA2 results in
massively swollen mitochondria that
are devoid of cristae membranes
indicating that it is required for normal
mitochondrial biogenesis. Same
mutation was identied in other
families with identical presentation
(Z.N.A.-H., M. Al-Dosary, M. Alfadhel,
E.F., M. Alsagob, R. Kenana,
R. Almas, O.S. Al-Harazi, H. Al-Hindi,
O.I. Malibari, F.B. Almutari,
T. Al-Sheddi, S. Tulbah, F. Alhadeq,
R. Alamro, A, AlAsmari, M. Almuntashri,
H. Alshaalan, F.A. Al-Mohanna, D. Colak,
N.K., unpublished data). Only surviving
variant and segregates in family.

ARV1 is required for normal

sphingolipid metabolism, a
process known to be defective
in other neurodegenerative
diseases such as Nieman-Pick
disease. Only surviving variant
and segregates in family.

CHAF1B is part of the chromatin

assembly complex deciency of
which results in loss of asymmetry
during nervous system development
in C. elegans. Only surviving variant
and segregates in family.

Gene is enriched in copy number

changes associated with intellectual
disability. Supported by a single
linkage peak. Only surviving variant
and segregates in family.

Gene is highly expressed in the

brain in response to oxidative
stress. Only surviving variant and
segregates in family.

Gene Description Supporting Evidence

xin actin-binding
repeat
containing 1

156 Cell Reports 10, 148161, January 13, 2015 2015 The Authors

global developmental NUP107

delay, light
complexion, early
onset focal segmental
glomerulosclerosis
TMEM92

cerebellar vermis
hypoplasia, DandyWalker malformation,
hydrocephalus,
developmental delay

10DG0934

11DG0417

14DG0221

cerebellar atrophy,
hydrocephalus, and
global developmental
(cognitive, speech,
and motor) delay

global developmental
delay, typical
Joubert syndrome,
MRI ndings

13DG0274

DPH1

PDPR

NM_001168215:
c.95+3A > G

NM_020401:c.303G >
A:p.M101I

NM_001383.3:c.701T >
C:p.L234P

NM_017990:c.1360G >
T:p.G454C

NM_001146312:
c.1252A > G:p.I418V

MYOCD

intellectual disability
and epilepsy

NM_016077.3:c.254A >
C:p.Q85P

13DG1549

PTRH2

Mutation

NM_015721:c.2452T >
C:p.W818R

global developmental
delay, hearing loss,
and ataxia

13DG0215

Gene

GEMIN4
08DG0048510DG070313DG1542 (08DG00485) global
developmental delay,
severe dystonia, and
congenital cataract
(10DG0703) global
developmental
delay and congenital
cataract(13DG1542)
global developmental
delay, congenital
cataract, tubulopathy,
and severe osteopenia

Phenotype

Continued

ID

Table 2.

splice site

splice site

missense

missense

missense

missense

missense

Mutation
Type

transmembrane
protein 92

nucleoporin
107 kDa

diphthamide
biosynthesis 1

pyruvate
dehydrogenase
phosphatase
regulatory
subunit

myocardin

gem (nuclear
organelle)
associated
protein 4

peptidyl-tRNA
hydrolase 2

PMID:
18765564

PMID:
12867591

PMID:
11914277

PMID:
18218778

Reference

(Continued on next page)

Only surviving variant and segregates

in family.

Only surviving variant and segregates

in family.

Diphthamide is a unique
posttranslationally modied
histidine found only in translation
elongation factor-2 (eEF2), which
is linked to spinocerebellar ataxia.
Diphthamide modication of eEF2
is essential for normal mouse
development. Only surviving variant
and segregates in family.

Only surviving variant and segregates

in family.

Only surviving variant and segregates

in family.

Maps to the only shared haplotype

in the three families. GEMIN4 is part
of the SMN complex, and reduced
SMN protein results in spinal
muscular atrophy. Supported by a
single linkage peak. Only surviving
variant and segregates in family.

Null mice show ataxia and weakness;

tissue examination revealed delayed
development. Only surviving variant
and segregates in family.

Gene Description Supporting Evidence

coarse facial features FAM120AOS NM_198841:c.743C >

T:p.T248I
(open mouth,
bilateral ptosis, and
hypertelorism),
scoliosis, pectus
excavatum, skin laxity,
hypotonia, GERD,
chronic lung disease,
undescended
testicles.
12DG0321

missense

family with
sequence
similarity 120A
opposite strand

Only surviving variant and segregates

in family.

Reference
Only surviving variant and segregates
in family.

Gene Description Supporting Evidence

erythrocyte
membrane
protein band
4.1 like 4A
missense
NM_022140:c.1298C >
T:p.S433L
EPB41L4A
spastic paraplegia,
failure to thrive.
10DG0840

Mutation
Type
Gene

Mutation
Phenotype
ID

Continued
Table 2.

yield. In 2011, Najmabadi et al. reported the identication of 50

novel candidate genes (Najmabadi et al., 2011). Although that
study, similar to ours, combines autozygosity mapping with
next-generation sequencing of candidate ROH, their cohort
was not enriched for novel gene discovery. Consequently, there
were numerous cases that, upon exome sequencing, identied a
gene that was classically linked to the clinical presentation (as
per their Table 1). More importantly, the pipeline used in that
study did not strictly require that the candidate variant be the
only one to survive ltering in that family; hence, 24 of their candidate genes (48%) were not the sole surviving variant. Of the
remainder, one variant (HIST3H3) is present at sufciently high
frequency in our collection of 485 exomes to be excluded (allele
frequency 0.0082).
Beyond our analytic pipeline, the candidacy of many of our
candidate genes can be corroborated through other lines of
evidence. In the case of INO80, KCTD3, GEMIN4, and ISCA2,
each of these genes maps to the only shared haplotype
genome-wide across multiple or extended multiplex families
(Figure S3). TUBA3E belongs to a family of proteins that are
known to be involved in brain malformation syndromes including
lissencephaly, which is present in the affected patient (Figure S4)
(Jaglin et al., 2009; Keays et al., 2007; Poirier et al., 2010). As
well, TSEN15 belongs to a family of proteins (t-RNA splicing endonucleases) that have been found mutated in patients with pontocerebellar hypoplasia and microcephaly, which is what we
observed in our patient (Budde et al., 2008; Cassandrini et al.,
2010; Namavar et al., 2011). SPDL1 controls poleward movement of chromosomes along the mitotic spindles, analogous to
many of the known primary microcephaly genes that are involved
in mitotic spindle regulation (Barisic et al., 2010; Chen et al.,
2014); our patient exhibits an extreme reduction in overall brain
volume resulting in an almost empty skull (Figure S4). ISCA2 is
involved in mitochondrial protein maturation, and depletion of
the protein results in massively swollen mitochondria that are
devoid of cristae membranes (Sheftel et al., 2012). This is consistent with the mitochondrial encephalopathy detected in our
patient. Table 2 summarizes what is known of the 33 candidate
genes and available evidence that supports candidacy.
We were unable to identify a causal mutation in 25.9% of
cases. Many of these have two or more variants that remained
after ltering (Table S1), leaving open the possibility that one of
these surviving variants may indeed be causal. Some of these
are strong candidates. For instance, case 11DG0375 with brain
atrophy has the rst reported null mutation in SNCG, which encodes synuclein-g, and, although a role in neurodegeneration
has been suspected, such a role has remained elusive (GretenHarrison et al., 2010). Similarly, TRIM4 is a member of a family
of proteins that has been implicated in a number of neurological
disorders (TRIM2 in Charcot-Marie-Tooth, TRIM32 in Limb-Girdle Muscular Dystrophy, and TRIM18 in Opitz-GBBB (a syndromic form of intellectual disability) (Balastik et al., 2008; Frosk
et al., 2002; Quaderi et al., 1997). Thus, the homozygous TRIM4
truncation we identied in 12DG2083 may prove causal if additional mutations in this gene are identied in the future.
Cases in which no variants survived our lters may harbor
classes of mutations that either the sequencing technology
or our analysis pipeline are biased against e.g., compound
Cell Reports 10, 148161, January 13, 2015 2015 The Authors 157

Figure 1. Pedigrees of Families with Novel Candidate Genes following WES Analysis
The family ID is presented above each pedigree. A red box indicates the affected family member who was submitted for WES, and asterisks denote all the family
members we had access to for conrming segregation.

heterozygous mutations, X-linked mutations, or mutations

involving introns, UTRs, regulatory elements, and repeats. Nonetheless, our study clearly shows the utility of our approach when
applied in the right population. Our pipeline as mentioned here
has been very successful historically. Many unpublished genes
we highlighted as novel and disease causing at the time of
data analysis were subsequently published and veried by
others, and this lends weight to the candidacy of the genes we
report here. Indeed, such reports appeared as recently as a
few months from this submission, e.g., DIAPH1, PIGQ, and
WWOX (Ercan-Sencicek et al., 2014; Mallaret et al., 2014; Martin
et al., 2014). This is also true for certain variants in unsolved
cases where more than one variant remained, e.g., SLC13A5
(Thevenon et al., 2014). Our aim in publishing these 33 candidate
genes is to accelerate the discovery of other independent mutations, thereby conrming pathogenicity and assisting in genetic
diagnosis. Scaling of our study is feasible given the availability
of the appropriate patient samples and the dropping cost of
sequencing technology, with the promise that all autosomal
158 Cell Reports 10, 148161, January 13, 2015 2015 The Authors

recessive neurogenetic disease genes can be mapped within

the timeframe required by the global brain initiatives.
EXPERIMENTAL PROCEDURES
Human Subjects
Consanguineous families with history of a neurological disorder were clinically
evaluated and recruited for this study using a King Faisal Specialist Hospital
and Research Center institutional-review-board-approved protocol (RAC#
2121053) with informed consent. All families were multiplex, and all parents
were conrmed to be healthy. Pedigrees of families with the 33 candidate
genes are given in Figure 1. For each family, blood was collected in EDTA
tubes from all available members, and families were only included if at least
one affected individual was available for sampling. DNA was extracted from
whole blood using standard protocols.
Autozygome-Guided Mutation Analysis
Genome-wide SNP genotyping and homozygosity mapping was performed
using the AxiomGWH SNP Chip platform (Affymetrix). See Supplemental Information for more details. Genes known to cause a neurological disorder
compatible with the patients phenotype, and present within the autozygome,

Figure 2. Schematic of the Experimental

Pipeline for This Study

Recruitment
Epilepsy

Ataxia

Ausm

Intellectual
Disability

Primary
Microcephaly/
Brain
malformaons

Enrichment step

Global
developmental
delay

A total of 143 multiplex consanguineous families

with history of a neurological disorder were
found negative for known disease genes following
autozygome-guided analysis and were recruited
for this study. The bar illustrates the breakdown
based on the number of cases, whereas the pie
chart is based on the number of distinct genes
identied in the solved cases.

WES

Autozygome-guided
sequencing of known
disease genes
is negative
n=143 families

Data ltraon and results

Unsolved: >2
remaining
variants (10)

Solved cases
(105)

Unsolved: no
remaining
variants (27)

were screened by PCR and Sanger sequencing. If no such genes existed, or

if they did exist but were excluded by sequencing, exome capture was performed. In total, individuals from 143 families were exome sequenced, and
these families formed the cohort for this study.

sa/bioinf/db/variants/dg/index.html. The likely pathogenic variants reported

in this study have also been uploaded to ClinVar, and this section will be
updated with the corresponding accession number.
SUPPLEMENTAL INFORMATION

Exome Sequencing
Exome capture was performed using the TruSeq Exome Enrichment kit
(Illumina). See Supplemental Information for more details. A summary of the
quality control data for exome sequencing is provided in Table S2.

Supplemental Information includes two gures and two tables and can be
found with this article online at http://dx.doi.org/10.1016/j.celrep.2014.12.015.
AUTHOR CONTRIBUTIONS

Analysis of Exomic Variants

Exome-derived data were ltered according to the schematic in Figure 2.
See Supplemental Information for full details. Segregation was assessed for
all surviving variants, using all family members we had access to (Figure 1).

A.M.A. collected and analyzed data and wrote the manuscript, N.P. collected
and analyzed data and wrote the manuscript, H.E.S. collected and analyzed
data and wrote the manuscript.

ACCESSION NUMBERS

ACKNOWLEDGMENTS

All variants within the 143 exomes in this study can be accessed through the
following link (part of the Saudi Variome Database): http://shgp.kfshrc.edu.

We thank all the families for their enthusiastic participation. This work was
funded in part by KACST 13-BIO1113-20 (F.S.A.). We acknowledge the Saudi

Cell Reports 10, 148161, January 13, 2015 2015 The Authors 159

Human Genome Project for infrastructure and informatics support relating to

work presented in this manuscript. We are also grateful to the sequencing
core facility, and to Salma Wakil and the genotyping core facility, at KFSH&RC
for their invaluable assistance. M.A.S. was supported by the Deanship of
Scientic Research, King Saud University, Riyadh, Saudi Arabia through
Research Group no. RGP-VPP-301.
Received: August 21, 2014
Revised: November 19, 2014
Accepted: December 8, 2014
Published: December 31, 2014
REFERENCES
Alazami, A.M., Alzahrani, F., Bohlega, S., and Alkuraya, F.S. (2014). SET binding factor 1 (SBF1) mutation causes Charcot-Marie-tooth disease type 4B3.
Neurology 82, 16651666.
Alkuraya, F.S. (2010). Homozygosity mapping: one more tool in the clinical
geneticists toolbox. Genet. Med. 12, 236239.
Alkuraya, F.S. (2012). Discovery of rare homozygous mutations from studies of
consanguineous pedigrees. Curr. Protoc. Hum. Genet. 6, 16.12.
Balastik, M., Ferraguti, F., Pires-da Silva, A., Lee, T.H., Alvarez-Bolado, G., Lu,
K.P., and Gruss, P. (2008). Deciency in ubiquitin ligase TRIM2 causes
accumulation of neurolament light chain and neurodegeneration. Proc.
Natl. Acad. Sci. USA 105, 1201612021.
Barisic, M., Sohm, B., Mikolcevic, P., Wandke, C., Rauch, V., Ringer, T., Hess,
M., Bonn, G., and Geley, S. (2010). Spindly/CCDC99 is required for efcient
chromosome congression and mitotic checkpoint regulation. Mol. Biol. Cell
21, 19681981.
Budde, B.S., Namavar, Y., Barth, P.G., Poll-The, B.T., Nurnberg, G., Becker,
C., van Ruissen, F., Weterman, M.A., Fluiter, K., te Beek, E.T., et al. (2008).
tRNA splicing endonuclease mutations cause pontocerebellar hypoplasia.
Nat. Genet. 40, 11131118.

tion and the rst prenatal screening of cathepsin D deciency (CLN10). Acta
Neuropathol. 117, 201208.
Frosk, P., Weiler, T., Nylen, E., Sudha, T., Greenberg, C.R., Morgan, K.,
Fujiwara, T.M., and Wrogemann, K. (2002). Limb-girdle muscular dystrophy
type 2H associated with mutation in TRIM32, a putative E3-ubiquitin-ligase
gene. Am. J. Hum. Genet. 70, 663672.
Greten-Harrison, B., Polydoro, M., Morimoto-Tomita, M., Diao, L., Williams,
A.M., Nie, E.H., Makani, S., Tian, N., Castillo, P.E., Buchman, V.L., and Chandra, S.S. (2010). abg-Synuclein triple knockout mice reveal age-dependent
neuronal dysfunction. Proc. Natl. Acad. Sci. USA 107, 1957319578.
Harismendy, O., Ng, P.C., Strausberg, R.L., Wang, X., Stockwell, T.B., Beeson, K.Y., Schork, N.J., Murray, S.S., Topol, E.J., Levy, S., and Frazer, K.A.
(2009). Evaluation of next generation sequencing platforms for population
targeted sequencing studies. Genome Biol. 10, R32.
Hawrylycz, M.J., Lein, E.S., Guillozet-Bongaarts, A.L., Shen, E.H., Ng, L.,
Miller, J.A., van de Lagemaat, L.N., Smith, K.A., Ebbert, A., Riley, Z.L., et al.
(2012). An anatomically comprehensive atlas of the adult human brain transcriptome. Nature 489, 391399.
Jaglin, X.H., Poirier, K., Saillour, Y., Buhler, E., Tian, G., Bahi-Buisson, N.,
Fallet-Bianco, C., Phan-Dinh-Tuy, F., Kong, X.P., Bomont, P., et al. (2009).
Mutations in the b-tubulin gene TUBB2B result in asymmetrical polymicrogyria. Nat. Genet. 41, 746752.
Kalman, B., Lautenschlaeger, R., Kohlmayer, F., Buchner, B., Kmiec, T., Klopstock, T., and Kuhn, K.A. (2012). An international registry for neurodegeneration with brain iron accumulation. Orphanet J. Rare Dis. 7, 66.
Keays, D.A., Tian, G., Poirier, K., Huang, G.-J., Siebold, C., Cleak, J., Oliver,
P.L., Fray, M., Harvey, R.J., Molnar, Z., et al. (2007). Mutations in a-tubulin
cause abnormal neuronal migration in mice and lissencephaly in humans.
Cell 128, 4557.
Kelley, L.A., and Sternberg, M.J. (2009). Protein structure prediction on the
Web: a case study using the Phyre server. Nat. Protoc. 4, 363371.

Carr, I.M., Bhaskar, S., OSullivan, J., Aldahmesh, M.A., Shamseldin, H.E.,
Markham, A.F., Bonthron, D.T., Black, G., and Alkuraya, F.S. (2013).
Autozygosity mapping with exome sequence data. Hum. Mutat. 34, 5056.

Mallaret, M., Synofzik, M., Lee, J., Sagum, C.A., Mahajnah, M., Sharkia, R.,
Drouot, N., Renaud, M., Klein, F.A., and Anheim, M. (2014). The tumour suppressor gene WWOX is mutated in autosomal recessive cerebellar ataxia
with epilepsy and mental retardation. Brain 137, 411419.

Cassandrini, D., Biancheri, R., Tessa, A., Di Rocco, M., Di Capua, M., Bruno,
C., Denora, P.S., Sartori, S., Rossi, A., Nozza, P., et al. (2010). Pontocerebellar
hypoplasia: clinical, pathologic, and genetic studies. Neurology 75, 1459
1464.

Martin, H.C., Kim, G.E., Pagnamenta, A.T., Murakami, Y., Carvill, G.L., Meyer, E.,
Copley, R.R., Rimmer, A., Barcia, G., Fleming, M.R., et al.; WGS500 Consortium
(2014). Clinical whole-genome sequencing in severe early-onset epilepsy reveals
new genes and improves molecular diagnosis. Hum. Mol. Genet. 23, 32003211.

Chen, J.-F., Zhang, Y., Wilde, J., Hansen, K.C., Lai, F., and Niswander, L.
(2014). Microcephaly disease gene Wdr62 regulates mitotic progression of
embryonic neural stem cells and brain size. Nat. Commun. 5, 3885.

Maulik, P.K., Mascarenhas, M.N., Mathers, C.D., Dua, T., and Saxena, S.
(2011). Prevalence of intellectual disability: a meta-analysis of populationbased studies. Res. Dev. Disabil. 32, 419436.

Coe, B.P., Girirajan, S., and Eichler, E.E. (2012). A genetic model for neurodevelopmental disease. Curr. Opin. Neurobiol. 22, 829836.

Miller, D.T., Adam, M.P., Aradhya, S., Biesecker, L.G., Brothman, A.R., Carter,
N.P., Church, D.M., Crolla, J.A., Eichler, E.E., Epstein, C.J., et al. (2010).
Consensus statement: chromosomal microarray is a rst-tier clinical diagnostic test for individuals with developmental disabilities or congenital anomalies. Am. J. Hum. Genet. 86, 749764.

Cormier-Daire, V., Amiel, J., Vuillaumier-Barrot, S., Tan, J., Durand, G.,
Munnich, A., Le Merrer, M., and Seta, N. (2000). Congenital disorders of glycosylation IIa cause growth retardation, mental retardation, and facial dysmorphism. J. Med. Genet. 37, 875877.
de Brouwer, A.P., Yntema, H.G., Kleefstra, T., Lugtenberg, D., Oudakker, A.R.,
de Vries, B.B., van Bokhoven, H., Van Esch, H., Frints, S.G., Froyen, G., et al.
(2007). Mutation frequencies of X-linked mental retardation genes in families
from the EuroMRX consortium. Hum. Mutat. 28, 207208.
de Ligt, J., Willemsen, M.H., van Bon, B.W., Kleefstra, T., Yntema, H.G., Kroes,
T., Vulto-van Silfhout, A.T., Koolen, D.A., de Vries, P., Gilissen, C., et al. (2012).
Diagnostic exome sequencing in persons with severe intellectual disability.
N. Engl. J. Med. 367, 19211929.
Ercan-Sencicek, A.G., Jambi, S., Franjic, D., Nishimura, S., Li, M., El-Fishawy,
P., Morgan, T.M., Sanders, S.J., Bilguvar, K., Suri, M., et al. (2014). Homozygous loss of DIAPH1 is a novel cause of microcephaly in humans. Eur. J.
Hum. Genet.
Fritchie, K., Siintola, E., Armao, D., Lehesjoki, A.-E., Marino, T., Powell, C.,
Tennison, M., Booker, J.M., Koch, S., Partanen, S., et al. (2009). Novel muta-

160 Cell Reports 10, 148161, January 13, 2015 2015 The Authors

Morrow, E.M., Yoo, S.-Y., Flavell, S.W., Kim, T.-K., Lin, Y., Hill, R.S., Mukaddes,
N.M., Balkhy, S., Gascon, G., Hashmi, A., et al. (2008). Identifying autism loci
and genes by tracing recent shared ancestry. Science 321, 218223.
Najmabadi, H., Hu, H., Garshasbi, M., Zemojtel, T., Abedini, S.S., Chen, W.,
Hosseini, M., Behjati, F., Haas, S., Jamali, P., et al. (2011). Deep sequencing
reveals 50 novel genes for recessive cognitive disorders. Nature 478, 5763.
Namavar, Y., Barth, P.G., Kasher, P.R., van Ruissen, F., Brockmann, K.,
Bernert, G., Writzl, K., Ventura, K., Cheng, E.Y., Ferriero, D.M., et al.; PCH
Consortium (2011). Clinical, neuroradiological and genetic ndings in pontocerebellar hypoplasia. Brain 134, 143156.
Poirier, K., Saillour, Y., Bahi-Buisson, N., Jaglin, X.H., Fallet-Bianco, C., Nabbout, R., Castelnau-Ptakhine, L., Roubertie, A., Attie-Bitach, T., Desguerre, I.,
et al. (2010). Mutations in the neuronal -tubulin subunit TUBB3 result in
malformation of cortical development and neuronal migration defects. Hum.
Mol. Genet. 19, 44624473.

Quaderi, N.A., Schweiger, S., Gaudenz, K., Franco, B., Rugarli, E.I., Berger,
W., Feldman, G.J., Volta, M., Andol, G., Gilgenkrantz, S., et al. (1997). Opitz
G/BBB syndrome, a defect of midline development, is due to mutations in a
new RING nger gene on Xp22. Nat. Genet. 17, 285291.
Rauch, A., Wieczorek, D., Graf, E., Wieland, T., Endele, S., Schwarzmayr, T.,
Albrecht, B., Bartholdi, D., Beygo, J., Di Donato, N., et al. (2012). Range of
genetic mutations associated with severe non-syndromic sporadic intellectual
disability: an exome sequencing study. Lancet 380, 16741682.
Shaheen, R., Faqeih, E., Alshammari, M.J., Swaid, A., Al-Gazali, L., Mardawi,
E., Ansari, S., Sogaty, S., Seidahmed, M.Z., AlMotairi, M.I., et al. (2013).
Genomic analysis of Meckel-Gruber syndrome in Arabs reveals marked
genetic heterogeneity and novel candidate genes. Eur. J. Hum. Genet. 21,
762768.
Shaheen, R., Faqeih, E., Ansari, S., Abdel-Salam, G., Al-Hassnan, Z.N., AlShidi, T., Alomar, R., Sogaty, S., and Alkuraya, F.S. (2014). Genomic analysis
of primordial dwarsm reveals novel disease genes. Genome Res. 24, 291299.

Sheftel, A.D., Wilbrecht, C., Stehling, O., Niggemeyer, B., Elsasser, H.-P.,
Muhlenhoff, U., and Lill, R. (2012). The human mitochondrial ISCA1, ISCA2,
and IBA57 proteins are required for [4Fe-4S] protein maturation. Mol. Biol.
Cell 23, 11571166.
Song, J., and Markley, J.L. (2007). Three-dimensional structure determined for
a subunit of human tRNA splicing endonuclease (Sen15) reveals a novel
dimeric fold. J. Mol. Biol. 366, 155164.
Tan, D., Rice, W.J., and Sosa, H. (2008). Structure of the kinesin13-microtubule ring complex. Structure 16, 17321739.
Tarpey, P.S., Smith, R., Pleasance, E., Whibley, A., Edkins, S., Hardy, C.,
OMeara, S., Latimer, C., Dicks, E., Menzies, A., et al. (2009). A systematic,
large-scale resequencing screen of X-chromosome coding exons in mental
retardation. Nat. Genet. 41, 535543.
Thevenon, J., Milh, M., Feillet, F., St-Onge, J., Duffourd, Y., Juge, C., Roubertie, A., Heron, D., Mignot, C., Raffo, E., et al. (2014). Mutations in SLC13A5
cause autosomal-recessive epileptic encephalopathy with seizure onset in
the rst days of life. Am. J. Hum. Genet. 95, 113120.

Cell Reports 10, 148161, January 13, 2015 2015 The Authors 161

Cell Reports

Report
Analysis of Intron Sequences Reveals Hallmarks
of Circular RNA Biogenesis in Animals
Andranik Ivanov,1 Sebastian Memczak,1,5 Emanuel Wyler,2,5 Francesca Torti,1,5 Hagit T. Porath,3 Marta R. Orejuela,1
Michael Piechotta,4 Erez Y. Levanon,3 Markus Landthaler,2 Christoph Dieterich,4 and Nikolaus Rajewsky1,*
1Laboratory

for Systems Biology of Gene Regulatory Elements, Berlin Institute for Medical Systems Biology, Max Delbruck Center for
Molecular Medicine, Robert Roessle Strae 10, 13125 Berlin-Buch, Germany
2Laboratory for RNA Biology and Posttranscriptional Regulation, Berlin Institute for Medical Systems Biology, Max Delbruck Center for
Molecular Medicine, Robert Roessle Strae 10, 13125 Berlin-Buch, Germany
3The Mina and Everard Goodman Faculty of Life Sciences, Bar-Ilan University, Ramat-Gan 5290002, Israel
4Max Planck Institute for Biology of Ageing, Cologne, Joseph Stelzmann Strae 9B, 50931 Ko
ln, Germany
5These authors contributed equally to this work
*Correspondence: [email protected]
http://dx.doi.org/10.1016/j.celrep.2014.12.019
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/).

SUMMARY

Circular RNAs (circRNAs) are a large class of animal

RNAs. To investigate possible circRNA functions,
it is important to understand circRNA biogenesis.
Besides human ALU repeats, sequence features
that promote exon circularization are largely unknown. We experimentally identied circRNAs in
C. elegans. Reverse complementary sequences between introns bracketing circRNAs were signicantly
enriched in comparison to linear controls. By scoring
the presence of reverse complementary sequences
in human introns, we predicted and experimentally
validated circRNAs. We show that introns bracketing
circRNAs are highly enriched in RNA editing or hyperediting events. Knockdown of the double-strand
RNA-editing enzyme ADAR1 signicantly and specifically upregulated circRNA expression. Together, our
data support a model of animal circRNA biogenesis
in which competing RNA-RNA interactions of introns
form larger structures that promote circularization
of embedded exons, whereas ADAR1 antagonizes
circRNA expression by melting stems within these interactions.
INTRODUCTION
Recently, several studies have revealed that the transcriptome of
animals contains many single-stranded exonic circular RNAs
(circRNAs) (Jeck et al., 2013; Jeck and Sharpless, 2014; Memczak et al., 2013; Salzman et al., 2012; Wang et al., 2014).
Although circRNAs have tissue- and stage-specic expression
(Memczak et al., 2013), the function of circRNAs is altogether unknown. The human circRNA CDR1as (Hansen et al., 2011;) can
act as a miRNA sponge (Hansen et al., 2013; Memczak et al.,
2013). However, we and others have proposed that generally
circRNAs might function in assembly of complexes, in transport,
170 Cell Reports 10, 170177, January 13, 2015 2015 The Authors

in trans (Memczak et al., 2013), by competing with linear splicing,

or as regulators of the local concentration of RNA-binding proteins (Ashwal-Fluss et al., 2014). In vitro studies (Braun et al.,
1996; Pasman et al., 1996) and a recent in vivo study (AshwalFluss et al., 2014) provided evidence that circRNAs are often
generated cotranscriptionally by head-to-tail splicing. In humans, circularized exons are typically bracketed by unusually
long introns (Jeck et al., 2013). Moreover, the circRNA SRY
(Capel et al., 1993) is bracketed by very long (15,000 nt), almost
perfectly complementary intronic sequences, and these reverse
complementary matches (RCMs) are required for SRY circularization (Dubin et al., 1995). Therefore, it has been proposed that
RCMs promote hairpin formation of the transcript. This would
explain how the 50 and 30 ends of an exon can be in spatial
proximity, perhaps thereby inducing head-to-tail splicing (Figure 1A). Jeck et al. (2013) reported that in humans, introns bracketing circRNAs are highly enriched in ALU repeats. The fact that
ALU repeats contain RCMs supports this model; however, ALU
repeats are specic to a small branch of vertebrates, and thus
the widespread existence of circRNAs in other animals remains
to be explained.
Caenorhabditis elegans, a well-annotated genome that is
not rich in repeats, offers the possibility of identifying conserved
features of circRNA biogenesis outside of vertebrates. We rst
sequenced RNA from several life stages of C. elegans and
were able to boost the number of annotated exonic circRNAs
from 300 (Memczak et al., 2013) to 1,100. Computational
analysis of the bracketing introns revealed that these circRNAs
are signicantly enriched for RCMs. We developed a simple
model for scoring circRNA biogenesis from intronic sequence
analysis and asked whether our model could predict novel human circRNAs. We successfully validated the predicted human
circRNAs.
The RNA-editing factor ADAR binds double-stranded RNA.
Thus, the model of circRNA biogenesis predicts that circRNAs
should be anked by intronic sequences that are enriched in
adenosine to inosine (A-to-I) editing. Moreover, it is known that
ALU elements are edited, often by ADAR1 (Athanasiadis et al.,
2004; Levanon et al., 2004; Osenberg et al., 2010; Ramaswami

ADAR2 knockdown in human cells affects circRNA expression,

we performed RNA sequencing (RNA-seq) and quantitative
RT-PCR (qRT-PCR). We observed a signicant and specic
upregulation of most circRNAs, whereas their linear host transcripts were less perturbed. Thus, our data reveal a surprising
function for ADAR as an antagonist of circRNA production. We
discuss the implications of these ndings.
RESULTS

Figure 1. RCMs between or within Introns and circRNA Biogenesis

(A) Model for the possible inuence of RCMs on circRNA biogenesis. RCMs
between different introns (lower panel) in competition with intron internal
RCMs (upper panel) may promote hairpin formation and circularization of the
embedded exon.
(B) Enrichment of RCMs and SMs in introns bracketing C. elegans circRNAs.
Left: distribution of RCM and SM counts per intron pair that anks circRNAs
(green) and length-matched controls (gray). Right: ratio of the number of intron
pairs bracketing circRNAs that contain only RCMs but no SMs to the number
of intron pairs bracketing circRNAs with only SMs but no RCMs (p value:
Fishers exact test).
See also Figures S1, S4, and S5.

et al., 2012). Also in C. elegans, long inverse repeats are enriched

in A-to-I editing (Morse et al., 2002). Indeed, we discovered signicant A-to-I editing, including hyperediting (Carmi et al., 2011),
in introns bracketing circRNAs. To test whether ADAR1 and

Identication and Characterization of circRNAs in

C. elegans
Recently, hundreds of circRNAs were reported in samples from
very early developmental stages of C. elegans (one/two-cell
embryo, oocytes, and sperm) (Memczak et al., 2013). To
include circRNAs expressed in later stages, we sequenced
ribosomal-depleted RNA from major life stages, including
adulthood (see Experimental Procedures). We computationally
identied circRNA candidates by applying our pipeline (Glazar
et al., 2014; Memczak et al., 2013; http://www.circbase.org)
and ensuring that the head-to-tail junctions precisely overlapped the annotated canonical splice sites (Experimental Procedures). This extended the published set of exonic circRNAs
from 300 to 1,111. Analogously to human circRNAs (Jeck
et al., 2013), the circRNA anking introns were much longer
(median 10-fold) than all of the C. elegans introns (Figure S1A).
Therefore, we asked whether intron length alone is sufcient
for circularization or additional sequence features are needed.
More precisely, in C. elegans we tested the idea that RCMs between introns bracketing circRNAs may induce larger hairpin
structures that promote the circularization of embedded exons
(Figure 1A).
In C. elegans, Introns Bracketing circRNAs Are Highly
Enriched for RCMs
For each intron pair that anked a circRNA, we aligned the
respective introns using Basic Local Alignment Search Tool
(BLAST; Experimental Procedures). RCMs were strongly enriched compared with length-matched introns (Figure 1B), with
a median of eight RCMs per intron pair (control pairs: one
RCM). We observed the same number and signicance of
matches on the same strand (sense matches [SMs]) of intron
pairs (Figure 1B). However, the exclusive occurrence of RCMs
on intron pairs bracketing circRNAs was signicantly enriched
compared with the exclusive occurrence of SMs (Fishers exact
test, p value < 0.0004; Figures 1B and S1C). This result suggests
that exon circularization is promoted by RCMs that can induce
basepairing between anking introns.
Predicting circRNAs from RCMs
We asked whether the occurrence of RCMs between circRNA
anking introns is signicant enough to predict circularized
exons. Therefore, we developed a simple probabilistic score H
to capture the likelihood that introns will basepair via RCMs
(Supplemental Experimental Procedures) and therefore promote
hairpin formation. Competition between RCMs is naturally taken
into account by this model. For each intron pair in a transcript
that contained one of the 1,111 circRNAs, we computed H.
Cell Reports 10, 170177, January 13, 2015 2015 The Authors 171

The resulting symmetric matrix (Figure 2A; Supplemental Experimental Procedures) shows that the intron pairs anking circularized exons stood out from all intron pairs. This was not the case
for the intron length-matched control transcripts (Figure 2A).
We then ranked, genome wide, all intron pairs of all transcripts
annotated in C. elegans by H and calculated the cumulative
fraction of known circRNAs as a function of intron-pair rank (Figure 2B). The top 4,200 (top 1%) of C. elegans intron pairs precisely bracketed 430 (38%) of the already annotated circRNAs,
a highly statistically signicant enrichment (hypergeometric test,
p value < 2.2 3 1016). Together, these results suggest that
RCMs provide highly signicantly improved accuracy in predicting circRNAs compared with intron length. However, we note
that (for C. elegans) the top BLAST score of RCMs in a pair of
introns yielded similar results.
To determine whether the presence of RCMs is a conserved
feature of circRNA formation, we next asked whether we could
predict circRNAs in human by intron sequence analysis.
The Presence of RCMs in Introns Flanking circRNAs Is a
Conserved Feature of circRNA Biogenesis
For human circRNAs, we used 1,067 exons overlapping circRNAs (Memczak et al., 2013). As in C. elegans, RCMs were highly
signicantly enriched (Fishers exact test, p value < 4 3 106;
172 Cell Reports 10, 170177, January 13, 2015 2015 The Authors

Figure 2. Prediction of circRNAs Based on

Sequence Analysis of Introns
(A) C. elegans. The matrix represents the average
circularization scores for intron pairs of circRNA
producing genes. (0,0) coordinates correspond to
circRNA anking introns. Upstream and downstream introns are enumerated with decreasing
and increasing numbers, respectively (with 0 being directly adjacent, 1/+1 being the preceding/
next intron, etc.; see Supplemental Experimental
Procedures).
(B) C. elegans. Different prediction methods are
used. The x axis is the fraction of all introns pairs
per gene sorted by H scores. Small (high) x values
denote intron pairs with high (low) scores for
bracketing a circRNA. The y axis is the fraction of
1,111 exonic circRNAs recovered by H score.
(C and D) Analyses for human circRNAs (Memczak
et al., 2013), analogous to (A) and (B).
See also Figures S4 and S5, and Tables S1, S2,
and S3.

Figures S1B and S1C). Again, the number

of intron pairs anking circularized exons
that contained only SM elements was
much lower compared with intron pairs
containing only RCMs (Figure S1C). The
median length of human RCMs was 20fold higher than that of C. elegans (Figures
S1D and S1E). We found that 88% of
the top-scoring RCMs overlapped with
ALU elements, 4.5% of RCMs overlapped
with L1/L2 repeats, and the remaining
RCMs did not overlap any annotated repetitive element and had the highest sequence conservation
(Figure S1F).
Further, we analyzed circRNA biogenesis conservation between mouse and human. We dened 71 circRNAs that are circularized in human and mouse (conserved circular expression,
Figure S1G). Since repetitive sequences such as ALU elements
are rapidly evolving, it is difcult to align them between species.
To circumvent this problem, we computed H scores by independently scoring the human and mouse introns (Figure S1G). As
controls, we dened (1) circRNAs that are circularized in humans
but have not been annotated as circular in mouse (nonconserved circRNAs), and (2) randomly selected exons that have
a similar bracketing intron length as human circRNAs, and have
positive H scores. Mouse H scores for conserved circRNAs
were signicantly higher (p value < 2.2 3 1016) compared with
the scores for exons that circularize only in human or the
length-matched controls (Figure S1G). Thus, the set of circRNAs
that are conserved between mouse and human are also
conserved in their biogenesis as quantied by H.
Human circRNAs Can Be Predicted Based on the
Sequence Composition of Their Flanking Introns
To predict circRNAs on a genome-wide level, we computed H
for all possible intron pairs in the UCSC RefSeq database

(Experimental Procedures). As a control, we used transcripts

with introns matched in length (Figure 2C; Supplemental Experimental Procedures) or predictions using the intron length or
simply the top BLAST score between intron pairs (Figure 2D).
For high-ranking intron pairs, the enrichment in circRNAs was
highest when the H score was used, followed by BLAST and
intron length. The top 20,000 (1% of all) predictions comprised
610 (9%) exonic circRNAs cataloged in the Memczak et al.
(2013), Jeck et al. (2013), and Zhang et al. (2014) data sets, a
highly signicant enrichment (hypergeometric test, p value < 2.2
3 1016). Predictions based on sequences associated with ALU
repeats yielded similar results (data not shown), showing that in
humans, ALU elements have likely contributed dominantly to
circRNA formation (Experimental Procedures). We note that
the number of false positives is very difcult to estimate because
many human circRNAs have not yet been reported. For example,
when we restricted the circRNA predictions to well-expressed
mRNAs in HEK293 cells, the success rate of our predictions
was much higher (Figure S1H). Therefore, experimental validation of predictions seems to be a better way to estimate falsepositive rates.
Predicted circRNAs Are Experimentally Validated in
HEK293 Cells
Since the circRNA predictions were not informed by expression
data, we expected that we could only validate circRNAs that are
isoforms of expressed transcripts. To test our predictions, we
considered the top 1% (H ranked) predicted circRNAs and
grouped them into three bins based on the expression of the
linear host transcripts in HEK293 cells (top 1% expressed,
medium, and bottom 50%). For experimental validation, we
randomly selected six, ten, and ve circRNAs from the high-,
medium-, and low-expression bins, respectively. As negative
controls, we used three well-expressed linear mRNAs and two
exons that are well expressed in HEK293 cells and have anking
introns but an H score of 0. The linear controls were not circularized (RNase R negative) and the two exons with an H score of
0 could not be amplied with divergent primers. However, 12
out of 16 circRNAs with top H scores and high/medium expression of the host transcripts were (1) resistant to RNase R and (2)
had the predicted head-to-tail junctions, as validated by Sanger
sequencing (Experimental Procedures; Figure 3), suggesting
that these 12 circRNAs exist in circularized form in HEK293
cells. Five candidates with decent H scores that we could
not conrm were, as expected, from the third, low-expression
bin. The predicted head-to-tail splicing of ten (out of 12) of
the RNase R-resistant circRNAs was validated by Sanger
sequencing (Experimental Procedures). We found that two
circRNA candidates were circularized (by RNase R assay) but
had an additional exon incorporated (as observed in Sanger
sequencing; Figure 3A). We note that ve of the tested circRNAs
had reasonable expression (1%10% of VCL expression; Experimental Procedures).
RNA Editing of Introns Flanking circRNAs
ADAR is a highly conserved RNA-editing enzyme that binds double-stranded RNA (Nishikura, 2010) and deaminates adenosine
bases to inosine. In humans, ADAR1 and ADAR2 interact with

double-stranded ALU repeats (Athanasiadis et al., 2004; Levanon et al., 2004; Nishikura, 2010; Ramaswami et al., 2012). We
compared A-to-I conversions (Ramaswami and Li, 2014) in
1,500 bp regions anking circRNA splice sites with (1) other
splice sites in transcripts that produce circRNAs, and (2)
length-matched introns (Figure 4A). A-to-I conversions nearby
circRNA splice sites were enriched compared with controls. In
general, ALU repeats in circRNAs anking introns were edited
signicantly higher compared with expression- and lengthmatched controls (Figures S2A and S2B). Since A-to-I editing
is a hallmark of basepaired RNA, we asked whether RCMs are
preferentially located at sites of editing. We compared the position of the RCMs (dened as the nearest RCMs that match between the pair of introns bracketing circRNA) with the same controls as before (Figure 4B). These results suggest that indeed Ato-I editing preferentially occurs at regions that are basepaired
and proximal (upstream and downstream 200600 nt) to the
splice sites of circularized exons.
To test whether ADAR proteins are involved in circRNA
biogenesis, we codepleted ADAR1 and ADAR2 in HEK293 cells
using RNAi (Supplemental Experimental Procedures). We used
two controls: untreated total RNA and codepletion of three
proteins of the APOBEC family (APOBEC3B, APOBEC3C, and
APOBEC3F), which are known to bind mRNAs (Baltz et al.,
2012). APOBEC enzymes are known to edit single-stranded
DNA or RNA (Vasudevan et al., 2013), but not double-stranded
RNA. The efcacy of the different knockdowns (KDs) was validated by western blotting and qRT-PCR (Figure S2C). Total
RNA extracted from the different experiments was depleted
from rRNAs and sequenced (Supplemental Experimental Procedures). We reproducibly observed that ADAR depletion resulted
in signicantly higher (p value < 2.2 3 1016) circRNA expression
compared with controls (Figure S2D), whereas the linear host
transcripts were less strongly affected (Figure 4C). For example,
84 circRNAs and 11 linear hosts were upregulated more than 2fold (Fishers exact test, p value 5.7 3 1013). This effect was
seen in independent biological replicates, as well as in a comparison of ADAR1/2 KD with APOBEC3 KD (Figures S2E and S2F).
We set out to validate these observations in independently
carried out ADAR1 and ADAR2 KD experiments (using the previous and an independent small interfering RNAs [siRNAs]
against ADAR1) followed by qRT-PCR assays (Experimental
Procedures; Figures 4D, 4E, and S2G). CircRNA candidates
were selected based on the increased fold changes observed
in sequencing data sets. circRNA expression was compared
with the expression of the respective linear host gene. Four
out of eight circRNAs (UBAC2, SMARCA, SPECC1, and
HIPK2) were upregulated upon ADAR1 depletion, whereas
the linear host RNAs did not show consistent expression
changes. PUM1 and CREBBP circRNA were upregulated to
the same level as their linear host transcripts. Consistent
with sequencing data, CDR1as was expressed 4-fold higher
in ADAR1 KD samples compared with control. Two circRNAs
(GAPVD1 and PDS5B) did not show the expression changes
observed in the sequencing data (Figures 4E and S2G). Based
on these ndings, we conclude that ADAR1 depletion can
induce upregulation of circRNAs independently of the expression level of the linear host circRNA.
Cell Reports 10, 170177, January 13, 2015 2015 The Authors 173

Figure 3. Experimental Validation of Human circRNAs Predicted by RCM Analyses

(A) CircRNA candidates from high-, medium-, and low-expression sets were assayed by qPCR with divergent primers and RNase R treatment. Linear control: VCL,
GAPDH, TFRC; positive control: a known circRNA (Memczak et al., 2013). Sanger sequencing of amplicons conrmed in all tested cases the predicted head-to-tail
junctions (candidates 9 and 11 contained an additional exon: marked with *). CircRNA candidates that were >10-fold resistant to RNase R treatment compared with
Vinculin were counted as positive (+). As negative controls, we selected exons from highly expressed genes with H score = 0. Error bars, SEM; n = 4.
(B) The chromatogram of a Sanger sequencing experiment conrms the presence of the predicted head-to-tail junction of the candidate circRNA from the
PTGES3 gene locus.
See also Figure S5 and Tables S1, S2, and S3.

To explore whether circRNA anking introns undergo extensive hyperediting (Carmi et al., 2011), we applied a computational pipeline that detects hyperediting events in the RNA
sequencing data sets used by Memczak et al. (2013) to predict
circRNAs, as well as in ADAR KD and control data sets. This
analysis (for details, see Porath et al. 2014) identied 165,000
unique hyperediting sites in the human genome (72.000,
49.000, 45.000, and 11.000 in Memczak et al. [2013] and
two control and ADAR KD samples, respectively). Notably, the
number of hyperedited sites identied in the ADAR KD sample
was 4- to 5-fold smaller compared with controls. We found
that 25% (19%) of circRNA upstream (downstream) introns
had at least one conversion, whereas only 6% of other introns
from the same genes had conversions. The number of conversions per base was also found to be elevated for the upstream
174 Cell Reports 10, 170177, January 13, 2015 2015 The Authors

(downstream) introns (Figures S3AS3C). A similar highly signicant enrichment of hyperediting events was detected for
C. elegans circRNA upstream introns (Supplemental Experimental Procedures; Figure S3D). Together, our data suggest
that A-to-I editing by ADAR is a universal hallmark of circRNA
biogenesis in animals.
DISCUSSION
Our analyses support a model in which RCMs between introns
that bracket an exon promote the circularization of that exon.
We note that this model is powerful enough to enable us to successfully predict and experimentally validate circRNAs. Our data
suggest that this model of circRNA biogenesis is conserved
across animals.

Figure 4. ADAR
Expression

An immediate follow-up question is, do RCMs have

sequence specicity? In general, our data do not support
sequence specicity in C. elegans, since the most prevalent
motif in RCMs between introns bracketing circRNAs was
clearly present in only 11% of all such RCMs. The equivalent
motif analysis in humans yielded (perhaps as expected) the
ALU element. For mouse circRNAs, we also found an ALUlike element. Sequence alignment between these motifs (Figure S4) suggests that there may be an ancient RNA sequence

Antagonizes

circRNA

(A) Normalized distribution of A-to-I conversions

upstream/downstream of the head/tail splice sites
of circRNAs (Memczak et al., 2013). As controls,
we selected length-matched introns and introns
from the same genes that produce circRNAs. Introns smaller than 1.5 kb were removed from the
analysis. The curves were smoothed within 10 bp
at each position and normalized to the total number of analyzed introns.
(B) Position of the nearest RCMs (or SMs) around
circRNA splice sites. For each intron pair, we
selected the top-scoring RCMs within the 1.5 kb
region around the splice site. The curves were
smoothed within a 20 bp window and normalized
to the total number of analyzed intron pairs.
(C) Comparison of the differential expression
(ADAR1 and ADAR2 codepletion against untreated control) of linear RNA and circRNA in two
independent replicates (merged). The y axis is the
log2 fold change of the reads (+5) supporting the
head-to-tail splice sites. The x axis is log2 fold
change of the reads supporting linear splice sites
of circRNA host genes. For the analysis, we
selected only circRNAs with at least ten reads in
ADAR1/2 KD or control samples. The numbers
tally the count of circRNAs/corresponding linear
hosts in each segment of the graph.
(D) Depletion of ADAR1 by RNAi. ADAR1 was
depleted from HEK293 cells for 96 hr using two
different siRNAs and compared with untreated
and ADAR2-depleted cells. Upper panel: qRTPCR of ADAR1 and ADAR2 transcripts, as well as
GAPDH and SNCA as controls. Lower panel:
western blot using ADAR1- and ADAR2-specic
antibodies. Error bars, SD; n = 3.
(E) Quantication of circRNAs and host transcript
levels. Using circRNA-specic primers (upper
panel) or exon-spanning primers that were not part
of the circRNA (lower panel), we quantied RNA
abundances relative to untreated cells by qRTPCR (Delta-Delta Ct values were normalized to
C. elegans spike-in). Error bars, SD; n = 3.
See also Figures S2, S3, and S5, and Tables S1,
S2, and S3.

that promotes circularization. This motif

awaits experimental testing.
RCMs can be generated by simple
random matches under neutral evolution. One possible scenario is that relatively quickly evolving RCMs constantly
generate a pool of circRNAs that can subsequently be
selected for and xed. Finding circRNAs that are under negative selection seems possible because, as we have shown,
RCMs between introns can be analyzed across species.
Thus, we think that cross-species comparisons of competing
RCMs will provide a way to nd circRNAs that may be functionally important. Indeed, our results show that our H score
is already able to link conserved circular expression to
conserved biogenesis.
Cell Reports 10, 170177, January 13, 2015 2015 The Authors 175

In humans, 88% of circRNAs have ALU repeats in their anking

introns, which, as we have shown, likely promote circularization
by RCMs. Therefore, it is interesting to speculate about the
possible functions of circRNAs, since ALU repeats have
expanded relatively recently in vertebrate evolution. We noticed
that in y, genes with neuronal functions often have long introns
and express circRNAs at relatively high levels (Ashwal-Fluss
et al., 2014). This also holds true for human brain tissues, where
dozens of circRNAs appear to be more highly expressed than
their (well-expressed) linear hosts (N.R., unpublished data). For
human circRNAs, we found that A-to-I editing events had a tendency to occur at intronic positions that were proximal (upstream
and downstream 200600 nt) to the splice sites of circularized
exons. These results link A-to-I editing to circRNA biogenesis
and predict that A-to-I editing events melt the stems that are
formed across introns that bracket a circRNA. Our KD experiments showed that indeed ADAR1 antagonizes circRNA biogenesis. The upregulation of circRNAs was stronger compared with
upregulation of their linear host transcripts, suggesting that
ADAR1 has a specic effect on circRNA biogenesis. However,
we cannot rule out that indirect effects explain upregulation of
circRNAs upon ADAR1 KD.
It is very interesting to think about the implications of these ndings. As a class, circRNAs may become more important in systems where ADAR1 expression is temporarily low. For example,
ADAR1 expression decreases in human embryonic stem cells
that are differentiating into the neuronal lineage (Osenberg et al.,
2010). Since circRNAs are unusually stable, this might be a mechanism to generate a long-term memory of past states. However,
one should not forget that other factors, such as muscleblind,
have been shown to promote the biogenesis of circRNAs
(Ashwal-Fluss et al., 2014). Therefore, circRNAs may also be
well expressed in systems in which ADAR1 expression is high.
Finally, it has been shown that circular splicing and linear splicing
can compete with each other (Ashwal-Fluss et al., 2014). Thus, it is
possible that regulation of circRNA biogenesis by ADAR1 serves
as a mechanism to regulate expression of the linear isoforms.
Note: while this paper was under review, two studies were
published describing RCM-dependent circularization in human
cells (Liang and Wilusz, 2014; Zhang et al., 2014). We also
acknowledge Starke et al. (2014), which presents related content
in this issue of Cell Reports.

(2013) with the additional ltering step of requiring unique alignments for
both of the read anchors prior to extension. Thereafter, we retained circRNAs
that overlap annotated internal splice sites.
Intron Alignments
We carried out intron alignments using BLAST (Altschul et al., 1990) with
the parameters -task blastn -word_size 6 for C. elegans and -task
blastn -word_size 11 for human. For further analysis, we considered
only alignments that exceeded BLAST score cutoffs of 20 and 100 for
C. elegans and human, respectively. For each intron pair, we calculated
circularization H as a top BLAST score multiplied by the probability of
forming at least one stem around the exon (Supplemental Experimental
Procedures).
Nonrepetitive RCMs were dened as BLAST matches that did not overlap
with any sequence present in UCSC RepeatMasker. Putative circRNAs with
nonrepetitive RCMs are shown in Table S3.
Conservation Analysis
To determine homologous exon groups, we used UCSC liftOver with minMatch = 0.1 option. We found that 71 out of 1,067 human circRNAs (Memczak
et al., 2013) overlapped annotated splice sites and were present in circular
form in mouse. We compared H score distributions using the Mann-Whitney
two-sided test.
qRT-PCR
We performed qPCR using the Maxima SYBR-Green/ROX qPCR master
mix (Thermo Scientic) and a StepOnePlus PCR system (Applied Biosystems). To detect putative head-to-tail junctions, we designed divergent
primers for each circRNA candidate. Ct values for mock/RNase R-treated
circRNAs were normalized to C. elegans spike-in RNA (for standard curves,
see Figures S5AS5C and Table S2). Amplicons were gel or bead puried
(Zymoclean gel DNA recovery kit [Zymo Research]; Agencourt AMPure
XP [Beckman Coulter]) and subjected to Sanger sequencing by LGC Genomics. Conrmed head-to-tail junctions are available in Figure S5D. A list of
the oligos is given in Table S1.
For more details, see the Supplemental Experimental Procedures.
ACCESSION NUMBERS
The data have been deposited in the NCBI Gene Expression Omnibus and are
available under accession numbers GSE63823 and SRP050149. circRNAs
identied in this study, as well as H scores, are available at http://www.
circbase.org.
SUPPLEMENTAL INFORMATION
Supplemental Information includes Supplemental Experimental Procedures,
ve gures, and three tables and can be found with this article online at
http://dx.doi.org/10.1016/j.celrep.2014.12.019.

CONCLUSIONS
AUTHOR CONTRIBUTIONS

In summary, in this work, we explored a specic model of

circRNA biogenesis and showed that this model seems to be
conserved across animals and is powerful enough to successfully predict circRNAs. Our data also link RNA editing to circRNA
biogenesis and suggest a function for ADAR1. These results
will enable the future detection and understanding of possible
circRNA functions, particularly in neuronal tissues.

A.I. performed all computational analyses except hyperediting. S.M. contributed experimental validation assays. E.W. designed and carried out ADAR
RNAi, RNA-seq, and qPCR experiments and initial analysis of the RNA-seq
data. H.P. and E.L. performed hyperediting detection. F.T. collected and prepared worm samples, carried out initial ADAR KDs, and helped in the design of
the validation experiments. M.O. annotated worm circRNAs. M.P. performed
an initial analysis of the RNA-seq data and was supervised by C.D. A.I. and
N.R. analyzed the data and wrote the paper with input from E.L.

EXPERIMENTAL PROCEDURES

ACKNOWLEDGMENTS

Identication of C. elegans circRNAs

We rst mapped worm sequencing data to rRNA to reduce the ribosomal
reads. The remaining reads were analyzed as described in Memczak et al.

A.I. and N.R. thank Sebastian Kadener (Hebrew University), Albrecht Bindereif
(University of Giessen), and Marvin Jens (N.R. lab) for helpful discussions. We
thank all members of the N.R. lab for discussions and support. This work was

176 Cell Reports 10, 170177, January 13, 2015 2015 The Authors

supported by German-Israeli-Foundation for Scientic Research and Development (G.I.F) and German Ministry for Education and Research (SatNet
program).

Levanon, E.Y., Eisenberg, E., Yelin, R., Nemzer, S., Hallegger, M., Shemesh,
R., Fligelman, Z.Y., Shoshan, A., Pollock, S.R., Sztybel, D., et al. (2004). Systematic identication of abundant A-to-I editing sites in the human transcriptome. Nat. Biotechnol. 22, 10011005.

Received: September 16, 2014

Revised: November 25, 2014
Accepted: December 9, 2014
Published: December 31, 2014

Liang, D., and Wilusz, J.E. (2014). Short intronic repeat sequences facilitate
circular RNA production. Genes Dev. 28, 22332247.

REFERENCES
Altschul, S.F., Gish, W., Miller, W., Myers, E.W., and Lipman, D.J. (1990). Basic
local alignment search tool. J. Mol. Biol. 215, 403410.
Ashwal-Fluss, R., Meyer, M., Pamudurti, N.R., Ivanov, A., Bartok, O., Hanan,
M., Evantal, N., Memczak, S., Rajewsky, N., and Kadener, S. (2014). circRNA
biogenesis competes with pre-mRNA splicing. Mol. Cell 56, 5566.
Athanasiadis, A., Rich, A., and Maas, S. (2004). Widespread A-to-I RNA editing
of Alu-containing mRNAs in the human transcriptome. PLoS Biol. 2, e391.
Baltz, A.G., Munschauer, M., Schwanhausser, B., Vasile, A., Murakawa, Y.,
Schueler, M., Youngs, N., Penfold-Brown, D., Drew, K., Milek, M., et al.
(2012). The mRNA-bound proteome and its global occupancy prole on protein-coding transcripts. Mol. Cell 46, 674690.
Braun, S., Domdey, H., and Wiebauer, K. (1996). Inverse splicing of a discontinuous pre-mRNA intron generates a circular exon in a HeLa cell nuclear
extract. Nucleic Acids Res. 24, 41524157.

Memczak, S., Jens, M., Elefsinioti, A., Torti, F., Krueger, J., Rybak, A., Maier,
L., Mackowiak, S.D., Gregersen, L.H., Munschauer, M., et al. (2013). Circular
RNAs are a large class of animal RNAs with regulatory potency. Nature 495,
333338.
Morse, D.P., Aruscavage, P.J., and Bass, B.L. (2002). RNA hairpins in noncoding regions of human brain and Caenorhabditis elegans mRNA are edited by
adenosine deaminases that act on RNA, Volume 99 (USA: Proc. Natl. Acad.
Sci), pp. 79067911.
Nishikura, K. (2010). Functions and regulation of RNA editing by ADAR deaminases. Annu. Rev. Biochem. 79, 321349.
Osenberg, S., Paz Yaacov, N., Safran, M., Moshkovitz, S., Shtrichman, R.,
Sherf, O., Jacob-Hirsch, J., Keshet, G., Amariglio, N., Itskovitz-Eldor, J., and
Rechavi, G. (2010). Alu sequences in undifferentiated human embryonic
stem cells display high levels of A-to-I RNA editing. PLoS ONE 5, e11173.
Pasman, Z., Been, M.D., and Garcia-Blanco, M.A. (1996). Exon circularization
in mammalian nuclear extracts. RNA 2, 603610.
Porath, H.T., Carmi, S., and Levanon, E.Y. (2014). A genome-wide map of hyper-edited RNA reveals numerous new sites. Nat. Commun. 5, 4726.

Capel, B., Swain, A., Nicolis, S., Hacker, A., Walter, M., Koopman, P., Goodfellow, P., and Lovell-Badge, R. (1993). Circular transcripts of the testis-determining gene Sry in adult mouse testis. Cell 73, 10191030.

Ramaswami, G., and Li, J.B. (2014). RADAR: a rigorously annotated database
of A-to-I RNA editing. Nucleic Acids Res. 42, D109D113.

Carmi, S., Borukhov, I., and Levanon, E.Y. (2011). Identication of widespread
ultra-edited human RNAs. PLoS Genet. 7, e1002317.

Ramaswami, G., Lin, W., Piskol, R., Tan, M.H., Davis, C., and Li, J.B. (2012).
Accurate identication of human Alu and non-Alu RNA editing sites. Nat.
Methods 9, 579581.

Dubin, R.A., Kazmi, M.A., and Ostrer, H. (1995). Inverted repeats are necessary
for circularization of the mouse testis Sry transcript. Gene 167, 245248.
Glazar, P., Papavasileiou, P., and Rajewsky, N. (2014). circBase: a database
for circular RNAs. RNA 20, 16661670.

Salzman, J., Gawad, C., Wang, P.L., Lacayo, N., and Brown, P.O. (2012). Circular RNAs are the predominant transcript isoform from hundreds of human
genes in diverse cell types. PLoS ONE 7, e30733.

Hansen, T.B., Wiklund, E.D., Bramsen, J.B., Villadsen, S.B., Statham, A.L.,
Clark, S.J., and Kjems, J. (2011). miRNA-dependent gene silencing involving
Ago2-mediated cleavage of a circular antisense RNA. EMBO J 30, 44144422.

Starke, S., Jost, I., Rossbach, O., Schneider, T., Schreiner, S., Hung, L.-H., and
Bindereif, A. (2014). Exon circularization requires canonical splice signals. Cel
Rep 10, this issue.

Hansen, T.B., Jensen, T.I., Clausen, B.H., Bramsen, J.B., Finsen, B., Damgaard, C.K., and Kjems, J. (2013). Natural RNA circles function as efcient
microRNA sponges. Nature 495, 384388.

Vasudevan, A.A.J., Smits, S.H.J., Hoppner, A., Haussinger, D., Koenig, B.W.,
and Munk, C. (2013). Structural features of antiviral DNA cytidine deaminases.
Biol. Chem. 394, 13571370.

Jeck, W.R., and Sharpless, N.E. (2014). Detecting and characterizing circular
RNAs. Nat. Biotechnol. 32, 453461.

Wang, P.L., Bao, Y., Yee, M.-C., Barrett, S.P., Hogan, G.J., Olsen, M.N., Dinneny, J.R., Brown, P.O., and Salzman, J. (2014). Circular RNA is expressed
across the eukaryotic tree of life. PLoS ONE 9, e90859.

Jeck, W.R., Sorrentino, J.A., Wang, K., Slevin, M.K., Burd, C.E., Liu, J., Marzluff, W.F., and Sharpless, N.E. (2013). Circular RNAs are abundant,
conserved, and associated with ALU repeats. RNA 19, 141157.

Zhang, X.-O., Wang, H.-B., Zhang, Y., Lu, X., Chen, L.-L., and Yang, L. (2014).
Complementary sequence-mediated exon circularization. Cell 159, 134147.

Cell Reports 10, 170177, January 13, 2015 2015 The Authors 177

Cell Reports

Report
ThermoMouse: An In Vivo Model
to Identify Modulators of UCP1 Expression
in Brown Adipose Tissue
Andrea Galmozzi,2,3 Si B. Sonne,1,3,4 Svetlana Altshuler-Keylin,1 Yutaka Hasegawa,1 Kosaku Shinoda,1
Ineke H.N. Luijten,1,5 Jae Won Chang,2 Louis Z. Sharp,1 Benjamin F. Cravatt,2 Enrique Saez,2,* and Shingo Kajimura1,*
1UCSF

Diabetes Center, Department of Cell and Tissue Biology, University of California, San Francisco, 35 Medical Center Way,
San Francisco, CA 94143, USA
2Department of Chemical Physiology and The Skaggs Institute for Chemical Biology, The Scripps Research Institute, La Jolla, CA 92037, USA
3Co-rst author
4Present address: Department of Biology, University of Copenhagen, Copenhagen 2200, Denmark
5Present address: The Wenner-Gren Institute, The Arrhenius Laboratories, Stockholm University, Stockholm 106 91, Sweden
*Correspondence: [email protected] (E.S.), [email protected] (S.K.)
http://dx.doi.org/10.1016/j.celrep.2014.10.066
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/).

SUMMARY

Obesity develops when energy intake chronically

exceeds energy expenditure. Because brown adipose tissue (BAT) dissipates energy in the form of
heat, increasing energy expenditure by augmenting
BAT-mediated thermogenesis may represent an
approach to counter obesity and its complications.
The ability of BAT to dissipate energy is dependent
on expression of mitochondrial uncoupling protein
1 (UCP1). To facilitate the identication of pharmacological modulators of BAT UCP1 levels, which
may have potential as antiobesity medications, we
developed a transgenic model in which luciferase
activity faithfully mimics endogenous UCP1 expression and its response to physiologic stimuli. Phenotypic screening of a library using cells derived from
this model yielded a small molecule that increases
UCP1 expression in brown fat cells and mice. Upon
adrenergic stimulation, compound-treated mice
showed increased energy expenditure. These tools
offer an opportunity to identify pharmacologic modulators of UCP1 expression and uncover regulatory
pathways that impact BAT-mediated thermogenesis.
INTRODUCTION
The epidemic of obesity poses a dire public health problem, for
obesity is a major risk factor for development of insulin resistance, type 2 diabetes, cardiovascular disease, and cancer.
Obesity is the result of a sustained energy imbalance in which
intake exceeds expenditure. Current antiobesity drugs work by
limiting energy intake, either through suppression of appetite
or inhibition of intestinal lipid absorption (Kim et al., 2014). These
medications are effective, but side effects often associated with
long-term use, such as depression or steatorrhea, limit patient

compliance. The discovery of brown adipose tissue (BAT) in

adult humans and its correlation with body mass index (Cypess
et al., 2009; Saito et al., 2009; van Marken Lichtenbelt et al.,
2009; Virtanen et al., 2009) suggests that active BAT may protect
from obesity. BAT dissipates energy in the form of heat, thus
increasing energy expenditure. It is thought that pharmacological activation of BAT thermogenesis may be an alternative
approach to alter energy balance, one complementary to existing obesity medications (Nedergaard and Cannon, 2010; Kajimura and Saito, 2014).
The ability of BAT to produce heat is dependent on expression
of the BAT-specic uncoupling protein 1 (UCP1). In response to
exposure to cold or a high-fat diet, UCP1 reduces the mitochondrial membrane potential and uncouples cellular respiration from
ATP synthesis, thereby generating heat. Because other UCP
proteins (e.g., UCP2 and UCP3) do not contribute to adaptive
thermogenesis (Golozoubova et al., 2001), UCP1 is thought to
be solely responsible for adaptive nonshivering thermogenesis.
UCP1-null mice are intolerant to cold (Enerback et al., 1997)
and develop obesity at thermoneutral conditions (Feldmann
et al., 2009). In contrast, transgenic expression of UCP1 in fat increases oxygen consumption in BAT and epididymal white adipose tissue (WAT) and reduces body weight gain (Kopecky
et al., 1995). Contrary to the mechanism of action of small-molecule mitochondrial uncouplers such as 2,4-dinitrophenol that
proved too toxic as weight loss agents (Grundlingh et al.,
2011), UCP1-mediated uncoupling is a highly regulated process
that requires direct binding of long-chain free fatty acids to UCP1
in response to physiologic cyclic AMP (cAMP) signaling (Fedorenko et al., 2012). A pharmacological approach to increase
UCP1 expression and activity in adipose tissue is thus likely to
constitute a safer avenue to enhance whole-body thermogenic
capacity and energy expenditure. To test this concept, it is of
great interest to identify small molecules that can stimulate
UCP1 expression in fat tissue.
Phenotypic screens with adipocytes have proven to be a
powerful method to isolate small molecules that ameliorate the
symptoms of metabolic syndrome through novel mechanisms

1584 Cell Reports 9, 15841593, December 11, 2014 2014 The Authors

of action (Waki et al., 2007; Dominguez et al., 2014). To facilitate

the identication of pharmacologic agents to modulate UCP1
expression in adipocytes, we have generated a transgenic reporter mouse, ThermoMouse, in which luciferase activity recapitulates the pattern of expression of UCP1 in vivo and allows
real-time visualization and quantication of UCP1 expression
in live animals. A chemical screen using brown adipocytes
derived from this model yielded a compound that can induce
UCP1 expression in cells and enhance UCP1 expression in vivo.
In response to adrenergic stimulation, mice treated with this
compound show increased energy expenditure. These results
demonstrate the utility of these models to identify pharmacological modulators of UCP1 levels. Discovery of compounds with
this ability is an important stride toward the goal of enhancing
BAT function in obese individuals with drug-like molecules.
RESULTS
Development of a Transgenic Model to Image UCP1
Expression In Vivo
To develop an in vivo reporter system to monitor endogenous
UCP1 expression in a noninvasive manner, we generated transgenic mice that express luciferase2 under the control of the
Ucp1 genetic locus. A luciferase2-T2A-tdTomato cassette was
inserted at the initiation codon of the Ucp1 gene in a 98.6 kb bacterial articial chromosome (BAC) containing the entire Ucp1
gene locus (Figure S1A) and proper targeting conrmed (Figure S1B). Next, we used the IVIS Spectrum Imaging System to
monitor luciferase activity in transgenic mice generated using
this BAC construct. 3D imaging detected robust luciferase
signals in interscapular BAT, perirenal BAT, and inguinal WAT
(Figure 1A). No signals were detected in adipose depots of wildtype mice (Figure S1C). To assess if luciferase activity recapitulated endogenous UCP1 protein levels, we measured luciferase
activity and UCP1 protein in BAT, WAT, liver, and muscle. A tight
correlation was found between luciferase activity and endogenous UCP1 protein expression (Figures 1B and 1C), indicating
that the reporter model mirrors the tissue distribution of UCP1.
To examine whether reporter mice responded to physiologic
stimuli known to induce UCP1 expression and BAT activity, we
monitored changes in luciferase activity during cold adaptation
(Figure 1D). Luciferase signal in interscapular BAT was low in
mice maintained at 28 C (Figure S1D) but increased robustly in
mice kept at 9 C for 24 hr (Figures 1D and 1E). Luciferase induction correlated with increases in UCP1 mRNA (Figure 1F) and
protein levels (Figure 1G). Similarly, subchronic (4 days) or acute
(1 day) treatment of reporter mice kept at 28 C with a specic b3
adrenergic receptor agonist (CL-316,243; 1 mg/kg) dramatically
enhanced the luciferase signal and endogenous UCP1 protein
levels in BAT (Figures 1H and 1I). To evaluate the response of
beige cells in inguinal WAT, imaging was performed from a
side angle. Strong luciferase activity was detected in inguinal
WAT in response to chronic CL-316,243 treatment (Figure 1J).
The increase in signal was paralleled by robust induction of
UCP1 protein in this depot (Figure 1K). These results show that
our Ucp1 luciferase reporter accurately indicates changes in
UCP1 elicited by physiological responses and is a useful tool
to quantify changes in UCP1 expression in vivo.

A Cell-Based Screening Platform to Monitor UCP1

Protein Expression
Next, to establish a cell-based system for quantifying UCP1
expression in fat cells suited for screening of small-molecule
or genomic libraries, we generated immortalized preadipocyte
lines from Ucp1 luciferase transgenics. From six cell lines
derived from interscapular BAT, we selected one that showed
high adipogenic capacity and signicant levels of UCP1 and
luciferase expression. Because PPARg ligands can stimulate
UCP1 expression (Sears et al., 1996), we tested the response
of this line to rosiglitazone. Luciferase activity was induced in a
dose-dependent manner and correlated tightly with endogenous UCP1 protein levels (Figures 2A and 2B). Treatment
with the cAMP-signaling activator forskolin also increased
luciferase activity and UCP1 protein expression (Figures 2C
and 2D).
To conrm that the Ucp1 luciferase reporter cell line preserved
the response to BAT activators, and that it could report sequential changes in UCP1 expression, nude mice were implanted with
Ucp1 luciferase preadipocytes subcutaneously and treated
6 days posttransplantation with saline or rosiglitazone (10 mg/
kg per day) for 7 days. Luciferase activity in transplants of rosiglitazone-treated mice was increased signicantly at 4 days of
treatment and thereafter (Figures 2E and 2F). Transplanted preadipocytes in mice treated with rosiglitazone formed discrete
adipose tissue containing multilocular adipocytes (Figure 2G,
upper left). These adipocytes were positive for UCP1 protein
(Figure 2G, upper right) and GFP (i.e., tdTomato; Figure 2G,
lower left), indicating that transplanted cells retained brown adipogenic capacity in vivo. These observations indicate that this
cell-based system is a robust surrogate to assess endogenous
UCP1 expression.
A Small-Molecule Screen Identies a Regulator of UCP1
Expression
To test the ability of our monitoring systems to identify regulators
of UCP1 expression, we performed a phenotypic screen using a
modestly sized library of small molecules, primarily carbamates
and triazole ureas (Adibekian et al., 2011; Bachovchin et al.,
2010). Ucp1 luciferase brown preadipocytes were seeded and
differentiated in 96-well plates. Mature adipocytes (day 8) were
incubated with compounds and luciferase activity quantied
16 hr later (Figure 3A). To evaluate the power of the assay to identify compounds that modulate UCP1 levels in either direction,
four hits with differing properties (activators and inhibitors)
were selected for study (Figure 3B). As it is often the case with
reporter-based screens, two hits could not be validated and
may represent, for example, nonspecic stabilizers of luciferase
signal. Of those that conrmed, compound 4 robustly increased
luciferase activity (Figure 3C). In contrast, compound 3 signicantly reduced luciferase signal. Importantly, compound 4
increased endogenous UCP1 protein expression to a similar
level to that induced by rosiglitazone (Figure 3D). None of the
compounds had any effect on brown preadipocytes (Figure S2).
Compound 4 (WWL113 in original nomenclature) was also a hit
in a different, image-based screen we recently described, in
which white preadipocytes were treated chronically (8 days)
during differentiation (Dominguez et al., 2014). WWL113 inhibits

Cell Reports 9, 15841593, December 11, 2014 2014 The Authors 1585

(legend on next page)

1586 Cell Reports 9, 15841593, December 11, 2014 2014 The Authors

two serine hydrolases expressed in adipocytes, carboxylesterase 3 (Ces3 or Ces1d) and Ces1f (CesML1). However, the ability
of WWL113 to induce UCP1 expression in brown adipocytes
does not appear to be mediated by Ces3/1f inhibition, for a
structurally distinct Ces3 inhibitor (WWL229) failed to have the
same effect on UCP1 expression, whereas the urea version of
WWL113 (WWL113U), which does not inhibit Ces3, retained
the ability to induce UCP1 (Figure S3A). Nonetheless, we
concluded that WWL113 could be a valuable chemical probe
to further validate our reporter systems.
Induction of UCP1 Expression by WWL113 Relies on
PPARa Signaling
To characterize the effect of WWL113 on endogenous UCP1
expression, we treated cultured brown adipocytes with several
doses of WWL113. At a dose of 1 mM and higher, WWL113 signicantly increased UCP1 protein expression (Figure 4A). The
effect of WWL113 on UCP1 protein levels was largely due to
activation of Ucp1 transcription, because WWL113 powerfully
increased expression of Ucp1 mRNA (Figure 4B). WWL113 treatment also stimulated expression of mRNAs for other thermogenic genes, such as Cidea, Pgc1a, and Cox7a1 (Figure 4B).
WWL113 treatment had no effect on expression of the adipogenic marker Adiponectin. WWL113 treatment for 24 hr was sufcient to activate endogenous Ucp1 mRNA expression without
affecting expression of multiple adipogenic markers (Figures
4C and S3B), indicating that WWL113 enhanced the BAT-selective thermogenic gene program in a cell-autonomous manner
without affecting adipogenesis per se. To test the functional consequences of WWL113 treatment, we examined the extent to
which WWL113 could sensitize brown adipocytes to physiologic
activators of BAT such as norepinephrine. WWL113 pretreatment enhanced the increase in Ucp1 mRNA expression normally
induced by norepinephrine (Figure 4D). More importantly,
WWL113-treated cells showed greater total and uncoupled
respiration in response to norepinephrine and greater uncoupled
basal respiration (Figures 4E and S3C).
Next, we explored the mechanism by which WWL113
increased Ucp1 transcription. Because PPARa is a critical regulator of thermogenic gene expression in BAT (Barbera et al.,
2001), we hypothesized that the action of WWL113 in BAT
could be mediated via PPARa. To test this notion, primary brown
adipocytes were treated with a selective PPARa antagonist
(GW6471) and/or a PPARg antagonist (GW9662) in the presence
or absence of WWL113. The capacity of WWL113 to increase
Ucp1 mRNA expression was largely blunted by treatment with

GW6471, but not with GW9662 (Figure 4F). The inhibitory effect
of GW6471 on WWL113-induced Ucp1 mRNA expression was
not altered when cells were cotreated with GW9662. These
results indicate that the ability of WWL113 to enhance Ucp1
expression is principally dependent on PPARa, but not PPARg,
activity. Because WWL113 is not a direct PPARa activator
(Dominguez et al., 2014), we tested the extent to which cotreatment of differentiated brown adipocytes with WWL113 and a
PPARa agonist (GW9578) would further boost Ucp1 mRNA
levels. Cotreatment with WWL113 and GW9578 modestly but
signicantly increased expression of Ucp1 relative to treatment
with either compound alone (Figure 4G), suggesting that
WWL113 cooperates with the PPARa pathway to enhance
UCP1 expression.
WWL113 Increases UCP1 Expression and the
Thermogenic Response In Vivo
We next tested if WWL113 could enhance UCP1 expression
in vivo. Ucp1 luciferase mice were treated with vehicle or
WWL113 (50 mg/kg once daily) for 5 days. This dose has been
shown to be effective in mice (Dominguez et al., 2014). A robust
and signicant increase (5-fold) in Ucp1-driven luciferase
expression was detected in the interscapular BAT depots of
transgenic mice treated with WWL113 (Figures 5A and 5B).
To conrm these ndings, we examined the ability of WWL113
to enhance thermogenic gene expression in C57BL/6J mice
treated with the compound for 5 days. WWL113 treatment
induced signicant increases in mRNA expression of Ucp1 and
thermogenic genes, such as Pgc1a and Dio2, in the BAT of
wild-type mice (Figure 5C). No change in the general marker
Pparg was seen. Importantly, UCP1 protein expression was
highly induced in vivo by WWL113 (Figure 5D). In contrast, no difference in mRNA expression of Ucp1, Pgc1a, Dio2, and Pparg
was observed in inguinal WAT of WWL113-treated mice (Figure S5A), indicating that the effects of WWL113 on the thermogenic gene program may be specic to brown fat.
Finally, we examined the effects of WWL113 on whole-body
energy expenditure. Mice treated with WWL113 for 7 days
showed no differences in basal energy expenditure, but upon
adrenergic stimulation (CL-316,243 injection), they responded
with a more robust increase in energy expenditure than controls
(Figure 5E). WWL113 did not affect locomotor activity (Figure 5F),
food intake (Figure 5G), or heart rate (Figure 5H). These data indicate that WWL113 increased UCP1 levels in vivo to a functionally
meaningful degree, increasing the adaptive thermogenic capacity of treated mice.

Figure 1. Luciferase Imaging of UCP1 Expression In Vivo

(A) 3D reconstruction of luciferase signals in Ucp1 luciferase reporter mice. Adipose depots with specic luciferase signal are indicated.
(B) Quantication of luciferase in adipose depots, skeletal muscle, and liver of mice kept at room temperature. Values normalized to protein content.
(C) UCP1 protein expression in tissue lysates from the mouse in (B).
(D) Luciferase activity in Ucp1 luciferase reporter mice kept at 28 C and subsequently kept at 9 C for 24 hr. Representative mice are shown.
(E) Quantication of luciferase signal in interscapular BAT of Ucp1 luciferase reporter mice shown in (D) (n = 9). ***p < 0.001. Data are expressed as mean SEM.
(F) Ucp1 mRNA expression in interscapular BAT of Ucp1 luciferase reporter mice kept at 28 C and 9 C for 24 hr. ***p < 0.001. Data are expressed as mean SEM.
(G) UCP1 protein expression in interscapular BAT of transgenic mice analyzed in (F).
(H) Representative image of luciferase signal in Ucp1 luciferase reporter mice treated with saline or CL-316,243 for 1 day (acute) or 4 days (subchronic; n = 3).
(I) UCP1 protein expression in BAT of transgenic mice analyzed in (H).
(J) Representative images of luciferase signal in inguinal WAT depots of Ucp1 luciferase reporter mice treated with saline or CL-316,243 for 4 days.
(K) UCP1 protein expression in inguinal WAT depots of Ucp1 luciferase mice treated with saline or CL-316,243 for 1 day or 4 days.

Cell Reports 9, 15841593, December 11, 2014 2014 The Authors 1587

(legend on next page)

1588 Cell Reports 9, 15841593, December 11, 2014 2014 The Authors

Figure 3. Small-Molecule Screen to Identify

Regulators of UCP1 Expression
(A) Screen scheme. Ucp1 luciferase brown preadipocytes were differentiated in 96-well plates
and treated at day 8 with compounds for 16 hr
(n = 3).
(B) Screen performance. Mean luciferase activity
of compounds (blue circles) plotted relative to the
value of DMSO-treated Ucp1 luciferase adipocytes (100%; green circles). Rosiglitazone served
as positive control (red circles). Ucp1 luciferase
preadipocytes (purple circles) were used to
determine background and induction of signal
upon differentiation. Hits selected for evaluation
(blue circles, red outline) are numbered according
to the scheme used in validation experiments.
(C) Luciferase activity in differentiated Ucp1 luciferase brown adipocytes treated with compounds
for 5 days. Rosiglitazone (0.5 mM) and CL-316,243
(10 nM) served as positive controls (n = 3). *p <
0.05; **p < 0.01; ***p < 0.001 versus control. Data
are expressed as mean + SD.
(D) UCP1 protein expression in cells from (C).

DISCUSSION
We have developed an in vivo monitoring system that allows to
quantitatively track sequential changes in UCP1 expression
within the same individual. Because UCP1 expression shows a
high level of interindividual variation (Boeuf et al., 2002) and
changes dynamically during circadian oscillation (Gerhart-Hines
et al., 2013), seasonal changes (Au-Yong et al., 2009), and
aging (Rogers et al., 2012), this model may prove of wide utility.
18
uoro-labeled 2-deoxy-glucose positron emission tomography (18FDG-PET) scanning has been applied to assess BAT activity in rodents and humans (Cypess et al., 2009; Saito et al.,
2009; van Marken Lichtenbelt et al., 2009; Virtanen et al.,
2009), but detection of 18FDG-PET signals in BAT depends
entirely on glucose uptake. In contrast, the level of UCP1 expression in BAT is a more-direct measure of the thermogenic capacity of this tissue (Nedergaard and Cannon, 2010). Hence, the
transgenic UCP1 reporter we describe provides an opportunity
to identify signaling pathways and transcriptional events that
control thermogenic capacity in brown adipocytes in vivo. Unlike
prior UCP1 models (Cassard-Doulcier et al., 1993), it also en-

ables characterization of modulators of BAT function in real

time in live animals.
Ucp1 luciferase brown adipocyte lines derived from this transgenic retain the characteristics of bona de brown fat cells.
A phenotypic screen using these cells identied a compound,
WWL113, that can increase UCP1 expression in vitro and in vivo.
It is important to note that UCP1s uncoupling activity is dependent
on sympathetic nerve activation and increased intracellular cAMP
levels. As BAT activity is stimulated by cold exposure, long-chain
free fatty acids supplied from cAMP-induced lipolysis and from
the circulation directly bind UCP1 and serve as a substrate
to transport protons into the mitochondrial matrix (Fedorenko
et al., 2012). Thus, small molecules that solely stimulate UCP1
expression are unlikely to induce substantial thermogenesis. In
agreement with this notion, we found that the effect of WWL113
treatment on cellular respiration was greater when cells were stimulated with norepinephrine. In that setting, this chemical probe
signicantly increased total and uncoupled cellular respiration.
More importantly, mice treated with WWL113 showed considerably enhanced energy expenditure, but this increase required
adrenergic stimulation. WWL113-treated mice showed no

Figure 2. Cell-Based System to Monitor UCP1 Expression

(A) Luciferase activity in immortalized Ucp1 luciferase brown adipocytes. Differentiated adipocytes were treated with DMSO (control) or rosiglitazone for 5 days
(n = 3).
(B) UCP1 protein expression in cells analyzed in (A).
(C) Luciferase activity in differentiated Ucp1 luciferase brown adipocytes treated with DMSO (control), forskolin, or rosiglitazone (0.5 mM) for 5 days (n = 3).
(D) UCP1 protein expression in cells analyzed in (C).
(E) Luciferase activity monitored at days 0, 1, 4, and 7 after the start of rosiglitazone treatment in mice implanted with Ucp1 luciferase immortalized preadipocytes.
Saline (n = 4) or rosiglitazone (n = 6; 10 mg/kg) treatment started 6 days postimplantation. Representative mice are shown.
(F) Sequential changes in luciferase activity measured in fat transplants from mice in (E).
(G) Hematoxylin and eosin and immunouorescent stains for UCP1 or GFP (i.e., tdTomato) in transplants from mice treated with rosiglitazone. Merged UCP1 and
GFP image counterstained with DAPI. The scale bar represents 50 mm. H&E, hematoxylin and eosin staining.
*p < 0.05; **p < 0.01; ***p < 0.001. Data are expressed as mean SEM.

Cell Reports 9, 15841593, December 11, 2014 2014 The Authors 1589

Figure 4. Effect of WWL113 on UCP1

Expression and Cellular Respiration
(A) UCP1 protein expression in differentiated Ucp1
luciferase brown adipocytes treated with WWL113
for 5 days. Rosiglitazone (0.5 mM) served as positive control.
(B) Expression of thermogenic genes in differentiated Ucp1 luciferase brown adipocytes treated
with WWL113 (10 mM) for 5 days. Rosiglitazone
(0.5 mM) served as positive control (n = 3). *p <
0.05; ***p < 0.001 versus control.
(C) Ucp1 mRNA expression in primary differentiated brown adipocytes treated with WWL113
(10 mM) or rosiglitazone (5 mM) for 24 or 48 hr (n =
3). ***p < 0.001 versus control.
(D) Ucp1 mRNA expression in primary differentiated brown adipocytes treated with WWL113
(10 mM) or rosiglitazone (5 mM) for 48 hr. Norepinephrine (0.1 mM) was added 8 hr prior to harvest
(n = 3). **p < 0.01; ***p < 0.001 versus control.
(E) Total and uncoupled (oligomycin-insensitive)
respiration of differentiated brown adipocytes (5 3
105 cells/sample) treated with WWL113 (10 mM) in
the presence or absence of norepinephrine
(0.1 mM; n = 3 to 4). *p < 0.05 versus control.
(F) Ucp1 mRNA expression in differentiated brown
adipocytes treated with vehicle or WWL113
(10 mM) for 24 hr with or without 30 min pretreatment with a PPARa-selective antagonist
(GW6471; 3 mM) and/or a PPARg-selective
antagonist (GW9662; 10 mM; n = 3). ***p < 0.001
versus control. ###p < 0.001 versus WWL113treated cells.
(G) Ucp1 mRNA expression in differentiated brown
adipocytes treated with vehicle, WWL113 (10 mM),
a PPARa-selective agonist (GW9578), or the
combination for 24 hr (n = 4). *p < 0.05; ***p <
0.001.
Data are expressed as mean + SD.

differences in locomotor activity, food intake, or heartbeat, indicating that the compound does not enhance basal sympathetic
drive. Together with the nding that WWL113 increases UCP1
expression in brown fat cells in a cell-autonomous manner, these
results suggest that WWL113 boosts energy expenditure primarily
by increasing the content of UCP1 in BAT that can be activated by
physiologic stimuli such as cold.
We previously reported that chronic administration (2 months)
of WWL113 to obese-diabetic mice reduced body weight gain,
improved systemic glucose and lipid homeostasis, and cleared
hepatic steatosis (Dominguez et al., 2014). We ascribed the
effects of WWL113 to inhibition of Ces3 in WAT and liver. Using
a distinct phenotypic screen as a starting point, in this study, we
have found that WWL113 can have Ces3-independent effects in

BAT. WWL113 is not a direct activator of

PPARa (Dominguez et al., 2014), but its
effects on UCP1 expression in brown fat
cells depend to a large extent, though
not completely, on PPARa signaling.
Although the molecular target(s) for
WWL113s action in BAT remains to be
claried, this tool compound has nonetheless demonstrated
the utility of our UCP1-monitoring systems to identify pharmacologic activators of UCP1 expression.
BAT is the major adipose depot that contains UCP1-positive
adipocytes (classical brown adipocytes), but rodents and humans also possess an inducible type of thermogenic fat cells,
termed beige or brite adipocytes (Sharp et al., 2012; Wu et al.,
2012; Cypess et al., 2013; Lidell et al., 2013). UCP1-positive
beige adipocytes emerge within WAT in response to external
cues, such as sustained cold exposure or exercise. Beige adipocytes are considered promising reservoirs for enhancing energy
expenditure, but current technologies (e.g., 18FDG-PET and MRI
scans) do not possess enough resolution to detect beige cells
in vivo. Our data indicate that our Ucp1 luciferase mouse may

1590 Cell Reports 9, 15841593, December 11, 2014 2014 The Authors

Figure 5. WWL113 Increases UCP1 Expression in Mice

(A) Luciferase activity in Ucp1 luciferase reporter
mice treated daily with WWL113 (50 mg/kg) or
vehicle for 5 days (n = 5). Representative mice are
shown.
(B) Quantication of luciferase signal in interscapular BAT of mice treated as in (A). Values
normalized to protein content and shown as fold
change relative to vehicle.
(C) Expression of thermogenic genes in interscapular BAT of C57BL/6 mice treated daily with
WWL113 (50 mg/kg) or vehicle for 5 days (n = 5).
*p < 0.05; **p < 0.01
(D) UCP1 protein expression in interscapular BAT
of C57BL/6 mice analyzed in (C).
(E) VO2 of wild-type mice treated daily with
WWL113 (50 mg/kg) or vehicle for 7 days (n = 6).
CL-316,243 (1 mg/kg) was injected to examine the
response to adrenergic stimulation. **p < 0.01.
N.S., not signicant.
(F) Locomotor activity of mice in (E).
(G) Food intake of mice in (E).
(H) Heart rate of mice in (E).
Data are expressed as means SEM.

serve as a tool to monitor the effects of compounds and biological factors on beige cells.
Although they share the thermogenic ability of brown adipocytes, beige adipocytes have a distinct, heterogeneous developmental origin that is not fully understood. Beige adipocytes in
multiple WAT depots can arise from Myf5-positive and negative
cells (Sanchez-Gurmaches and Guertin, 2014), and a subset of
inguinal beige adipocytes originates from a smooth-muscle
lineage (Long et al., 2014). Prior work has shown that regulation
of Ucp1 is distinct in the two types of thermogenic adipocytes
(Guerra et al., 1998; Koza et al., 2000; Xue et al., 2007). We found
that WWL113 could activate UCP1 expression in BAT, but
not in inguinal WAT, implying that WWL113-initiated signaling
events that regulate UCP1 expression may be unique to brown
adipocytes.
BAT and liver are the major organs responsible for triglyceride
uptake (Bartelt et al., 2011), and emerging evidence points to a
close connection between BAT activity, liver function, and systemic lipid homeostasis. For example, defects in thermogenesis
caused by BAT-specic knockout of the enzyme euchromatic
histone-lysine N-methyltransferase 1 resulted in hepatic steatosis and insulin resistance, even when weight-matched mice

were compared (Ohno et al., 2013).

Conversely, activation of BAT thermogenesis, for example by overexpression
of PTEN, reduced hepatic lipid accumulation and enhanced systemic insulin sensitivity (Ortega-Molina et al., 2012). Thus,
the therapeutic benets of increased
BAT thermogenesis may not be limited
to effects on weight but could include
improvements in lipid homeostasis and
whole-body insulin sensitivity. The set of
tools we have developed should aid the development of pharmacologically tractable approaches to activate BAT function as a
therapy against obesity and its complications.
EXPERIMENTAL PROCEDURES
Animals
Experiments were approved by the Institutional Animal Care and Use Committees of the University of California, San Francisco, and The Scripps Research
Institute. Unless stated, mice were kept at room temperature. WWL113 was
administered either intraperitoneally once daily (50 mg/kg in a 4:1 PEG300:Tween 80 vehicle solution) or orally (50 mg/kg in 0.5% hydroxypropylmethylcellulose). At the conclusion of treatment, BAT, WAT, skeletal muscle, and liver
were snap frozen for luciferase assays and RNA and protein analysis. Energy
balance studies were performed over the course of 7 days using a Comprehensive Lab Animal Monitoring System (Columbus Instruments) as described
(Ohno et al., 2013). Data were normalized to body weight, as there were no differences between groups. Heart rate in the conscious state was measured by
the indirect tail cuff method with a SC1000 MSP (Hatteras Instruments).
Generation of Ucp1 Luciferase Reporter Mice
A 98.6 kb BAC (bMQ353d13; Source BioScience) containing the entire Ucp1
gene locus was obtained and a luciferase2-T2A-tdTomato reporter cassette
(Addgene 32904) inserted at the initiation codon of the Ucp1-coding sequence
located in exon 1 using BAC recombineering techniques (Warming et al.,

Cell Reports 9, 15841593, December 11, 2014 2014 The Authors 1591

2005). Ucp1 luciferase BAC DNA was microinjected into single-cell FVB
embryos and transgenic founders and their offspring identied by PCR
(primers provided in Table S1). Segregation patterns indicate that the transgene inserted into the Y chromosome. Transgenics display no decrease in
fertility or any other abnormalities.
In Vivo Luciferase Imaging
Luciferase activity was monitored using an IVIS Spectrum Instrument (Caliper
Life Sciences). For 3D reconstruction, six images (exposure time: 180 s;
binning: L; F/Stop:1; emission lters: 560660 nm; eld of view: C) were
collected starting 8 min after injection of 150 mg/kg luciferin (Goldbio). In other
experiments, one image per scan (exposure time: 300 s; binning: M; F/Stop:1;
emission lters: open; eld of view: D) was acquired 15 min after luciferin injection. 3D reconstructions and luciferase activity were calculated using Living
Image Software (Caliper Life Sciences).
Immortalized Ucp1 Luciferase Adipocyte Lines
The stromal vascular fraction from interscapular BAT of 3-week-old male
UCP1 luciferase transgenic mice was isolated (Ohno et al., 2012) and cells
infected with a retrovirus expressing large T antigen (pBabe SV40 Large T antigen; Addgene) and selected in puromycin (2 mg/ml). Immortalized preadipocytes were cultured in Dulbeccos modied Eagles medium (DMEM), 10%
fetal bovine serum (FBS), penicillin, and streptomycin. Upon 100% conuence
(day 0), differentiation was induced with medium containing 10% FBS, 5 mg/ml
insulin, 1 nM T3, 0.125 mM indomethacin, 2 mg/ml dexamethasone, and
0.125 mM 3-isobutyl-1-methylxanthine for 2 days. From day 2 on, cells were
cultured only in the presence of insulin and T3.
Phenotypic Screen
A library of approximately 300 compounds (Adibekian et al., 2011; Bachovchin
et al., 2010) was screened. Immortalized Ucp1 luciferase brown preadipocytes
were seeded in 96-well plates and differentiated as described above. At day 8,
cells were exposed to compounds (5 mM for carbamates, 2.5 mM for triazole
ureas) in serum-free DMEM. Rosiglitazone (0.5 mM) served as positive control.
After 16 hr, media was replaced with Glo Lysis Buffer (Promega) and luciferase
activity quantied in a PHERAstar reader (BMG Labtech). Activity was normalized relative to the signal in DMSO-treated cells in each plate. Assays were
performed in triplicate. Compounds inducing >50% increase or decrease in
luciferase activity were selected for follow-up.
Luciferase Assays
Cells or tissues were lysed in Cell Culture Lysis Reagent (Promega) and luciferase activity quantied using the Luciferase Assay System (Promega). Activity
was measured in an Optocomp I reader (MGM Instruments) and normalized to
total protein content.
Gene Expression
Total RNA was isolated using RiboZol reagents (AMRESCO) and reversed
transcribed using an iScript cDNA synthesis kit (Bio-Rad) and quantitative
real-time PCR performed using SYBR green and an ABI ViiA7 machine. Relative mRNA expression was determined by the DD-Ct method with TATAbinding protein as a normalization control. Primer sequences are provided in
Table S1.
Metabolic Studies
Whole-body energy expenditure was measured at ambient temperature using
a Comprehensive Lab Animal Monitoring System (Columbus Instruments) at
the University of California, San Francisco (UCSF) animal metabolic core facility. Wild-type mice were treated daily with WWL113 (50 mg/kg) or vehicle for
7 days (n = 6). The mice were injected intraperitoneally with a b3-adrenergic
receptor-specic agonist CL-316,243 at a dose of 1 mg/kg to examine the
response to adrenergic stimulation. VO2 was normalized by body weight.
Western Analysis
Proteomes were separated by SDS-PAGE. Antibodies used: UCP1 (Abcam;
ab10983); Akt (Cell Signaling Technology; 9272); a-tubulin (Sigma; T8203);
and b-actin (Sigma-Aldrich; AC-15).

Cellular Respiration
Oxygen consumption rate was measured in a MT200A Cell Respirometer
(Strathkelvin), as previously described (Kajimura et al., 2009). Briey, differentiated brown adipocytes treated with DMSO or WWL113 (10 mM) for 48 hr were
trypsinized and incubated in serum-free medium in the presence or absence of
norepinephrine. Uncoupled and nonmitochondrial cellular respiration were
measured using oligomycin (1 mM) and antimycin A (1 mM).
Fat Transplantation
Preadipocytes were implanted as described (Kajimura et al., 2009). Briey,
cultured immortalized Ucp1 luciferase preadipocytes were trypsinized,
washed, and resuspended in PBS. Preadipocytes in a volume of 300 ml (4
3 107 cells) were injected subcutaneously into NCr nude mice (Taconic).
Six days after transplantation, mice were injected with either saline (n = 4) or
rosiglitazone (n = 6; 10 mg/kg) twice daily for 7 days. Luciferase activity was
monitored on days 0, 1, 4, and 7.
Histology
Tissues were xed in 4% paraformaldehyde and embedded in parafn. Sections (7 mm) were analyzed with hematoxylin and eosin staining and immunouorescence to detect UCP1 (ab10983 1:1,000 in 5% goat serum) or GFP
(GFP-1020 1:500 in 5% goat serum) as described (Sharp et al., 2012).
Statistics
Statistical signicance was dened as p < 0.05 and determined by two-tailed
Student t tests, Wilcoxon, or ANOVA, with Dunnes multiple comparison post
hoc analysis.
SUPPLEMENTAL INFORMATION
Supplemental Information includes four gures and one table and can be
found with this article online at http://dx.doi.org/10.1016/j.celrep.2014.10.066.
AUTHOR CONTRIBUTIONS
A.G., S.B.S., E.S., and S.K. designed experiments. J.W.C. and B.F.C. provided
libraries and synthesized compounds. S.K. and S.B.S. generated transgenic
model and cell lines. A.G. performed screen and follow-up. A.G., S.B.S.,
S.A.-K., Y.H., K.S., and S.K. performed in vivo and vitro experiments. L.Z.S.
and I.H.N.L. provided technical help. A.G., S.B.S., E.S., and S.K. wrote the
manuscript.
ACKNOWLEDGMENTS
We thank Haemin Hong, Kathleen Jay, Christophe Paillart, and Denis Glenn for
assistance. Work was supported by NIH grants DK087853 and DK97441 to
S.K. and DK099810 and CA179489 to E.S. S.K. acknowledges support from
the DERC center grant (DK63720), UCSF PBBR program, the Pew Charitable
Trust, and PRESTO from Japan Science and Technology Agency. S.B.S. was
supported by a fellowship from the Alfred Benzon Foundation and A.G. by
fellowship 14POST18200019 from the American Heart Association.
Received: April 2, 2014
Revised: October 13, 2014
Accepted: October 24, 2014
Published: November 26, 2014
REFERENCES
Adibekian, A., Martin, B.R., Wang, C., Hsu, K.L., Bachovchin, D.A., Niessen,
S., Hoover, H., and Cravatt, B.F. (2011). Click-generated triazole ureas as
ultrapotent in vivo-active serine hydrolase inhibitors. Nat. Chem. Biol. 7,
469478.
Au-Yong, I.T., Thorn, N., Ganatra, R., Perkins, A.C., and Symonds, M.E. (2009).
Brown adipose tissue and seasonal variation in humans. Diabetes 58, 2583
2587.

1592 Cell Reports 9, 15841593, December 11, 2014 2014 The Authors

Bachovchin, D.A., Ji, T., Li, W., Simon, G.M., Blankman, J.L., Adibekian, A.,
Hoover, H., Niessen, S., and Cravatt, B.F. (2010). Superfamily-wide portrait
of serine hydrolase inhibition achieved by library-versus-library screening.
Proc. Natl. Acad. Sci. USA 107, 2094120946.
Barbera, M.J., Schluter, A., Pedraza, N., Iglesias, R., Villarroya, F., and Giralt,
M. (2001). Peroxisome proliferator-activated receptor alpha activates transcription of the brown fat uncoupling protein-1 gene. A link between regulation
of the thermogenic and lipid oxidation pathways in the brown fat cell. J. Biol.
Chem. 276, 14861493.

Kopecky, J., Clarke, G., Enerback, S., Spiegelman, B., and Kozak, L.P. (1995).
Expression of the mitochondrial uncoupling protein gene from the aP2 gene
promoter prevents genetic obesity. J. Clin. Invest. 96, 29142923.
Koza, R.A., Hohmann, S.M., Guerra, C., Rossmeisl, M., and Kozak, L.P. (2000).
Synergistic gene interactions control the induction of the mitochondrial uncoupling protein (Ucp1) gene in white fat tissue. J. Biol. Chem. 275, 3448634492.
Lidell, M.E., Betz, M.J., Dahlqvist Leinhard, O., Heglind, M., Elander, L., Slawik,
M., Mussack, T., Nilsson, D., Romu, T., Nuutila, P., et al. (2013). Evidence for
two types of brown adipose tissue in humans. Nat. Med. 19, 631634.

Bartelt, A., Bruns, O.T., Reimer, R., Hohenberg, H., Ittrich, H., Peldschus, K.,
Kaul, M.G., Tromsdorf, U.I., Weller, H., Waurisch, C., et al. (2011). Brown adipose tissue activity controls triglyceride clearance. Nat. Med. 17, 200205.

Long, J.Z., Svensson, K.J., Tsai, L., Zeng, X., Roh, H.C., Kong, X., Rao, R.R.,
Lou, J., Lokurkar, I., Baur, W., et al. (2014). A smooth muscle-like origin for
beige adipocytes. Cell Metab. 19, 810820.

Boeuf, S., Keijer, J., Franssen-Van Hal, N.L., and Klaus, S. (2002). Individual
variation of adipose gene expression and identication of covariated genes
by cDNA microarrays. Physiol. Genomics 11, 3136.

Nedergaard, J., and Cannon, B. (2010). The changed metabolic world with human brown adipose tissue: therapeutic visions. Cell Metab. 11, 268272.

Cassard-Doulcier, A.M., Gelly, C., Fox, N., Schrementi, J., Raimbault, S.,
Klaus, S., Forest, C., Bouillaud, F., and Ricquier, D. (1993). Tissue-specic
and b-adrenergic regulation of the mitochondrial uncoupling protein gene:
control by cis-acting elements in the 50 -anking region. Mol. Endocrinol. 7,
497506.

Ohno, H., Shinoda, K., Spiegelman, B.M., and Kajimura, S. (2012). PPARg agonists induce a white-to-brown fat conversion through stabilization of
PRDM16 protein. Cell Metab. 15, 395404.
Ohno, H., Shinoda, K., Ohyama, K., Sharp, L.Z., and Kajimura, S. (2013).
EHMT1 controls brown adipose cell fate and thermogenesis through the
PRDM16 complex. Nature 504, 163167.

Cypess, A.M., Lehman, S., Williams, G., Tal, I., Rodman, D., Goldne, A.B.,
Kuo, F.C., Palmer, E.L., Tseng, Y.H., Doria, A., et al. (2009). Identication
and importance of brown adipose tissue in adult humans. N. Engl. J. Med.
360, 15091517.

Ortega-Molina, A., Efeyan, A., Lopez-Guadamillas, E., Munoz-Martin, M., Gomez-Lopez, G., Canamero, M., Mulero, F., Pastor, J., Martinez, S., Romanos,
E., et al. (2012). Pten positively regulates brown adipose function, energy
expenditure, and longevity. Cell Metab. 15, 382394.

Cypess, A.M., White, A.P., Vernochet, C., Schulz, T.J., Xue, R., Sass, C.A.,
Huang, T.L., Roberts-Toler, C., Weiner, L.S., Sze, C., et al. (2013). Anatomical
localization, gene expression proling and functional characterization of adult
human neck brown fat. Nat. Med. 19, 635639.

Rogers, N.H., Landa, A., Park, S., and Smith, R.G. (2012). Aging leads to a programmed loss of brown adipocytes in murine subcutaneous white adipose tissue. Aging Cell 11, 10741083.

Dominguez, E., Galmozzi, A., Chang, J.W., Hsu, K.L., Pawlak, J., Li, W., Godio,
C., Thomas, J., Partida, D., Niessen, S., et al. (2014). Integrated phenotypic
and activity-based proling links Ces3 to obesity and diabetes. Nat. Chem.
Biol. 10, 113121.
Enerback, S., Jacobsson, A., Simpson, E.M., Guerra, C., Yamashita, H.,
Harper, M.E., and Kozak, L.P. (1997). Mice lacking mitochondrial uncoupling
protein are cold-sensitive but not obese. Nature 387, 9094.
Fedorenko, A., Lishko, P.V., and Kirichok, Y. (2012). Mechanism of fatty-aciddependent UCP1 uncoupling in brown fat mitochondria. Cell 151, 400413.
Feldmann, H.M., Golozoubova, V., Cannon, B., and Nedergaard, J. (2009).
UCP1 ablation induces obesity and abolishes diet-induced thermogenesis in
mice exempt from thermal stress by living at thermoneutrality. Cell Metab. 9,
203209.
Gerhart-Hines, Z., Feng, D., Emmett, M.J., Everett, L.J., Loro, E., Briggs, E.R.,
Bugge, A., Hou, C., Ferrara, C., Seale, P., et al. (2013). The nuclear receptor
Rev-erba controls circadian thermogenic plasticity. Nature 503, 410413.
Golozoubova, V., Hohtola, E., Matthias, A., Jacobsson, A., Cannon, B., and
Nedergaard, J. (2001). Only UCP1 can mediate adaptive nonshivering thermogenesis in the cold. FASEB J. 15, 20482050.
Grundlingh, J., Dargan, P.I., El-Zanfaly, M., and Wood, D.M. (2011). 2,4-dinitrophenol (DNP): a weight loss agent with signicant acute toxicity and risk
of death. J. Med. Toxicol. 7, 205212.

Saito, M., Okamatsu-Ogura, Y., Matsushita, M., Watanabe, K., Yoneshiro, T.,
Nio-Kobayashi, J., Iwanaga, T., Miyagawa, M., Kameya, T., Nakada, K., et al.
(2009). High incidence of metabolically active brown adipose tissue in healthy
adult humans: effects of cold exposure and adiposity. Diabetes 58, 1526
1531.
Sanchez-Gurmaches, J., and Guertin, D.A. (2014). Adipocyte lineages: tracing
back the origins of fat. Biochim. Biophys. Acta 1842, 340351.
Sears, I.B., MacGinnitie, M.A., Kovacs, L.G., and Graves, R.A. (1996). Differentiation-dependent expression of the brown adipocyte uncoupling protein
gene: regulation by peroxisome proliferator-activated receptor gamma. Mol.
Cell. Biol. 16, 34103419.
Sharp, L.Z., Shinoda, K., Ohno, H., Scheel, D.W., Tomoda, E., Ruiz, L., Hu, H.,
Wang, L., Pavlova, Z., Gilsanz, V., and Kajimura, S. (2012). Human BAT possesses molecular signatures that resemble beige/brite cells. PLoS ONE 7,
e49452.
van Marken Lichtenbelt, W.D., Vanhommerig, J.W., Smulders, N.M., Drossaerts, J.M., Kemerink, G.J., Bouvy, N.D., Schrauwen, P., and Teule, G.J.
(2009). Cold-activated brown adipose tissue in healthy men. N. Engl. J. Med.
360, 15001508.
Virtanen, K.A., Lidell, M.E., Orava, J., Heglind, M., Westergren, R., Niemi, T.,
Taittonen, M., Laine, J., Savisto, N.J., Enerback, S., and Nuutila, P. (2009).
Functional brown adipose tissue in healthy adults. N. Engl. J. Med. 360,
15181525.

Guerra, C., Koza, R.A., Yamashita, H., Walsh, K., and Kozak, L.P. (1998).
Emergence of brown adipocytes in white fat in mice is under genetic control.
Effects on body weight and adiposity. J. Clin. Invest. 102, 412420.

Waki, H., Park, K.W., Mitro, N., Pei, L., Damoiseaux, R., Wilpitz, D.C., Reue, K.,
Saez, E., and Tontonoz, P. (2007). The small molecule harmine is an antidiabetic cell-type-specic regulator of PPARgamma expression. Cell Metab. 5,
357370.

Kajimura, S., and Saito, M. (2014). A new era in brown adipose tissue biology:
molecular control of brown fat development and energy homeostasis. Annu.
Rev. Physiol. 76, 225249.

Warming, S., Costantino, N., Court, D.L., Jenkins, N.A., and Copeland, N.G.
(2005). Simple and highly efcient BAC recombineering using galK selection.
Nucleic Acids Res. 33, e36.

Kajimura, S., Seale, P., Kubota, K., Lunsford, E., Frangioni, J.V., Gygi, S.P.,
and Spiegelman, B.M. (2009). Initiation of myoblast to brown fat switch by a
PRDM16-C/EBP-beta transcriptional complex. Nature 460, 11541158.

Wu, J., Bostrom, P., Sparks, L.M., Ye, L., Choi, J.H., Giang, A.H., Khandekar,
M., Virtanen, K.A., Nuutila, P., Schaart, G., et al. (2012). Beige adipocytes are a
distinct type of thermogenic fat cell in mouse and human. Cell 150, 366376.

Kim, G.W., Lin, J.E., Blomain, E.S., and Waldman, S.A. (2014). Antiobesity
pharmacotherapy: new drugs and emerging targets. Clin. Pharmacol. Ther.
95, 5366.

Xue, B., Rim, J.S., Hogan, J.C., Coulter, A.A., Koza, R.A., and Kozak, L.P.
(2007). Genetic variability affects the development of brown adipocytes in
white fat but not in interscapular brown fat. J. Lipid Res. 48, 4151.

Cell Reports 9, 15841593, December 11, 2014 2014 The Authors 1593

Cell Reports

Report
TRIM28 Represses Transcription of Endogenous
Retroviruses in Neural Progenitor Cells
Liana Fasching,1 Adamandia Kapopoulou,2 Rohit Sachdeva,1 Rebecca Petri,1 Marie E. Jonsson,1 Christian Manne,1
Priscilla Turelli,2 Patric Jern,3 Florence Cammas,4 Didier Trono,2 and Johan Jakobsson1,*
1Laboratory

of Molecular Neurogenetics, Department of Experimental Medical Science, Wallenberg Neuroscience Center and Lund
Stem Cell Center, Lund University, 221 84 Lund, Sweden
2School of Life Sciences, Ecole Polytechnique Fe
derale de Lausanne (EPFL), 1015 Lausanne, Switzerland
3Science for Life Laboratory, Department of Medical Biochemistry and Microbiology, Uppsala University, 751 23 Uppsala, Sweden
4IRCM, Institut de Recherche en Cance
rologie de Montpellier, INSERM, U896, Universite Montpellier; Institut Regional du Cancer Montpellier,
Montpellier 34298, France
*Correspondence: [email protected]
http://dx.doi.org/10.1016/j.celrep.2014.12.004
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/).

SUMMARY

TRIM28 is a corepressor that mediates transcriptional silencing by establishing local heterochromatin. Here, we show that deletion of TRIM28 in neural
progenitor cells (NPCs) results in high-level expression of two groups of endogenous retroviruses
(ERVs): IAP1 and MMERVK10C. We nd that NPCs
use TRIM28-mediated histone modications to
dynamically regulate transcription and silencing of
ERVs, which is in contrast to other somatic cell types
using DNA methylation. We also show that derepression of ERVs inuences transcriptional dynamics in
NPCs through the activation of nearby genes and
the expression of long noncoding RNAs. These ndings demonstrate a unique dynamic transcriptional
regulation of ERVs in NPCs. Our results warrant
future studies on the role of ERVs in the healthy and
diseased brain.
INTRODUCTION
The mammalian brain is an extremely complex organ harboring
more than a thousand different types of neurons that serve a
wide variety of functions. How this complexity is achieved
remains largely unknown. However, epigenetic mechanisms
such as DNA methylation, histone modication, and noncoding
RNAs are thought to be important in establishing a high diversity
of gene expression from the same template, leading to a spatial
pattern of transcription. How distinct transcriptional programs
are established in different neuronal populations remains poorly
understood, but one interesting recently proposed hypothesis
suggests transposable elements (TEs) to be involved in this process (Muotri et al., 2007; Reilly et al., 2013). TEs are repetitive
mobile genetic elements that were originally considered to be
parasitic DNA without any function, popularly termed junk
DNA. Today, it is becoming increasingly clear that TEs can
act as gene regulatory elements by serving as hubs for chromatin
20 Cell Reports 10, 2028, January 6, 2015 2015 The Authors

modications and by acting as transcriptional start sites for

noncoding RNAs. Consequently, TEs are very well suited to inuence gene expression and may play an important role in controlling and ne-tuning gene networks in the brain (Jern and Cofn,
2008; Cowley and Oakey, 2013).
Retroviruses are found in most vertebrates and can transform
their genetic material and integrate into the host genome as proviruses to produce new viruses. Occasionally, retroviruses infect
germline cells allowing the integrated proviruses to be passed on
to the offspring as an endogenous retrovirus (ERV). Around
8%10% of the human and mouse genome are composed of
this type of TE, and, despite up to millions of years since their
integration in host germline, many ERVs contain sequences
that can serve as transcriptional start sites or as cis-acting regulatory elements in the host genomes (Jern and Cofn, 2008). The
large amount of ERVs in mammalian genomes suggest that they
play important roles in the host organisms, for instance, by inuencing gene regulatory networks (Kunarso et al., 2010; Feschotte, 2008), but ERVs have also been linked to diseases. In
humans, aberrant expression of ERVs has been found in both
cancer and brain disorders, although causality remains to be
established (Jern and Cofn, 2008; Douville et al., 2011). Thus,
ERVs may contribute both benecial and detrimental effects,
which have been balanced throughout evolution, to the host
organism.
ERVs are silenced during the rst few days of embryogenesis
by TRIM28 (tripartite motif-containing protein 28, also known as
KAP1 or TIF1beta), a transcriptional corepressor essential for
early mouse development (Cammas et al., 2000; Rowe et al.,
2010). During the extensive genome reprogramming that takes
place at this period, TRIM28 is recruited to ERVs via
sequence-specic Kruppel-associated box zinc-nger proteins
(KRAB-ZFPs), a family of transcription factors that has undergone a rapid expansion in mammalian genomes in parallel with
the expansion of ERVs (Wolf and Goff, 2009; Thomas and
Schneider, 2011). TRIM28 then induces repressive histone modications by recruiting multiprotein complexes including the
H3K9 methyltransferase SETDB1 (also known as ESET), the
histone deacetylase-containing NuRD complex, and heterochromatin protein 1 (HP1) (Schultz et al., 2002; Sripathy et al.,

2006). Deletion of Trim28 or Setdb1 in ESCs leads to loss of the

H3K9me3-mark at ERVs, resulting in transcriptional activation of
these elements (Matsui et al., 2010; Rowe et al., 2010).
However, KRAB-ZFP/TRIM28 histone-based repression of
ERVs rapidly gives place to a more permanent silencing mechanism, as the TRIM28-mediated recruitment of de novo DNA
methyltransferases leads to cytosine methylation at CpG dinucleotides (Ellis et al., 2007; Wiznerowicz et al., 2007; Rowe and
Trono, 2011). The maintenance DNA methyltransferase complex
then ensures that DNA methylation is maintained, alleviating the
need for sequence-specic KRAB-ZFPs and TRIM28. In mouse
embryonic broblasts as well as in all adult tissues examined so
far, TRIM28 depletion has no impact on ERV silencing, which is
instead released by drugs such as 5-azacytidine or by deletion of
DNA methyltransferases (Jackson-Grusby et al., 2001; Hutnick
et al., 2010).
DNA methylation has long been considered as a stable epigenetic mark resulting in maintenance of DNA-methylation patterns
throughout the lifespan of an organism. However, several recent
studies demonstrate a unique dynamic regulation of DNAmethylation patterns in the brain (Sweatt, 2013). There is also
evidence that retroelements and transposons are highly active
during brain development and in neural progenitor cells (NPCs)
(Muotri et al., 2005, 2010; Baillie et al., 2011; Evrony et al.,
2012; Li et al., 2013; Perrat et al., 2013). For example, LINE-1
elements have been found to be transcriptionally active and to
retrotranspose in NPCs (Muotri et al., 2005, 2010; Coufal et al.,
2009). In addition, we have previously found that deletion of
TRIM28 in postmitotic forebrain neurons results in complex
behavioral alterations, including vulnerability to stress (Jakobsson et al., 2008). In the present work, we demonstrate that
NPCs use TRIM28-mediated histone modications to dynamically regulate the transcription and silencing of ERVs, rather
than the DNA methylation at play in other somatic tissues. We
also unveil that derepression of ERVs inuences transcriptional
dynamics in NPCs, by activating nearby genes and the expression of long noncoding RNAs (lncRNAs).
RESULTS
TRIM28-Decient NPCs Express High Levels of ERVs
To investigate if TRIM28 contributes to ERV silencing in NPCs we
established Trim28-decient NPC cultures. We crossed transgenic NestinCre mice (Tronche et al., 1999) with mice carrying
oxed Trim28-alleles (Trim28/) (Weber et al., 2002), resulting
in excision of Trim28 in neural progenitors at the time when
Nestin-expression is initiated, starting around embryonic day
10 (E10). At E13.5, we collected embryos, dissected the forebrain, and established NPC cultures from individual embryos
(Figures 1A and 1B). We conrmed the deletion of Trim28 by
genotyping for the excised allele and by verifying the absence
of TRIM28 protein (Figures 1C and 1D). We collected RNA
from Trim28/ NPCs and wild-type controls and performed
RNA extraction followed by deep sequencing (RNA-seq). The resulting reads were mapped against reference sequences from
Repbase, a database containing consensus sequences for
known repetitive elements (Jurka et al., 2005). We found that
several ERVs were highly upregulated in Trim28/ NPCs,

including, e.g., Mus musculus ERV using tRNALys type 10C

(MMERVK10C) and intracisternal A-particles class 1 (IAP1) (Figure 1E; Tables S1 and S2). Other retroelements such as MusD
and LINE-1 were modestly upregulated, whereas several other
types of common repetitive elements were unaffected (Figure 1E;
Tables S1 and S2).
We conrmed increased transcription of MMERVK10C and
IAP1 elements using quantitative RT-PCR (qRT-PCR) (Figure 1F).
In contrast, when we used primer pairs designed to recognize
the consensus sequence of the entire IAP-family, including
more ancient IAP elements, we detected only a modest upregulation (Figure 1F). This nding is in line with the results of the
RNA-seq, which indicated that only certain types of IAP elements were upregulated in Trim28/ NPCs. Also in agreement
with the RNA-seq, qRT-PCR analyses indicated that deletion
of Trim28 in NPCs only modestly increased the expression of
other retroelements such as LINE-1 or MusD (Figure 1F). We
conrmed these results in cultures derived from two separate
embryos (data not shown).
Trim28/ NPCs proliferated at a similar rate compared to
cells generated from wild-type and heterozygous siblings and
could be expanded for more than 60 passages. However, we
observed that Trim28/ NPCs were growing in dense clusterlike formations, which seemed to attach less to the ask surface
compared to the wild-type control. Trim28/ NPCs could also
be differentiated to both neurons and astrocytes suggesting
that TRIM28 has no major inuence on the self-renewal and
differentiation of NPCs (Figures 1G and 1H).
MMERVK10C Elements Are Controlled by TRIM28
The RNA-seq analysis indicated that MMERVK10C elements
were among the most upregulated ERVs following Trim28deletion in NPCs. MMERVK10C is a beta-like ERV similar to
HERVK (HML2), one of the most recent ERVs to invade the
human genome (Reichmann et al., 2012) (Belshaw et al., 2005).
MMERVK10C sequences anked by RLTR10C make up putative
proviral sequences of around 8.4 kb. In the mouse genome,
MMERVK10C is present in a few complete provirus loci (20)
and more than 1,000 incomplete loci (Reichmann et al., 2012).
We performed sequence analysis of the MMERVK10C provirus
for the presence of retroviral features using the RetroTector
software (Sperber et al., 2007). Based on this analysis, we
designed primers recognizing the LTRs, gag, pol, and env of
the MMERVK10C provirus and investigated expression levels
in Trim28/ NPCs (schematics in Figure 2A). We found that
transcripts over the entire region of the provirus were increased,
including a massive expression of env sequences when compared to wild-type controls (170-fold; Figure 2B).
Ascertaining that the ERV induction observed in NPCs isolated
from Trim28/ animals was not secondary to more general
developmental anomalies, knocking down TRIM28 in wild-type
NPCs by lentivector-mediated RNA interference led to a marked
upregulation of these retroelements (Figure 2C). Furthermore,
increased ERV expression was detected in forebrain tissue
from Trim28/ embryos (Figure 2D).
In ESCs, TRIM28 controls ERV expression via histone modications including H3K9 trimethylation (Rowe et al., 2010),
whereas it is DNA methylation that instead prevails in somatic
Cell Reports 10, 2028, January 6, 2015 2015 The Authors 21

Figure 1. Establishment of Trim28-Decient Neural Progenitor Cultures

(A) Illustration of the experimental approach.
(B) Representative images of early passage Trim28/ NPCs.
(C) PCR analysis of genomic DNA from wild-type and Trim28/ NPCs demonstrates the presence of the 152 and 290 bp products corresponding to loxP-anked
or excised Trim28, respectively.
(D) Verication of a complete lack of TRIM28 protein via immunocytochemistry.
(E) RNA-seq analysis. The graph shows KO samples plotted versus wild-type samples, where each dot represents a Repbase sequence.
(F) qRT-PCR of RNA isolated from wild-type and Trim28/ NPCs.
(G) Trim28-decient NPCs display a homogenous expression of NESTIN.
(H) Immunouorescent analysis of differentiated NPCs.
Data are presented as mean of relative values SEM. **p < 0.01, ***p < 0.001, Students t test. Scale bars, 200 (A) and 50 (B) mm. See also Tables S1 and S2.

tissues. In NPCs, we found that the MMERVK10C provirus was

enriched in H3K9me3, and that this repressive mark was markedly reduced in Trim28/ NPCs (Figure 2E).
Because MMERVK10C appeared to be under TRIM28 control
in NPCs, we hypothesized that at least a proportion of these retroelements escaped DNA methylation in these cells. To probe
this issue, we examined the DNA methylation status of full-length
MMERVK10C, which were among the most highly upregulated
retroelements in Trim28/ NPCs. Bisulte sequencing of a
CpG-island located in the 30 region of MMERVK10C revealed
several clones with some unmethylated CpGs (17% unmethylated CpGs, Figure 2F) in NPCs, whereas this region was almost
22 Cell Reports 10, 2028, January 6, 2015 2015 The Authors

fully methylated in DNA extracted from mouse tail (7% unmethylated CpGs, Figure 2F, Fishers exact test one-sided p < 0.05).
Moreover, we found no difference in the level of CpG methylation
between wild-type and Trim28/ NPCs. In summary, these data
suggest that a proportion of the MMERVK10C elements are
spared from undergoing DNA methylation specically in NPCs
during early development.
Increased Expression of IAP1 Results in ERV-Derived
Protein Expression
IAP1 elements, which lose H3K9me3 marks and were also
highly upregulated in Trim28/ NPCs (Figures 3A and 3B), are

Figure
2. Analysis
of
MMERVK10C Provirus

the

Putative

(A) Schematic drawing of the MMERVK10C

provirus and approximate primer positions.
(B) Quantitative analysis of transcript levels of
different regions of the MMERVK10C provirus in
Trim28/ and wild-type NPCs.
(C) qRT-PCR analysis of MMERVK10C following
TRIM28-shRNA knockdown.
(D) qRT-PCR analysis of E13.5 forebrain dissected
from intercrosses of NestinCre Trim28oxed mice.
(E) ChIP for H3K9me3 in Trim28/ and
wild-type NPCs.
(F) Bisulte sequencing analysis of the 30 end
region of MMERVK10C. Empty and full circles
represent unmethylated and methylated CpGs,
respectively.
Data are presented as mean of relative values
SEM. *p < 0.05, Students t test.

internalized env-lacking mouse ERVs that demonstrate a large

degree of polymorphism among different mouse strains and
maintain the capacity to retrotranspose. Using immunocytochemistry with an IAP-specic antibody, we found a uniform,
high-level IAP-gag expression located to the cytoplasm in
Trim28/ NPCs (Figure 3C).
Taken together, these data demonstrate that deletion of
TRIM28 in NPCs results in a massive transcriptional increase
of ERVs, including MMERVK10C and IAP1. NPCs thus appear
to constitute a cellular environment distinct from that of other
somatic cells studied so far, with the TRIM28-induced histonebased repressive mechanism playing a role in ERV control.
Activation of ERVs Correlates with Increased
Transcription of Nearby Genes
The ability of ERVs to attract transcription factors and silencing
complexes has led to a reassessment of their role in the host
genome. ERVs are now considered to be important transcriptional regulatory elements that shape and inuence gene expression during early development (Isbel and Whitelaw, 2012).
For example, we have recently found that TRIM28 controls
the expression of developmental genes by repressing ERVassociated enhancers in pluripotent cells (Rowe et al., 2013).
Twenty-six MMERVK10C proviruses and 361 IAP proviruses
that were upregulated in Trim28/ NPCs were mapped to

precise genomic locations (Figure S1).

Out of these 387 proviruses, 90 were
situated close to genes (<50 kb). We
found that 25 of those genes (28%)
demonstrated signicantly increased expression, whereas expression of only six
of them was decreased (7%) (Figure 4A).
We also found that those 90 genes
located close to upregulated ERVs
(ERV-up genes) were on average 3-fold
upregulated in Trim28/ cells (Figure 4B).
In contrast, a random selection of ERVs
that was not upregulated in Trim28/
cells (n = 129, MMERVK10C and IAP1 elements) did not affect
nearby genes (ERVs-ctrl genes, n = 50, Figure 4B). Interestingly,
we also found that ERV-up genes were expressed at low
levels in wild-type cells (Figure 4C), which is in agreement
with a model where ERVs mediate repressive regulation of
nearby genes caused by the attraction of the TRIM28 silencing
complex to ERV sequences. We validated the increased expression of ve ERV-up genes in Trim28/ cells using qRT-PCR
(Figure 4D).
ERVs Produce Long Noncoding RNAs
We looked in detail at BC048671, which is a protein-coding transcript that is induced in Trim28/ NPCs but completely absent
in wild-type NPCs. BC048671 is located 5 kb downstream of an
IAP element, which is also highly upregulated in Trim28/
NPCs. The RNA-seq data show that transcriptional initiation at
the IAP element results in the formation of a long transcript
(>10 kb) that extends into the coding sequence of BC048671
(Figure 4E). The presence of high levels of this long transcript
was veried using qRT-PCR primers located both upstream
and within the coding sequence of BC048671 (Figure 4F).
Thus, readthrough of an ERV-derived transcript into another
locus is likely to be one of several mechanisms by which nearby
gene expression can be affected (see also Figure S2). This
nding supports the notion that a general feature of ERVs might
Cell Reports 10, 2028, January 6, 2015 2015 The Authors 23

Figure 3. Analysis of IAP1 Expression

(A) Quantitative analysis of transcript levels of different regions of IAP1 provirus
in Trim28/ and wild-type NPCs.
(B) ChIP for H3K9me3 in Trim28/ and wild-type NPCs.
(C) Confocal analysis of immunouorescence staining for IAP-gag on
Trim28/ and wild-type NPCs. Scale bar, 10 mm.
Data are presented as mean of relative values SEM. *p < 0.05, Students
t test.

be to act as transcriptional start sites for long noncoding RNAs

(lncRNAs). Indeed, when we scrutinized ERV elements located
in gene free regions, we found that both IAP and MMERVK10C
elements serve as start sites for lncRNAs (Figures 4G and 4I).
Using qRT-PCR, we conrmed high-level expression of two
ERV-derived lncRNAs in Trim28/ NPCs (Figures 4H and 4J).
The length of the ERV-derived lncRNAs did in many cases
extend 25 kb (Figure 4K). These data demonstrate that derepression of ERVs in NPCs can result in the expression of multiple
lncRNAs. The functional role of lncRNAs in NPCs remains largely
unexplored, but they are thought to play important regulatory
roles and have been implicated as scaffolds for nuclear protein
complexes and as antisense transcripts in the control of epigenetic pathways (Guttman and Rinn, 2012).
DISCUSSION
In pluripotent stem cells, TRIM28 is a master corepressor of
retroelements including ERVs (Matsui et al., 2010; Rowe
et al., 2010). When these cells differentiate into various somatic
cell types, DNA methylation is instated on ERV sequences,
which ultimately results in stable silencing that is no longer
dependent on TRIM28 (Wiznerowicz et al., 2007; Rowe et al.,
2013). Thus, when TRIM28 is deleted from various somatic
cell types such as broblasts, hepatocytes, and white blood
cells, no increased ERV expression is detected (Rowe et al.,
24 Cell Reports 10, 2028, January 6, 2015 2015 The Authors

2010; Bojkowska et al., 2012; Santoni de Sio et al., 2012a,

2012b). Here, we describe an exception to this rule. When
TRIM28 is deleted in NPCs, several ERVs become highly expressed. This nding unravels a unique transcriptional regulation of ERVs in NPCs.
ERVs regulated by TRIM28 in NPCs are recent invaders of the
mouse genome. IAP1 is the most recent member of the wellstudied IAP ERVs (Qin et al., 2010). IAPs are ERVs that have
lost the env gene and adopted an intracellular life cycle (Ribet
et al., 2008). IAP1 has been shown to retrotranspose and has
distinct integration patterns in different strains of laboratory
mice (Li et al., 2012). MMERVK10C, another ERV massively
upregulated in Trim28/ NPCs, is poorly characterized, and it
is unclear if it is still endowed with retrotransposition potential,
whether on its own or with the support of factors provided
in trans. A previous study that analyzed the structure of
MMERVK10C elements in the mouse genome found that the majority of these elements have 30 deletions removing the start of
the gag open reading frame as well as the major part of env
(Reichmann et al., 2012). Our data demonstrate that, in NPCs,
TRIM28 controls the rare copies of env-containing MMERVK10C
elements, which are most likely to be the youngest ones, raising
the possibility that these recent invaders of the mouse genome
contain cis-acting genomic elements that allow them to escape
DNA methylation in NPCs.
The classic view of repetitive mobile genetic elements as parasitic DNA without benecial function to the host is challenged in
many ways. There are a number of recent studies indicating that
transposable elements (TEs) play important roles in establishing
and rewiring gene networks (Kunarso et al., 2010; Chuong et al.,
2013). TEs have been shown to act as enhancers, repressors,
and alternative promoters. In addition, TEs can affect splicing
patterns and produce peptides with important functional roles
(Jern and Cofn, 2008). In this study, we demonstrate that
activated ERVs can inuence gene expression of nearby genes,
such as BC048671, and serve as start sites for lncRNAs. Taken
together, our ndings indicate that ERVs participate in the
control of gene networks in the brain.
We have previously demonstrated that deletion of Trim28 in
postmitotic forebrain neurons results in complex behavioral
changes (Jakobsson et al., 2008). In addition, heterozygous
germline deletion of Trim28 has been described to result in
abnormal behavioral phenotypes (Whitelaw et al., 2010). In this
study, we found that deletion of Trim28 during brain development is lethal (Figure S3). In addition, we also noted that heterozygous deletion of Trim28 during brain development resulted in
behavioral changes characterized by hyperactivity (Figure S3).
Together, these ndings demonstrate that disruption of
TRIM28 levels in the mouse brain results in behavioral changes
that are similar to impairments found in humans with certain
psychiatric disorders. With this in mind, it is noteworthy that
increased levels of ERV transcripts have been detected in
patients with several neurological and psychiatric disorders
(Jeong et al., 2010; Douville et al., 2011; Li et al., 2012; Karlsson
et al., 2001). The signicance of these ndings has been questioned because the human genome does not appear to harbor
ERVs with known retrotransposing capacity (Jern and Cofn,
2008). However, the increasing evidence that derepression of

Figure 4. Activation of ERVs Inuences

Expression of Nearby Genes and Results in
the Expression of lncRNAs
(A) Transcriptional change of genes located close
(<50 kb) to ERVs in Trim28/ NPCs.
(B) Mean transcriptional change of genes located to
ERVs with increased transcription (ERV-up genes)
and genes located close to unchanged ERVs
(ERV-ctrl genes) in Trim28/ NPCs.
(C) Absolute expression level of ERV-up genes and
ERV-ctrl genes in wild-type NPCs.
(D) qRT-PCR of RNA isolated from wild-type and
Trim28/ NPCs.
(E) Screen shot from the USCS genome browser
(mm9) showing induced transcription of BC048671
in Trim28/ NPCs.
(G, I, and K) Activation of ERVs results in the
expression of lncRNAs. Screen shot from the USCS
genome browser (mm9).
(F, H, and J) qRT-PCR of RNA isolated from
wild-type and Trim28/ NPCs. Primers are indicated as green bars and include primers over the
ERV junction as well as close and more distant from
the 30 end of the ERVs.
Data are presented as mean of relative values
SEM. *p < 0.05, Students t test. See also Figures
S1 and S2.

Cell Reports 10, 2028, January 6, 2015 2015 The Authors 25

ERVs inuence gene networks, including the ndings presented

here, provides a potential mechanistic explanation for these
observations.
In summary, our data suggest that ERVs may be involved in
the regulation of gene expression in NPCs and may hereby offer
a link between ERVs and brain disorders. It seems unlikely that
behavioral phenotypes would arise from the derepression of a
single ERV-induced gene. Instead, the presence of ERVs in multiple copies scattered throughout the genome allows for a
powerful network-like control of gene expression, where dysregulation could result in widespread consequences. However,
due to the large numbers of ERVs present in the mouse and
human genome and their sequence variation, it is currently
unfeasible to demonstrate a causal role for ERVs in controlling
complex behavior or brain disorders using loss-of-function
approaches, such as gene targeting and small hairpin RNA
(shRNA) knockdown. Instead, improving our knowledge of critical host factors and networks controlling ERVs is essential to
appreciate their impact on the genome and pathologies that
may stem from their dysregulation. The demonstration that there
is an ongoing dynamic TRIM28-mediated regulation of ERVs in
NPCs is a step in this direction and warrants future studies of
epigenetic and posttranscriptional regulation of ERVs in the
healthy and diseased brain.
EXPERIMENTAL PROCEDURES
Detailed experimental procedures can be found in the Supplemental
Experimental Procedures.
Procedures
Transgenic Animals
All animal-related procedures were approved by and conducted in accordance with the committee for use of laboratory animals at Lund University.
NestinCre and oxed Trim28 mice have been described previously (Weber
et al., 2002; Tronche et al., 1999).
Cell Culture
NPC was established from embryonic day 13.5 (E13.5) forebrain and cultured
as previously described (Conti et al., 2005).
Immunouorescence
Immunouorescence was performed as previously described (Thompson
et al., 2005; Sachdeva et al., 2010).
RNA Studies
RNA-seq and qRT-PCR was performed as previously described (Rowe et al.,
2010). The 50-base-paired end reads were mapped onto the RepBase version
16.08 (Jurka et al., 2005) and to the mouse genome (mm9) assembly. Mapping
was done using the bowtie short read aligner (Langmead et al., 2009).
Chromatin Immunoprecipitation
Chromatin immunoprecipitation was performed with iDeal chromatin immunoprecipitation sequencing (ChIP-seq) kit (Diagenode) according to suppliers
recommendations.
DNA-Methylation Analysis
Bisulte sequencing was performed with the EpiTect bisulte kit (QIAGEN)
according to the suppliers recommendations. Sequence data were analyzed
with the QUantication tool for Methylation Analysis (Kumaki et al., 2008).
Statistical Analysis
An unpaired t test was performed in order to test for statistical signicance.
Data are presented as mean SEM.

SUPPLEMENTAL INFORMATION
Supplemental Information includes Supplemental Experimental Procedures,
three gures, and three tables and can be found with this article online at
http://dx.doi.org/10.1016/j.celrep.2014.12.004.
AUTHOR CONTRIBUTIONS
L.F., A.K., R.S., R.P., M.E.J., and C.M. designed and performed research and
analyzed data. P.J., P.T., and D.T. designed research and analyzed data.
F.C. contributed reagents. J.J. designed and coordinated the project and
analyzed data. L.F. and J.J. wrote the paper, and all authors reviewed the
manuscript.
ACKNOWLEDGMENTS
We are grateful to A. Bjorklund, S. Quenneville, and all members of the J.J. and
Parmar laboratories for stimulating discussions. We thank U. Jarl, A. Josefsson, C. Isaksson, I. Nilsson, A.-K. Olden, E. Ling, S. Smiljanic, M. Sparrenius,
and E. Tjon for technical assistance. This study was supported by grants from
Swedish Research Council (J.J.), Formas (P.J.), the Swedish Cancer Foundation (J.J.), the Lundqvist, Jeansson, and Crafoord foundations (J.J.), the
Swedish Government Initiative for Strategic Research Areas MultiPark (J.J.),
the French government, CNRS and INSERM (F.C.), and the French Agence
Nationale pour la Recherche (F.C.).
Received: June 10, 2014
Revised: October 28, 2014
Accepted: December 1, 2014
Published: December 24, 2014
REFERENCES
Baillie, J.K., Barnett, M.W., Upton, K.R., Gerhardt, D.J., Richmond, T.A., De
Sapio, F., Brennan, P.M., Rizzu, P., Smith, S., Fell, M., et al. (2011). Somatic
retrotransposition alters the genetic landscape of the human brain. Nature
479, 534537.
Belshaw, R., Dawson, A.L., Woolven-Allen, J., Redding, J., Burt, A., and
Tristem, M. (2005). Genomewide screening reveals high levels of insertional
polymorphism in the human endogenous retrovirus family HERV-K(HML2):
implications for present-day activity. J. Virol. 79, 1250712514.
Bojkowska, K., Aloisio, F., Cassano, M., Kapopoulou, A., Santoni de Sio, F.,
Zangger, N., Offner, S., Cartoni, C., Thomas, C., Quenneville, S., et al.
(2012). Liver-specic ablation of Kruppel-associated box-associated protein
1 in mice leads to male-predominant hepatosteatosis and development of liver
adenoma. Hepatology 56, 12791290.
Cammas, F., Mark, M., Dolle, P., Dierich, A., Chambon, P., and Losson, R.
(2000). Mice lacking the transcriptional corepressor TIF1beta are defective in
early postimplantation development. Development 127, 29552963.
Chuong, E.B., Rumi, M.A., Soares, M.J., and Baker, J.C. (2013).
Endogenous retroviruses function as species-specic enhancer elements in
the placenta. Nat. Genet. 45, 325329.
Conti, L., Pollard, S.M., Gorba, T., Reitano, E., Toselli, M., Biella, G., Sun, Y.,
Sanzone, S., Ying, Q.L., Cattaneo, E., and Smith, A. (2005). Nicheindependent symmetrical self-renewal of a mammalian tissue stem cell.
PLoS Biol. 3, e283.
Coufal, N.G., Garcia-Perez, J.L., Peng, G.E., Yeo, G.W., Mu, Y., Lovci, M.T.,
Morell, M., OShea, K.S., Moran, J.V., and Gage, F.H. (2009). L1 retrotransposition in human neural progenitor cells. Nature 460, 11271131.

ACCESSION NUMBERS

Cowley, M., and Oakey, R.J. (2013). Transposable elements re-wire and
ne-tune the transcriptome. PLoS Genet. 9, e1003234.

The RNA-seq data were deposited in the NCBI Gene Expression Omnibus and
are available under accession number GSE45930.

Douville, R., Liu, J., Rothstein, J., and Nath, A. (2011). Identication of active
loci of a human endogenous retrovirus in neurons of patients with amyotrophic
lateral sclerosis. Ann. Neurol. 69, 141151.

26 Cell Reports 10, 2028, January 6, 2015 2015 The Authors

Ellis, J., Hotta, A., and Rastegar, M. (2007). Retrovirus silencing by an

epigenetic TRIM. Cell 131, 1314.
Evrony, G.D., Cai, X., Lee, E., Hills, L.B., Elhosary, P.C., Lehmann, H.S.,
Parker, J.J., Atabay, K.D., Gilmore, E.C., Poduri, A., et al. (2012). Single-neuron
sequencing analysis of L1 retrotransposition and somatic mutation in the
human brain. Cell 151, 483496.
Feschotte, C. (2008). Transposable elements and the evolution of regulatory
networks. Nat. Rev. Genet. 9, 397405.
Guttman, M., and Rinn, J.L. (2012). Modular regulatory principles of large
non-coding RNAs. Nature 482, 339346.
Hutnick, L.K., Huang, X., Loo, T.C., Ma, Z., and Fan, G. (2010). Repression of
retrotransposal elements in mouse embryonic stem cells is primarily mediated
by a DNA methylation-independent mechanism. J. Biol. Chem. 285, 21082
21091.
Isbel, L., and Whitelaw, E. (2012). Endogenous retroviruses in mammals:
an emerging picture of how ERVs modify expression of adjacent genes.
Bioessays 34, 734738.
Jackson-Grusby, L., Beard, C., Possemato, R., Tudor, M., Fambrough, D.,
Csankovszki, G., Dausman, J., Lee, P., Wilson, C., Lander, E., and Jaenisch,
R. (2001). Loss of genomic methylation causes p53-dependent apoptosis
and epigenetic deregulation. Nat. Genet. 27, 3139.
Jakobsson, J., Cordero, M.I., Bisaz, R., Groner, A.C., Busskamp, V., Bensadoun, J.C., Cammas, F., Losson, R., Mansuy, I.M., Sandi, C., and Trono, D.
(2008). KAP1-mediated epigenetic repression in the forebrain modulates
behavioral vulnerability to stress. Neuron 60, 818831.
Jeong, B.H., Lee, Y.J., Carp, R.I., and Kim, Y.S. (2010). The prevalence of
human endogenous retroviruses in cerebrospinal uids from patients with
sporadic Creutzfeldt-Jakob disease. J. Clin. Virol. 47, 136142.

Muotri, A.R., Marchetto, M.C., Coufal, N.G., Oefner, R., Yeo, G., Nakashima,
K., and Gage, F.H. (2010). L1 retrotransposition in neurons is modulated by
MeCP2. Nature 468, 443446.
Perrat, P.N., DasGupta, S., Wang, J., Theurkauf, W., Weng, Z., Rosbash, M.,
and Waddell, S. (2013). Transposition-driven genomic heterogeneity in the
Drosophila brain. Science 340, 9195.
Qin, C., Wang, Z., Shang, J., Bekkari, K., Liu, R., Pacchione, S., McNulty, K.A.,
Ng, A., Barnum, J.E., and Storer, R.D. (2010). Intracisternal A particle genes:
Distribution in the mouse genome, active subtypes, and potential roles as species-specic mediators of susceptibility to cancer. Mol. Carcinog. 49, 5467.
Reichmann, J., Crichton, J.H., Madej, M.J., Taggart, M., Gautier, P., GarciaPerez, J.L., Meehan, R.R., and Adams, I.R. (2012). Microarray analysis of
LTR retrotransposon silencing identies Hdac1 as a regulator of retrotransposon expression in mouse embryonic stem cells. PLoS Comput. Biol. 8,
e1002486.
Reilly, M.T., Faulkner, G.J., Dubnau, J., Ponomarev, I., and Gage, F.H. (2013).
The role of transposable elements in health and diseases of the central nervous
system. J. Neurosci. 33, 1757717586.
Ribet, D., Harper, F., Dupressoir, A., Dewannieux, M., Pierron, G., and
Heidmann, T. (2008). An infectious progenitor for the murine IAP retrotransposon: emergence of an intracellular genetic parasite from an ancient retrovirus.
Genome Res. 18, 597609.
Rowe, H.M., and Trono, D. (2011). Dynamic control of endogenous retroviruses during development. Virology 411, 273287.
Rowe, H.M., Jakobsson, J., Mesnard, D., Rougemont, J., Reynard, S., Aktas,
T., Maillard, P.V., Layard-Liesching, H., Verp, S., Marquis, J., et al. (2010).
KAP1 controls endogenous retroviruses in embryonic stem cells. Nature
463, 237240.

Jurka, J., Kapitonov, V.V., Pavlicek, A., Klonowski, P., Kohany, O., and
Walichiewicz, J. (2005). Repbase Update, a database of eukaryotic repetitive
elements. Cytogenet. Genome Res. 110, 462467.

Rowe, H.M., Kapopoulou, A., Corsinotti, A., Fasching, L., Macfarlan, T.S.,
Tarabay, Y., Viville, S., Jakobsson, J., Pfaff, S.L., and Trono, D. (2013).
TRIM28 repression of retrotransposon-based enhancers is necessary to preserve transcriptional dynamics in embryonic stem cells. Genome Res. 23,
452461.

Karlsson, H., Bachmann, S., Schroder, J., McArthur, J., Torrey, E.F., and
Yolken, R.H. (2001). Retroviral RNA identied in the cerebrospinal uids and
brains of individuals with schizophrenia. Proc. Natl. Acad. Sci. USA 98,
46344639.

Sachdeva, R., Jonsson, M.E., Nelander, J., Kirkeby, A., Guibentif, C., Gentner,
B., Naldini, L., Bjorklund, A., Parmar, M., and Jakobsson, J. (2010). Tracking
differentiating neural progenitors in pluripotent cultures using microRNAregulated lentiviral vectors. Proc. Natl. Acad. Sci. USA 107, 1160211607.

Kumaki, Y., Oda, M., and Okano, M. (2008). QUMA: quantication tool for
methylation analysis. Nucleic Acids Res. 36 (Web Server issue), W170-5.

Santoni de Sio, F.R., Barde, I., Offner, S., Kapopoulou, A., Corsinotti, A.,
Bojkowska, K., Genolet, R., Thomas, J.H., Luescher, I.F., Pinschewer, D.,
et al. (2012a). KAP1 regulates gene networks controlling T-cell development
and responsiveness. FASEB J. 26, 45614575.

Jern, P., and Cofn, J.M. (2008). Effects of retroviruses on host genome function. Annu. Rev. Genet. 42, 709732.

Kunarso, G., Chia, N.Y., Jeyakani, J., Hwang, C., Lu, X., Chan, Y.S., Ng, H.H.,
and Bourque, G. (2010). Transposable elements have rewired the core regulatory network of human embryonic stem cells. Nat. Genet. 42, 631634.
Langmead, B., Trapnell, C., Pop, M., and Salzberg, S.L. (2009). Ultrafast and
memory-efcient alignment of short DNA sequences to the human genome.
Genome Biol. 10, R25.
Li, J., Akagi, K., Hu, Y., Trivett, A.L., Hlynialuk, C.J., Swing, D.A., Volfovsky, N.,
Morgan, T.C., Golubeva, Y., Stephens, R.M., et al. (2012). Mouse endogenous
retroviruses can trigger premature transcriptional termination at a distance.
Genome Res. 22, 870884.
Li, W., Prazak, L., Chatterjee, N., Gruninger, S., Krug, L., Theodorou, D., and
Dubnau, J. (2013). Activation of transposable elements during aging and
neuronal decline in Drosophila. Nat. Neurosci. 16, 529531.

Santoni de Sio, F.R., Massacand, J., Barde, I., Offner, S., Corsinotti, A., Kapopoulou, A., Bojkowska, K., Dagklis, A., Fernandez, M., Ghia, P., et al. (2012b).
KAP1 regulates gene networks controlling mouse B-lymphoid cell differentiation and function. Blood 119, 46754685.
Schultz, D.C., Ayyanathan, K., Negorev, D., Maul, G.G., and Rauscher, F.J.,
3rd. (2002). SETDB1: a novel KAP-1-associated histone H3, lysine 9-specic
methyltransferase that contributes to HP1-mediated silencing of euchromatic
genes by KRAB zinc-nger proteins. Genes Dev. 16, 919932.
Sperber, G.O., Airola, T., Jern, P., and Blomberg, J. (2007). Automated recognition of retroviral sequences in genomic dataRetroTector. Nucleic Acids
Res. 35, 49644976.

Matsui, T., Leung, D., Miyashita, H., Maksakova, I.A., Miyachi, H., Kimura, H.,
Tachibana, M., Lorincz, M.C., and Shinkai, Y. (2010). Proviral silencing in
embryonic stem cells requires the histone methyltransferase ESET. Nature
464, 927931.

Sripathy, S.P., Stevens, J., and Schultz, D.C. (2006). The KAP1 corepressor
functions to coordinate the assembly of de novo HP1-demarcated microenvironments of heterochromatin required for KRAB zinc nger protein-mediated
transcriptional repression. Mol. Cell. Biol. 26, 86238638.

Muotri, A.R., Chu, V.T., Marchetto, M.C., Deng, W., Moran, J.V., and Gage,
F.H. (2005). Somatic mosaicism in neuronal precursor cells mediated by L1
retrotransposition. Nature 435, 903910.

Sweatt, J.D. (2013). The emerging eld of neuroepigenetics. Neuron 80,

624632.

Muotri, A.R., Marchetto, M.C., Coufal, N.G., and Gage, F.H. (2007). The necessary junk: new functions for transposable elements. Hum. Mol. Genet. 16 Spec
No. 2, R159R167.

Thomas, J.H., and Schneider, S. (2011). Coevolution of retroelements and

tandem zinc nger genes. Genome Res. 21, 18001812.
Thompson, L., Barraud, P., Andersson, E., Kirik, D., and Bjorklund, A. (2005).
Identication of dopaminergic neurons of nigral and ventral tegmental area

Cell Reports 10, 2028, January 6, 2015 2015 The Authors 27

subtypes in grafts of fetal ventral mesencephalon based on cell morphology,

protein expression, and efferent projections. J. Neurosci. 25, 64676477.
Tronche, F., Kellendonk, C., Kretz, O., Gass, P., Anlag, K., Orban, P.C., Bock,
R., Klein, R., and Schutz, G. (1999). Disruption of the glucocorticoid
receptor gene in the nervous system results in reduced anxiety. Nat. Genet.
23, 99103.
Weber, P., Cammas, F., Gerard, C., Metzger, D., Chambon, P., Losson, R., and
Mark, M. (2002). Germ cell expression of the transcriptional co-repressor
TIF1beta is required for the maintenance of spermatogenesis in the mouse.
Development 129, 23292337.

28 Cell Reports 10, 2028, January 6, 2015 2015 The Authors

Whitelaw, N.C., Chong, S., Morgan, D.K., Nestor, C., Bruxner, T.J., Ashe, A.,
Lambley, E., Meehan, R., and Whitelaw, E. (2010). Reduced levels of two
modiers of epigenetic gene silencing, Dnmt3a and Trim28, cause increased
phenotypic noise. Genome Biol. 11, R111.
Wiznerowicz, M., Jakobsson, J., Szulc, J., Liao, S., Quazzola, A., Beermann,
F., Aebischer, P., and Trono, D. (2007). The Kruppel-associated box repressor
domain can trigger de novo promoter methylation during mouse early embryogenesis. J. Biol. Chem. 282, 3453534541.
Wolf, D., and Goff, S.P. (2009). Embryonic stem cells use ZFP809 to silence
retroviral DNAs. Nature 458, 12011204.

Cell Reports

Article
Comparative Hi-C Reveals that CTCF Underlies
Evolution of Chromosomal Domain Architecture
Matteo Vietri Rudan,1 Christopher Barrington,1 Stephen Henderson,1 Christina Ernst,2 Duncan T. Odom,2 Amos Tanay,3
and Suzana Hadjur1,*
1Research

Department of Cancer Biology, Cancer Institute, University College London, 72 Huntley Street, London WC1E 6BT, UK
Research UK, Cambridge Research Institute, Li Ka Shing Centre, Robinson Way, Cambridge CB2 0RE, UK
3Department of Computer Science and Applied Mathematics, Department of Biological Regulation, Weizmann Institute, Rehovot 76100,
Israel
*Correspondence: [email protected]
http://dx.doi.org/10.1016/j.celrep.2015.02.004
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/).
2Cancer

SUMMARY

Topological domains are key architectural building

blocks of chromosomes, but their functional importance and evolutionary dynamics are not well
dened. We performed comparative high-throughput
chromosome conformation capture (Hi-C) in four
mammals and characterized the conservation and
divergence of chromosomal contact insulation and
the resulting domain architectures within distantly
related genomes. We show that the modular organization of chromosomes is robustly conserved in
syntenic regions and that this is compatible with conservation of the binding landscape of the insulator
protein CTCF. Specically, conserved CTCF sites
are co-localized with cohesin, are enriched at strong
topological domain borders, and bind to DNA motifs
with orientations that dene the directionality of
CTCFs long-range interactions. Conversely, divergent CTCF binding between species is correlated
with divergence of internal domain structure, likely
driven by local CTCF binding sequence changes,
demonstrating how genome evolution can be linked
to a continuous ux of local conformation changes.
We also show that large-scale domains are reorganized during genome evolution as intact modules.
INTRODUCTION
The discovery of a topological-domain-like three-dimensional
organization in metazoan chromosomes (Sexton et al., 2012;
Dixon et al., 2012; Nora et al., 2012; Hou et al., 2012)
is re-shaping our understanding of genome structure and
function. This new layer of large-scale genome organization
provides insights into the way by which sparsely embedded
regulatory elements could interact to drive long-range transcriptional regulation. However, the extent by which the
multi-scale domain architecture facilitates long-range regulation or is implied by it, as well as the precise mecha-

nisms organizing chromosomes into domains, is not truly

understood.
Currently, the best-characterized mechanism for domain organization involves long-range interactions between insulator
proteins (CCCTC-binding factor [CTCF] in mammals) and the
cohesin complex (Phillips-Cremins et al., 2013; Sofueva et al.,
2013; Zuin et al., 2014). CTCF is a DNA-binding protein that
engages its 11 zinc ngers to bind to DNA at a large, information-rich consensus motif (Kim et al., 2007). CTCF is a critical
transcriptional regulator, originally described as a repressor of
the myc oncogene (Filippova et al., 1996) and subsequently
shown to function as an enhancer blocker and an insulator
element (Bell et al., 1999). The insulator activity of CTCF depends
on cohesin (Parelho et al., 2008; Wendt et al., 2008), an essential
protein complex required for sister chromatid cohesion during
mitosis (Michaelis et al., 1997; Guacci et al., 1997), which also
functions in gene regulation (Rollins et al., 1999; Pauli et al.,
2008). Together, CTCF and cohesin exert their effects on gene
regulation primarily through the formation or stabilization of
long-range chromatin loops (Hadjur et al., 2009; Mishiro et al.,
2009; Nativio et al., 2009; Seitan et al., 2011). Such CTCF/cohesin-anchored loops are distributed throughout the genome,
creating a network of long-range contacts spanning multiple
scales, including not only loops that dene the borders of
strongly demarcated topological domains but also loops within
such domains (Phillips-Cremins et al., 2013; Seitan et al., 2013;
Sofueva et al., 2013; Zuin et al., 2014). While CTCF binding
specicity depends to a large extent on specic DNA sequence
elements, the specicity and directionality of CTCF/cohesin
long-range contacts (Sofueva et al., 2013) and the way by which
specic sites are assembled to dene topological domains are
not fully understood.
The dependency of CTCF recruitment on DNA sequence elements and the role for this insulator in mediating long-range
chromosomal organization suggest that CTCF may function as
a key link between genome sequence and the evolution of chromosomal domain organization. Indeed, some conservation of
chromosomal domain structures has been reported between human and mouse through both linear epigenomic analysis (Yaffe
et al., 2010) and high-throughput chromosome conformation
capture (Hi-C) comparisons (Dixon et al., 2012). Moreover, a
comparative analysis of CTCF binding in several mammalian
Cell Reports 10, 12971309, March 3, 2015 2015 The Authors 1297

genomes suggests its evolutionary dynamics are context dependent, and conservation can be interrupted by mobile element
activity (Schmidt et al., 2012). Despite these observations, a
link between the evolutionary dynamics of CTCF binding and
the evolution of chromosomal domain organization is yet to be
explored.
Studies that have tracked the evolution of different transcription factor (TF) binding patterns have shown that sequence
evolution alone is incapable of fully explaining the evolutionary
dynamics of TF binding landscapes (Dermitzakis and Clark,
2001; Birney et al., 2007; Borneman et al., 2007; Schmidt
et al., 2010). TF binding landscapes and large-scale chromosomal organization may function cooperatively to drive the evolution of genome regulation. These observations highlight the
importance of multi-species comparative chromosomal structure analysis and its integration with insulator binding proles
across evolution. If the binding patterns of trans-factors such
as CTCF are indeed strong drivers of domain organization,
then their evolutionary dynamics should drive evolutionary conservation and divergence of chromosome domains.
With this in mind, we performed comparative Hi-C in noncycling primary liver cells and analyzed the data together with
CTCF binding proles from the same species and tissue. Analysis of four mammalian Hi-C maps allowed us to explore how
the evolution of CTCF binding proles correlates, and in some
cases likely drives, the evolution of chromosomal topologies.
We nd that the large-scale chromosomal domain structure is
highly conserved between species, in a way that is correlated
with the conservation of both the CTCF binding site and the
orientation of its motif, resulting in directional long-range interactions that demarcate conserved domains. On the other hand, internal domain structure is observed to be more dynamic, and we
discover remarkable correlation between evolutionary dynamics
of CTCF sites and divergence of local insulation structure. Since
the evolution of CTCF binding proles is strongly driven at the
nucleotide level within cis elements, our data suggest that internal domain structure can be modulated exibly through local
sequence evolution. Conversely, we show that interruption of
large-scale domain structure is rare, and we suggest that instead
of local sequence divergence, evolutionary manipulation of
global chromosomal topologies is driven by processes involving
duplications or rearrangements such as inversions, insertions/
deletions, and translocations. We demonstrate this by charting
cases of evolutionary domain shufing in mouse and dog.
RESULTS
Sequence-Driven Evolution of CTCF Binding Proles
CTCF binding is strongly correlated with the topological architecture of mammalian chromosomes and participates in longrange chromatin loops, thereby underlying global contact insulation. We analyzed mouse (Mus musculus [Mmus]), dog (Canis
familiaris [Cfam]), and macaque (Macaca mulatta [Mmul]) CTCF
chromatin immunoprecipitation sequencing (ChIP-seq) proles
from primary liver cells (Schmidt et al., 2012), aiming to dene
how conservation and divergence of the insulator binding landscape co-evolve with chromosomal topology. Pairwise CTCF
ChIP-seq analysis identied conserved or divergent CTCF bind1298 Cell Reports 10, 12971309, March 3, 2015 2015 The Authors

ing sites within syntenic chromosomal regions (Figures 1A, 1B,

and S1). Sites with the strongest CTCF binding intensities were
highly conserved (77% of the top 0.1 percentile), while lower-intensity CTCF binding sites were enriched for divergent binding
(57% of mouse-divergent sites) (Figure 1B). We computed the
sequence afnity of the different classes of binding sites to
the canonical CTCF consensus motif in mouse and found that
the levels of motif afnity for conserved sites were overall higher
than the level for the mouse-divergent sites (Figure S1).
To understand the relationship between sequence afnity and
CTCF binding at conserved or divergent sites, we correlated
changes in CTCF binding with changes in CTCF sequence motif
afnity among species. For this analysis, we used the canonical
consensus motif from mouse that is the same in the other
species (Schmidt et al., 2012). Remarkably, we found a direct association between sequence divergence and CTCF binding
divergence. Conserved CTCF binding sites showed overall
high motif afnities and a high degree of afnity conservation.
Conversely, motifs underlying the divergent sites were evolutionarily dynamic and diverged in strong correlation to divergent
binding intensity (Figures 1C and S1). The data show that when
strong motifs in CTCF binding sites diverge, CTCF binding itself
is concomitantly gained or lost. Interestingly, 65% of the sites
that were conserved between mouse and dog were also
conserved in macaque, while macaque-specic and dog-specic sites constituted another two populations of 775 and 891
sites, respectively, with weaker, more evolutionarily plastic motifs. Together, these data suggest that the CTCF insulator landscape is evolving under two regimes: the rst involves a tight
conservation of both sequence and binding landscape, and the
second shows a dynamic interplay between divergence of specic cis elements and consequential evolution of the CTCF binding trait. The relatively direct inuence of motif divergence on
CTCF binding forms a potential link between sequence evolution
and large-scale genome evolution.
CTCF Binding Site Evolution Is Correlated with the
Mouse Hi-C Domain Structure
To investigate how the different classes of CTCF binding conservation correlate with chromosomal structure, we prepared
Hi-C datasets on mouse liver cells (Figure S2). Filtering and
normalization of the Hi-C ligation products was performed as
before (Sofueva et al., 2013), revealing the characteristic chromosomal domain structure in these cells. Visualization of the
CTCF occupancy groups with the Hi-C contact maps suggested
that conserved CTCF binding sites were found at the borders
of large-scale Hi-C domains while species-specic CTCF sites
are located internal to domains (Figure 1D). This observation
was supported with a genome-wide analysis whereby the
relative position of conserved and divergent CTCF sites was
determined with respect to all domains in the mouse genome
(Figure 1E).
To further characterize chromosomal contacts around conserved and divergent CTCF sites, we analyzed the average contact distribution around these sites globally, measuring contact
insulation by quantifying the decrease in contact probability between multiple elements separated by a CTCF site (Sofueva
et al., 2013). Analysis of the composite contact insulation at

Figure 1. Evolution of CTCF Binding Correlates with Hi-C Domain Structure

250 Kb

Mmus
CTCF

Cfam
CTCF

Mmul
CTCF

4
6
2

5
4

0
3
-2

Mmus

Cfam CTCF ChIP signal

(A) Representative CTCF ChIP-seq tracks on

mouse (Mmus) chromosome 5; dog (Cfam) and
macaque (Mmul) tracks are shown after liftOver
to the mouse genome (mm10). Sites conserved
across all three species (arrowhead) or specic
to a single species (asterisks) are indicated as
examples.
(B) A pairwise comparison of mouse CTCF ChIP
and dog CTCF ChIP (liftOver track), identifying
conserved or divergent binding sites (see Experimental Procedures).
(C) A comparison of the interspecies difference in
CTCF ChIP signal against the difference in CTCF
motif afnity in mouse versus dog. Scatterplots are
highlighted for conserved (Mmus+/Cfam+, blue),
mouse-divergent (Mmus+/Cfam, black), and
dog-divergent sites (Mmus/Cfam+, green).
(D) A representative 2-Mb region from chromosome 5 of the mouse Hi-C contact maps. Also
shown are the mouse (Mmus) CTCF ChIP track
and conserved (Mmus+/Cfam+, blue) and divergent (Mmus+/Cfam, gray) CTCF sites.
(E) Genome-wide relative position of conserved
and divergent CTCF sites within mouse Hi-C domains.
(F) Contact insulation analysis for conserved
(Mmus+/Cfam+) and mouse-divergent (Mmus+/
Cfam) CTCF sites in the mouse Hi-C data. See
also Figure S1.

1
0
0

-4
-4

-6

-2

Mmus

Mmus CTCF ChIP signal

Cfam ChIP signal

Mmus CTCF
Mmus+/Cfam+
Mmus+/Cfam-

Mmus+/Cfam+
Mmus+/Cfam-

Normalized count

2.5

Contact enrichment
(log2 scale)

52.5 16
14
11

52 -

7
2

dynamics of CTCF binding sites and

mouse chromosome topology, indicating
the possibility of a direct link between
insulator site divergence and the evolution of topological domain structure.

1.5

51.5 0

20

40

60

80

100

Relative position in Mmus domains

Comparative Hi-C Reveals the

Evolution of Chromosome
Topologies
chr 5
We used comparative Hi-C to examine
conservation and divergence of chromo0.3
Mmus+/Cfam+
Mmus+/CfamF
some topology and to test how evolution
0.1
80K
40K
0
of CTCF binding sites might underpin
20K
10K
-0.1
this. We collected liver cells from ma5K
-0.3
caque,
rabbit (Oryctolagus cuniculus
-200 -100
0
100 200
-200 -100
0
100 200
-0.5
[Ocun]),
and
dog and processed them usDistance from site (Kb)
Distance from site (Kb)
ing the same approach applied to mouse,
yielding chromosomal contact maps for
multiple distance ranges, indicated strong insulation proles for each of these species (Figures S2 and S3). Evaluation of the
conserved CTCF sites, further supporting the idea that these overall topological structure within the three newly proled
conserved, high-intensity CTCF sites were co-occurring with species rst indicated the integrity of their reference genome
the borders of large-scale domain (reminiscent of topological structures and generated a resource for future renement of
chromosomal domains) (Figure 1F, left panel). In comparison, such assemblies. More importantly, the data showed that the
the lower-intensity mouse-divergent sites showed a signicantly chromosome topologies in macaque, dog, and rabbit are charweaker, more localized insulation prole (Figure 1F, right panel). acterized by a chromosomal domain structure that is similar to
Similar trends were also observed when classifying CTCF sites the one inferred before for human and mouse (Dixon et al.,
according to their conservation in macaque (Figure S1). In 2012). For example, comparison of a 9-Mb syntenic region highsummary, we found strong correlation between the evolutionary lighted the extensive conservation of chromosomal structure
Contact enrichment
(log2 scale)

Band

51 -

Cell Reports 10, 12971309, March 3, 2015 2015 The Authors 1299

Ocun

Mmus
Crossover 160kb
Crossover 80kb
Orth. genes
Clusters
96 42 98 chr 12

chr 10

40 -

38 -

100 -

102 Contact enrichment

(log2 scale)

36 104 -

Cfam

Mmul
Crossover 160kb
Crossover 80kb
Orth. genes

16
14
11
7
2

Clusters
156 -

64 -

66 chr 12

chr 4

154 -

152 -

68 -

150 -

70 -

148 -

Mmus
Prk

Ocun

Hs3st5
Hdac2

Mmul

Hs3st5
Hdac2

Prk
102

150

36.5

102.5

149.5

36

103

35.5

Hs3st5
Hdac2

Prk

37

Cfam
Hs3st5
Hdac2

Prk
70

70.5
149

103.5

148.5

104

148

71

35
34.5

71.5

10

10

12

Mmus interactions

14

12

Cfam interactions

12

Mmul interactions

Ocun interactions

10

10

12

Mmus interactions

14

12

10

10

12

14

Mmus interactions
(legend on next page)

1300 Cell Reports 10, 12971309, March 3, 2015 2015 The Authors

across all species (Figure 2A). The maps also revealed evidence
of intra-domain differences between species (Figure 2B). We
quantied the extent of structural conservation genome-wide
using a computational approach that allowed us to comprehensively describe domain structure at multiple scales. This pairwise
approach revealed extensive genome-wide interspecies conservation of chromosome structure (Figures 2C and S3). A systematic analysis of paired domains in mouse and dog revealed that
conserved domains are smaller in size compared to other domains and are classied as both active and passive clusters (Figure S4). Together, these data facilitated extensive analysis of the
evolution of chromosomal topologies within regions that did not
go through substantial genome rearrangement, allowing examination of the evolution of both large-scale domain borders and
the insulation structure within domains.
Divergent CTCF Binding Drives Local Structural Change
within Domains
Hi-C maps from liver cells of different species allowed us to ask
how the evolutionary dynamics of CTCF correlate with conservation or divergence of domain structure. Analysis of specic loci
showed that conserved CTCF sites were typically located at
the borders of large-scale chromosomal domains that were
themselves conserved between mouse and dog (Figure 3A). To
test these observations globally, we computed the contact insulation proles from either the mouse or dog Hi-C maps around
conserved CTCF sites, showing that these sites indeed globally
served as conserved insulation points (Figure 3B). Similar results
were derived using a comparison of mouse and macaque (Figure S5). We also observed that conserved CTCF sites were
strongly enriched for Rad21 in mouse (79% of conserved sites
compared to 51% mouse-divergent sites co-localize with
Rad21) and that CTCF/cohesin co-occupied sites exhibited
strong contact insulation in all three species (data not shown).
In contrast to these highly stable sites, our data showed that
divergent CTCF sites were located primarily within domains
and exhibited local contact insulation. Comparative analysis of
contact insulation at divergent CTCF sites revealed that indeed
these sites correlated with divergent contact insulation proles.
For example, dog-divergent CTCF sites (Mmus/Cfam+) exhibited local contact insulation specically in the dog genome,
whereas these same sites exhibited background levels of contact insulation when examined in the mouse Hi-C data (Figures
3C and S5). Importantly, the change in insulation following
CTCF binding site evolution was stronger at the local (20-kb)
scale, but was not signicant at the higher (80-kb) scale (Fig-

ure 3D), suggesting that large-scale domain changes either are

not affected by CTCF evolution or are under strong negative selection and are therefore not observed. These observations were
further strengthened when we examined CTCF binding sites
that were partially conserved. CTCF sites that were bound in
mouse and dog, but lost in macaque, were associated with
reduced contact insulation in the macaque genome. Thus, the
data demonstrate a relationship between CTCF binding divergence and divergence of local insulation structure and therefore
point to a role for CTCF in driving structural change in the
genome.
The continuous evolutionary dynamics of intra-domain looping can play a key role in tuning promoter-enhancer contacts
within domains. Consistent with this, we observe long-range
contacts between divergent CTCF sites and enhancers or
transcription start sites (TSSs) (Figure S6). Furthermore, analysis
of transcription data from mouse, dog, and macaque liver reveals that divergent CTCF sites are contacting differentially expressed genes with a greater frequency than non-differentially
expressed genes (Kolmogorov-Smirnov test, p < 0.05) (Figure S6). Together, these data support the hypothesis that emergence of divergent CTCF binding sites can contribute to changes
in gene expression.
Conserved CTCF Sites Are Directional and Interact with
Other Conserved Sites
While it is known that CTCF binding specicity greatly depends
on its specic DNA consensus sequence, the specicity and
directionality of CTCF/cohesin long-range contacts (Sofueva
et al., 2013) and the way by which specic sites are assembled
to dene topological domains are not fully understood. As our
data indicated that conserved CTCF binding sites have
conserved motif afnities (Figure 1C), and because it is known
that the CTCF consensus motif is non-symmetric, we asked
whether conserved sites could also be conserved for the orientation of the CTCF motif. Indeed, 94% (3,265/3,483) of CTCF
binding sites that are conserved between mouse and dog are
also conserved in their orientation. To explore this further, we
proled contact insulation around conserved CTCF binding sites
grouped according to the strand that the consensus motif was
found on. We observed an asymmetric insulation behavior that
was mirrored when the orientation of the motif was reversed (Figure 4A). This analysis uncoupled insulation (blue) from preferential contacts (red) and revealed that preferential contacts are
made on one side of the oriented CTCF binding site, indicating
that the orientation of the motif likely contributes to directionality

Figure 2. Chromosomal Domain Structure Is Robustly Conserved in Mammals

(A) Representative Hi-C contact maps of a 9-Mb syntenic region from mouse (Mmus), rabbit (Ocun), macaque (Mmul), and dog (Cfam). Maps are colored according to technically corrected contact enrichment. For scale purposes, the same size region is shown for all species, but in the macaque and dog genome, the
syntenic region is reduced in size. The portion of the map that is outside of the synteny boundary is shaded out. Shown above each map is the quantication of
contacts for distance bands of 160480 kb (crossover 160 kb) and 80240 kb (crossover 80 kb), as well as a track of orthologous genes, highlighting their
conserved distribution in chromosomal domains.
(B) A zoom in of the domain highlighted with a white dashed box in (A) reveals differences in its internal organization across species.
(C) Global quantication of the correlation between the maps. Genome-wide contact enrichments between elements separated by 160480 kb in rabbit, macaque, and dog were compared to the mouse genome after liftOver. Spearman correlation values are shown inside the plots. Axes units are contact enrichments
[log2(observed/expected)].
See also Figures S2 and S3.

Cell Reports 10, 12971309, March 3, 2015 2015 The Authors 1301

Cfam Hi-C

Mmus Hi-C

Mmus CTCF

Cfam CTCF

Mmus+/Cfam+

Mmus+/Cfam+
Mmus-/Cfam+

Mmus+/Cfam-

62.5 90.5 Contact enrichment

(log2 scale)

63 90 63.5 89.5 64 -

16
14
11
7
2

89 64.5 -

chr 8

chr 2

Hi-C
Mmus
Mmus+
Cfam+

Cfam

80K
40K
20K
10K
5K

-200

-100

100

200

-200

-100

100

200

CTCF ChIP

C
Mmus+
Cfam-

80K
40K
20K
10K
5K

-200

MmusCfam+

-100

100

200

-200

-100

100

200

80K
40K
20K
10K
5K

-200

-100

100

200

-200

Distance from site (Kb)

-100

100

Contact enrichment
(log2 scale)

CTCF ChIP

0.3
0.1
0
-0.1
-0.3
-0.5

200

Distance from site (Kb)

Mmus - Cfam
contact enrichment (80kb)

contact enrichment (20kb)

3
2
1
0
-1
-2
-3

2
1
0
-1
-2
-3

Mmus+
Cfam-

Mmus+
Cfam+

MmusCfam+

Mmus+
Cfam-

Mmus+ MmusCfam+ Cfam+

(legend on next page)

1302 Cell Reports 10, 12971309, March 3, 2015 2015 The Authors

of CTCFs long-range interactions. Consistent with this, we proled the genome-wide relative position within chromosomal
domains of Mmus+/Cfam+ conserved CTCF sites grouped according to the orientation of their binding motif as above. We
observed that the conserved CTCF binding sites that are enriched at the edges of conserved domains (Figure 1E) have a
specic orientation of their motif relative to chromosomal domains (Figure 4B). These observations were replicated when
we compared mouse and macaque.
To characterize the contact relationship between evolutionarily stable or exible CTCF sites and to further understand
how they contribute to the evolution of chromosome domain
structure, we performed a high-resolution high-throughput
circular chromosome conformation capture (4C-seq) study. We
designed four 4C-seq viewpoints to a series of neighboring
conserved CTCF binding sites bordering conserved domains in
the mouse and dog as well as to a mouse-specic site. The results showed that each conserved CTCF site engages in very
strong and directional interactions with neighboring conserved
CTCF sites (Figure 4C). Remarkably, the specic interactions
mediated by conserved sites in the mouse genome were themselves precisely conserved in the dog genome (Figure 4D) and
dene the underlying domain structure. In each case, the longrange interaction was anchored by a pair of conserved CTCF
sites whereby one CTCF site had an orientation on the + strand
and the other on the  strand and could provide the basis for
the observed directionality of CTCF-mediated interactions.
Moreover, a viewpoint designed to a mouse-divergent site exhibited weak interactions within the mouse domain, analogous
to the local insulation behavior observed in Figure 3B (Figure 4C).
Importantly, the mouse-divergent viewpoint had no prominent
interactions in the dog genome, conrming the specicity of its
interaction network.
Global analysis of Hi-C contacts between pairs of CTCF binding sites stratied according to genomic distances in cis (Sofueva et al., 2013) conrmed the 4C-seq observation systematically (Figure 4E). Consistent with the high-resolution 4C-seq
proles, Hi-C trends showed that conserved CTCF sites strongly
contacted one another within the same domain. Divergent CTCF
sites engaged in signicantly weaker contacts with other divergent sites, even when stratifying thoroughly for genomic distances. Importantly, little to no contact was observed in the
mouse genome between dog-divergent sites. These results
show that evolutionarily stable CTCF sites are engaged in strong
contacts with one another and suggest that in so doing, they
create an interaction network that may support the conservation

of domain structure. On the other hand, divergent CTCF sites are

involved in weaker interactions, perhaps reecting the evolutionary exibility of the binding sites themselves.
Domains Maintain Their Integrity during Chromosomal
Rearrangements
Our data suggested that large-scale domain re-organization
does not typically occur following insulator divergence. How
then can it still be observed? Our interspecies comparative HiC data allowed us to ask what happens to the integrity of
conserved chromosomal domains when genomes are challenged by structural rearrangements. If chromosomal domains
act as modular units (e.g., to regulate gene expression), then
large-scale rearrangements would be expected to occur at
domain borders, so as to maintain the integrity of these structures. We scanned the mouse and dog genomes for differences
in the distance between contiguous orthologous genes in the
two species. Our analysis uncovered a number of complex rearrangements between the mouse and dog genomes involving
insertions, inversions, and duplications. In each case, we discovered that the rearrangement occurred at the border between two
chromosomal domains. This is exemplied in the Hi-C map from
chromosome 15 in dog (Figure 5). Here, we found two domains,
one containing the Slc5a9 gene and the other containing the
Trabd2b gene (highlighted by red dots). Comparison of this region to the mouse genome revealed that a 2-Mb insertion
occurred in the mouse genome that contains the Skint gene cluster, which is rapidly evolving and unique to the mouse lineage
(Boyden et al., 2008). Remarkably, the insertion occurred directly
between two neighboring dog domains in such a way as to
perfectly maintain their integrity. A similar rearrangement event
occurred at the Mrgpr gene cluster in the mouse genome
(Dong et al., 2001), again preserving the structure of the neighboring domains (Figure S7). In another example, we observed
a large-scale 5.5-Mb insertion in the dog genome containing
multiple domains, and again, the domains on either side of the
insertion have been maintained intact (Figure S8). These examples suggest that domains function as modular units and are
selected against breakage during genome rearrangements.
DISCUSSION
In this study, we examined Hi-C contact maps and CTCF binding
proles from four mammalian species to understand the relationship between the evolution of CTCF binding sites and chromosome structures. Our data reveal that CTCF binding sites have

Figure 3. Conserved and Divergent CTCF Sites Show Differential Contact Insulation Behavior
(A) Representative Hi-C contact maps for a 2-Mb syntenic region in mouse (left panel) and dog (right panel). Also shown are the CTCF ChIP-seq tracks in each
species as well as the conserved (Mmus+/Cfam+, blue) and divergent (Mmus+/Cfam, gray or Mmus/Cfam+, green) CTCF sites.
(B) Average contact insulation analysis at conserved (Mmus+/Cfam+) CTCF sites in mouse (leftmost panel) and dog (rightmost panel) Hi-C datasets. Arrows
indicate liftOver of sites.
(C) Same as for (B), but for mouse-divergent (Mmus+/Cfam) and dog-divergent (Mmus/Cfam+) sites in both genomes. Divergent sites appear to mediate
weaker, shorter-range insulation that disappears in the species where the site is not bound.
(D) Distribution of the difference in contact insulation between mouse and dog at conserved or divergent CTCF sites at 2060 kb (left panel, 20-kb band) or 80240
kb (right panel, 80-kb band) scales. Divergent sites exhibit a signicant (Kolmogorov-Smirnov test, p < 1 3 1012, marked by an asterisk) shift in their distributions
in the 20-kb band, but not in the 80-kb band, compatible with them mediating insulation at a local scale, but not at higher ranges.
See also Figure S5.

Cell Reports 10, 12971309, March 3, 2015 2015 The Authors 1303

(legend on next page)

1304 Cell Reports 10, 12971309, March 3, 2015 2015 The Authors

evolved under two regimes, whereby some CTCF elements are

constrained both at the level of DNA sequence as well as in their
binding while other CTCF elements exhibit signicantly more
exibility. While both groups can mediate contact insulation,
conserved CTCF elements are enriched at large-scale domain
borders that tend to be themselves conserved. Meanwhile,
evolutionarily exible CTCF sites tend to be located internal of
large-scale domains and mediate local structural change
uniquely in that lineage. Our data thereby point to a strong correlation between the evolution of CTCF binding and chromosomal
structure and extend on our current understanding of contextdependent CTCF binding sites and their specic roles in chromosomal domain architecture (Dixon et al., 2012; Dowen et al.,
2014). Importantly, since CTCF binding information is encoded
in high specicity cis elements, the intra-domain insulator dynamics we observe directly link local sequence evolution with
chromosomal architectures. This direct linkage has strong
implications for the study of CTCF and genome function and
for our understanding of the evolutionary dynamics in complex
genomes.
A central causal role for CTCF/cohesin in establishing domain
structure is widely hypothesized, but direct experimental evidence has proven difcult to attain. Previous studies have
observed a correlation between insulator binding and domain
borders (Sexton et al., 2012; Dixon et al., 2012; Nora et al.,
2012; Hou et al., 2012), and knockout experiments have suggested a quantitative link between loss of chromosomal looping
structure and loss of the CTCF/cohesin binding landscapes (Sofueva et al., 2013; Zuin et al., 2014; Seitan et al., 2013). Given
the pervasive impact of CTCF/cohesin on nuclear organization
and gene regulation, it is difcult to identify the mechanisms
of their action through classic genetic perturbation. Instead,
the evolutionary comparison used here offers us thousands of
naturally occurring genomic perturbations that can be identied
and characterized at both the sequence and chromosomal
topology levels. This strategy has yielded strong evidence of
a direct link between the gain/loss of CTCF binding sites
and a corresponding gain/loss of local domain insulation. Our
comparative Hi-C analysis therefore strongly supports the
idea that CTCF is causally connected to chromosomal looping
structures.
The comparative chromosomal domain analysis described
here has revealed a spectrum of evolutionary consequences,
ranging from the conservation of essential large-scale chromo-

somal domains to the exibility of continuous genomic adaptation. CTCF and cohesin complexes are deeply evolutionarily
conserved, and the data here show that their role in mediating
chromosome topologies and, even more remarkably, the
large-scale building blocks of such topologies are also highly
conserved. Our data suggest that the orientation of the CTCF
motif may underlie the observed directionality of CTCF/cohesin-mediated long-range contacts and provide a rationale by
which specic sites are assembled to dene topological domains. Given that CTCF binding is strongly inuenced by its
consensus sequence, our data suggest that the assembly of
domain structure is hardwired in the genome. This also
has implications for further understanding the nature of the
relationship between CTCF and cohesin, since biochemical
studies have revealed that cohesin subunits interact with
CTCF primarily through its C-terminal tail (Xiao et al., 2011),
placing cohesin on a particular side of the chromosomal
domain.
Interestingly, while we were able to observe cases whereby
local sequence evolution perturbed CTCF binding and disrupted
chromosomal looping, the structures that were affected due to
this insulator divergence were primarily local loops. Cases of
large-scale topological domains that were split or fused due to
insulator divergence were not observed. We hypothesize that
this stability is achieved by a combination of both local purifying
selection on key CTCF binding sites and by buffering of major topological loops by additional factors. Strikingly, the cases of
large-scale domain divergence that we were able to characterize
were all linked with evolutionary genome rearrangements and revealed a mechanism that can reshufe whole domains such that
the rearranged chromosomal modules are aligned with existing
domain borders. It is still, however, formally possible that rearrangements take place between CTCF sites that are mediating
strong interactions.
In addition to the importance of topological domain and insulator conservation described here, the evolutionary dynamics
that couple intra-domain CTCF divergence with changes in the
local domain structure emerge as potentially fundamental for
genome regulation. Loops contained within domains link enhancers (and their bound trans-factors) to target gene promoters. While it is still unclear how such targeting is regulated
and how evolution can manipulate it, based on our data, we hypothesize that exible CTCF binding sites within domains can
inuence looping from the promoter or enhancer as well by

Figure 4. Conserved CTCF Sites Engage in Strong, Directional Interactions with Other Conserved Sites
(A) Average contact insulation analysis in both Mmus and Cfam genomes for conserved (Mmus+/Cfam+) CTCF binding sites grouped according to the orientation
of their binding motif. The consensus motif shown was generated from mouse binding sites.
(B) Genome-wide relative position within chromosomal domains of Mmus+/Cfam+ conserved CTCF sites grouped according to the orientation of their binding
motif as above.
(C and D) (C) 4C-seq analysis and Hi-C maps of a 1.2-Mb syntenic region in the mouse and (D) dog genomes. Shown are 4C-seq viewpoints designed to four
conserved (Mmus+/Cfam+) CTCF binding sites proximal to Hi-C domain borders and one 4C-seq viewpoint located at a mouse-divergent (Mmus+/Cfam) CTCF
site. The symbol above each 4C-seq bait indicates the strand (and orientation) of each viewpoint. Each 4C-seq experiment is represented by the median
normalized 4C-seq coverage in a sliding window of 5 kb (top) and a multi-scale domainogram indicating normalized mean coverage in windows ranging between
2 kb and 50 kb.
(E) Relative intra-domain contact enrichment between CTCF sites (solid lines) as a function of distance when the two sites are <5 kb away from a CTCF binding
site (solid lines) or where only one site is <5 kb away from a CTCF site (dotted lines). Shown are the relative contact enrichments within repressive domains for
conserved (Mmus+/Cfam+, blue), mouse-divergent (Mmus+/Cfam, gray) and dog-divergent (Mmus/Cfam+, green) CTCF sites in mouse Hi-C (left) and dog
Hi-C (right).

Cell Reports 10, 12971309, March 3, 2015 2015 The Authors 1305

Cfam
chr15

Mmus
chr4
108

112

114.2 116Mb

12.6

14.1Mb

Skint gene
family
insertion

B
Cfam

Mmus
Mmus
CTCF
Mmus+
Cfam+
Mmus+
Cfam-

Cfam
CTCF
Mmus+
Cfam+
MmusCfam+

Genes

Genes

116 16 115 15 114 14 113 13 112 12 111 11 110 10 109 9chr 4

chr 15
Contact enrichment
(log2 scale)
2

11 14 16

Figure 5. Chromosomal Domains Evolve as Modular Units

(A) A schematic of the rearrangement between mouse (gray) and dog (green).
(B) Hi-C maps from a syntenic region in mouse (left) and dog (right). Shown also are CTCF ChIP tracks, conserved and divergent sites, and genes in the region. A
cluster of non-orthologous Skint genes (black dots) has been inserted in mouse between the orthologous genes Slc5a9 and Trabd2b (highlighted as red dots). The
inserted region (bordered in yellow) forms its own nested chromosomal domain structure, probably as a result of gene-duplication events. Highlighted in blue are
other orthologous genes in the region.
See also Figures S7 and S8.

demarcating the implicated functional elements. As CTCF sites

are sufciently sequence-specic to be directly tunable by local
nucleotide substitutions, it is intriguing to speculate that the
intra-domain looping structure is a key and evolvable feature
affecting gene regulation. Such a trait, if indeed quantitatively
important, should be further studied between and within populations and species.
1306 Cell Reports 10, 12971309, March 3, 2015 2015 The Authors

EXPERIMENTAL PROCEDURES
Liver Homogenization and Fixation
Fresh or frozen liver from mouse, rabbit, macaque, and dog were processed
for Hi-C or 4C-seq libraries. With the exception of mouse, the samples used
for the Hi-C libraries were the same as the material used for CTCF ChIP-seq
(Schmidt et al., 2012). Livers were xed in 10% formalin for 20 min, and
1 g was cut and processed with a Dounce homogenizer (ten strokes with a

Table 1. 4C-Seq Primers Used in This Study

Viewpoint

CTCF Peak

Reading Primer

Non-reading Primer

chr10:94609250

50 -CCATCTGTTTGAACAAGATC-30

Mouse 4C-seq primers (mm10)

Mmus+ Cfam+ 1 (Figure 4)
Mmus+ Cfam+ 2 (Figure 4)
Mmus+ Cfam+ 3 (Figure 4)
Mmus+ Cfam+ 4 (Figure 4)
Mmus+ Cfam (Figure 4)

chr10:94623583
chr10:94958324
chr10:94991353

5 -AGTCAGATGGAATGCAGATC-3
0

5 -ATTGCTTTCTCTGGTTGATC-3

50 -CAAGAGAGAGTGGAAACAGG-30
0

50 -CTAGATACAGCAATCAGCCC-30
50 -AGTCACTCCTGCTCCTGTAA-30

50 -AAGCATTGTCCTACGTGATT-30

50 -CCCTTCCCTTCTATGTTTCT-30

5 -GTTTCTGTTGGTTCACGATC-3

chr10:95218005

5 -CTACTCTGGCTTCTATGATC-3

chr15:34606369

50 -GCTCTTGCTCTAAACTGATC-30

Dog 4C-seq primers (canFam3)

Mmus+ Cfam+ 1 (Figure 4)

50 -TGGACCTCACCTCTCCTA-30
50 -GTCGCATCACTTACTGGG-30

Mmus+ Cfam+ 2 (Figure 4)

chr15:34596229

5 -TGAGGTCCAGCAGAGATC-3

Mmus+ Cfam+ 3 (Figure 4)

chr15:34269944

50 -CTCCACTGAGCATTAAGATC-30

50 -GCGGGATAGTTCTTTTCTCT-30

Mmus+ Cfam+ 4 (Figure 4)

chr15:34244336

50 -CTTATGTGCTCCTCCAGATC-30

50 -AATCATATGCCTCCTCCTCT-30

Mmus+ Cfam (Figure 4)

chr15:33989305

50 -AAAGTAATCCCACCCAGATC-30

50 -CTGAAGGAAACAACAATGTCA-30

loose pestle followed by ten strokes with the tight pestle). After ltration
through a 70-mm nylon cell strainer, the sample was washed twice with PBS,
spinning down at 852 rcf for 5 min at 4 C to collect the cells between washes.
15 3 107 liver cells were then xed for a second time in xation buffer (1%
formaldehyde, 750 mg/ml BSA in DMEM/Hams F12 [Invitrogen]) for 10
30 min at room temperature. The xation reaction was quenched using
0.125 M glycine for 5 min at room temperature. Samples were washed twice
with 10 ml PBS, pelleted into 1 3 107 cells aliquots, and stored at 80 C.
Mouse Hi-C libraries were prepared from fresh liver samples of biological replicates (9-week-old C57/BL6 mouse and the pooled livers from 2- to 4-weekold outbred mice. The libraries for the other three organisms were technical
replicates.
Propidium Iodide Staining of Hepatocytes
Formaldehyde-xed liver cells were lysed on ice in a hypotonic buffer (10 mM
Tris-HCl [pH 8], 10 mM NaCl, 0.2% Igepal CA-640, EDTA-free protease inhibitors) for 30 min. Nuclei were stained with a propidium iodide (PI) staining buffer
(100 mg/ml PI, 50 mg/ml RNase A, 0.05% Triton X-100) for 60 min on ice. Samples were analyzed on a MoFlo cell sorter (Beckman Coulter).
High-Throughput Mapping of Chromatin Interactions via Hi-C
The Hi-C method previously used (Sofueva et al., 2013) was modied to
accommodate primary liver samples. Hepatocytes were lysed in Hi-C lysis
buffer (10 mM Tris-HCl [pH 8], 10 mM NaCl, 0.2% Igepal CA-640, EDTA-free
protease inhibitors) for 30 min on ice. The sample was transferred to Protein
LoBind tubes (Eppendorf) and the nuclei were permeabilized by incubation
with 0.1%0.6% SDS for 1 hr at 37 C with 800 rpm shaking. The reaction
was quenched with 0.67%4% Triton X-100, 1 hr at 37 C, 800 rpm shaking.
Nuclei were digested in 500 ml 1X NEBuffer 2 with 1500 U HindIII (New England
Biolabs) and monitored for maximal digestion of the chromatin template, thus
digestion times ranged from 24-72 hr. All other parts of the Hi-C protocol,
including library preparation were performed as previously described. 75 bp
paired-end sequencing was performed for each library according to manufacturers conditions using the Illumina Hi-seq platform.
Hi-C Interaction Matrix Generation and Domain Calling
Sequencing reads were aligned to the mouse (mm10), rabbit (oryCun2), macaque (rheMac2), and dog (canFam3) genome assemblies using Bowtie
0.12.8 (Langmead et al., 2009). The parameters used for the alignment allowed
a maximum of three mismatches and strictly one alignment per read. Processing of the aligned reads and normalization of the interaction matrices were
performed as previously described (Yaffe and Tanay, 2011; Sofueva et al.,
2013). The pipeline produced normalized matrices of interactions binning the
genome at different resolutions. Interaction matrices for each library were
generated displaying seven different resolutions simultaneously (12,500,
25,000, 50,000, 100,000, 250,000, 500,000, and 1,000,000 bp). Domains

were identied and clustered as described (Sexton et al., 2012) with the modication that scaling factors were inferred using fends 100400 kb apart, to
account for the lower resolution of the mouse map compared to the Drosophila
map. Domain borders were called using the 95% percentile of the scaling
track. A domain-level map was partitioned into two clusters, and clusters
were assigned as passive/active according to Lamin B mouse embryonic
broblast (MEF) data, as before. For the rabbit, macaque, and dog genome,
the Lamin B MEF track for mouse was lifted over to the corresponding genome
to label domain clusters. Domain calls in mouse and dog are available in
Table S1.
ChIP-Seq Analysis
We used previously published ChIP-seq data for CTCF from mouse, macaque,
and dog livers (Schmidt et al., 2012) and for Rad21 for mouse liver (Faure et al.,
2012). Rad21 ChIP-seq data for macaque and dog was prepared as for CTCF.
Mouse, macaque, and dog ChIP-seq reads were mapped using bowtie. Alignment was followed by extension of sequenced tags to 300-bp fragments and
pileup into 50-bp bins. We normalized ChIP-seq coverage by computing the
distribution of pile-up coverage on 50-bp bins and transforming each coverage
value v into log10 (1-quantile(v)). To dene binding sites, we used a simple
threshold on the sum of values from two biological replicates for each CTCF
dataset and for the macaque Rad21 data. Rad21 ChIP data from mouse and
dog were done in single, and the data were thresholded. Thresholds used
were as follows: mouse CTCF = 2.2, macaque CTCF = 2.4, dog CTCF = 2.2,
mouse Rad21 = 2.3, macaque Rad21 = 2.5, dog Rad21 = 3. Different thresholds did not change the results. Binding site width was standardized at 200 bp,
and the ChIP-seq intensity for each site was calculated as the maximum value
across the 200 bp. The relative distribution of CTCF within topological domains
(Figures 1E and S5) was calculated as the distance of each CTCF site from the
center of its domain. Half the size of the domain was added to convert it to a
measure of distance from the edge of the domain, and this number was then
divided by the size of the domain.
Interspecies Comparison of CTCF Sites
Macaque and dog CTCF ChIP-seq libraries were converted to mouse genome
coordinates using the liftOver tool from UCSC. To reduce the chance of inaccurate liftOver, a number of lters were implemented: sites within low-mappability regions, repeats, or windows of 100 kb with insufcient synteny were
excluded. To estimate mappability, each genome was broken into 50-bp reads
and the whole-genome sequence was split into articial reads and then mapped back to the genome. For each 50-bp bin, the mappability score was then
dened to be the portion of articial reads mapped uniquely to that bin. To estimate the level of synteny in the 100 kb around a CTCF site, the mappability
tracks for macaque and dog were converted to the mouse genome using liftOver and all bins for which liftOver was not possible were converted to zeroes.
The converted tracks were subsequently smoothed over 100 kb, and CTCF

Cell Reports 10, 12971309, March 3, 2015 2015 The Authors 1307

sites falling in regions below the top quartile of such smoothed tracks were
excluded from all subsequent analysis. Divergent CTCF sites in mouse and
dog are available in Table S1.
CTCF Binding Energy Function
A CTCF DNA-binding energy function from the Cortex CTCF binding sites
(ENCODE Cortex CTCF mouse, GSM769019; Shen et al., 2012) was used to
prole all genomes for their similarity to the CTCF consensus motif. The
consensus motif is very highly conserved across all species (Schmidt et al.,
2012). Given a set of genomic sites, we compute for each site the maximal energy value within a 200-bp window centered on the point.
Motif Orientation Analysis
Orientation of the motifs underneath conserved CTCF peaks was obtained using MEME (http://meme.nbcr.net/meme/), (Bailey and Elkan, 1994) with the
parameters -revcomp -dna -nmotifs 1 -w 20 -mod zoops -maxsize 100,000.
Crossover Analysis
Crossover analysis was performed as described previously (Sofueva et al.,
2013). The bands used were 57.5, 7.511.25, 1015, 1522.5, 2030, 30
45, 4060, 6090, and 80120 kb.
Distal Contact Analysis
To calculate the average interaction proles for a group of genomic landmarks,
HindIII fragment ends were grouped into classes by associating each end with
a genomic element located within 5 kb and then grouping all fragment ends
associated with an element of the same class. For the mouse, macaque,
and dog genomes, three classes of CTCF sites (conserved, divergent present,
divergent absent) and TSS sites were dened. These classes were further
divided to sites within active or passive Hi-C domains. The remaining fragment
end (not classied given other landmarks) was dened as the background.
4C-Seq
Preparation of 4C-seq samples, libraries, sequencing analysis, and normalization were all performed as previously described (Sofueva et al., 2013). Primer
sequences were chosen to viewpoint sites that were as close as possible to
CTCF ChIP-seq peaks (Table 1). Mouse primers were designed according to
the genome-wide 4C-seq primer database from (van de Werken et al.,
2012). For dog primers, a similar database was generated for the regions of
interest.
ACCESSION NUMBERS
The data analyzed in this study have been deposited in the GEO database with
the accession number GSE65126.
SUPPLEMENTAL INFORMATION
Supplemental Information includes eight gures and one table and can be
found with this article online at http://dx.doi.org/10.1016/j.celrep.2015.02.004.
AUTHOR CONTRIBUTIONS
M.V.R., D.T.O., and S.H. initiated the project. M.V.R. performed the Hi-C and
4C-Seq experiments, including library preparations. D.T.O. and C.E. provided
the liver samples for all species (except mouse) and sequenced the Hi-C libraries. M.V.R., C.B., and A.T. processed and statistically analyzed the data.
M.V.R., A.T., and S.H. wrote the manuscript, with contributions from all
authors.
ACKNOWLEDGMENTS
The authors wish to thank Sevil Sofueva for help with Hi-C and 4C-seq library
preparations; Wen-Ching Chan for analysis; Christine Feig for Rad21 ChIP-seq
data in macaque and dog; Bianca Schmidt and the Cancer Research UK CI
Genomic Core facility for technical assistance with Hi-C sequencing and Pe-

1308 Cell Reports 10, 12971309, March 3, 2015 2015 The Authors

dro Olivares for advice on analysis and data manipulation. We would also
like to acknowledge all members of the Hadjur group for discussions. This
work was supported by the Medical Research Council UK (G0900491/1 and
G1001649) (S.H.), the EPIGENESYS EU NoE (S.H. and A.T.), and Cancer
Research UK (studentship to M.V.R.). D.T.O. is supported by Cancer
Research UK.
Received: September 27, 2014
Revised: December 9, 2014
Accepted: January 29, 2015
Published: February 26, 2015
REFERENCES
Bailey, T.L., and Elkan, C. (1994). Fitting a mixture model by expectation maximization to discover motifs in bipolymers. Proc. Int. Conf. Intell. Syst. Mol. Biol.
2, 2836.
Bell, A.C., West, A.G., and Felsenfeld, G. (1999). The protein CTCF is required
for the enhancer blocking activity of vertebrate insulators. Cell 98, 387396.
Birney, E., Stamatoyannopoulos, J.A., Dutta, A., Guigo, R., Gingeras, T.R.,
Margulies, E.H., Weng, Z., Snyder, M., Dermitzakis, E.T., Stamatoyannopoulos, J.A., et al. (2007). Identication and analysis of functional elements in
1% of the human genome by the ENCODE pilot project. Nature 447, 799
816. http://dx.doi.org/10.1038/nature05874.
Borneman, A.R., Gianoulis, T.A., Zhang, Z.D., Yu, H., Rozowsky, J., Seringhaus, M.R., Wang, L.Y., Gerstein, M., and Snyder, M. (2007). Divergence of
transcription factor binding sites across related yeast species. Science 317,
815819. http://dx.doi.org/10.1126/science.1140748.
Boyden, L.M., Lewis, J.M., Barbee, S.D., Bas, A., Girardi, M., Hayday, A.C., Tigelaar, R.E., and Lifton, R.P. (2008). Skint1, the prototype of a newly identied
immunoglobulin superfamily gene cluster, positively selects epidermal gammadelta T cells. Nat. Genet. 40, 656662. http://dx.doi.org/10.1038/ng.108.
Dermitzakis, E.T., and Clark, A.G. (2001). Differential selection after duplication
in mammalian developmental genes. Mol. Biol. Evol. 18, 557562.
Dixon, J.R., Selvaraj, S., Yue, F., Kim, A., Li, Y., Shen, Y., Hu, M., Liu, J.S., and
Ren, B. (2012). Topological domains in Mamm. Genomes identied by analysis
of chromatin interactions. Nature 485, 376380. http://dx.doi.org/10.1038/nature11082.
Dong, X., Han, S., Zylka, M.J., Simon, M.I., and Anderson, D.J. (2001). A
diverse family of GPCRs expressed in specic subsets of nociceptive sensory
neurons. Cell 106, 619632.
Dowen, J.M., Fan, Z.P., Hnisz, D., Ren, G., Abraham, B.J., Zhang, L.N., Weintraub, A.S., Schuijers, J., Lee, T.I., Zhao, K., and Young, R.A. (2014). Control of
cell identity genes occurs in insulated neighborhoods in mammalian chromosomes. Cell 159, 374387. http://dx.doi.org/10.1016/j.cell.2014.09.030.
Faure, A.J., Schmidt, D., Watt, S., Schwalie, P.C., Wilson, M.D., Xu, H., Ramsay, R.G., Odom, D.T., and Flicek, P. (2012). Cohesin regulates tissue-specic
expression by stabilizing highly occupied cis-regulatory modules. Genome
Res. 22, 21632175. http://dx.doi.org/10.1101/gr.136507.111.
Filippova, G.N., Fagerlie, S., Klenova, E.M., Myers, C., Dehner, Y., Goodwin,
G., Neiman, P.E., Collins, S.J., and Lobanenkov, V.V. (1996). An exceptionally
conserved transcriptional repressor, CTCF, employs different combinations of
zinc ngers to bind diverged promoter sequences of avian and mammalian cmyc oncogenes. Mol. Cell. Biol. 16, 28022813.
Guacci, V., Koshland, D., and Strunnikov, A. (1997). A direct link between sister
chromatid cohesion and chromosome condensation revealed through the
analysis of MCD1 in S. cerevisiae. Cell 91, 4757.
Hadjur, S., Williams, L.M., Ryan, N.K., Cobb, B.S., Sexton, T., Fraser, P.,
Fisher, A.G., and Merkenschlager, M. (2009). Cohesins form chromosomal
cis-interactions at the developmentally regulated IFNG locus. Nature 460,
410413. http://dx.doi.org/10.1038/nature08079.
Hou, C., Li, L., Qin, Z.S., and Corces, V.G. (2012). Gene density, transcription,
and insulators contribute to the partition of the Drosophila genome into

physical domains. Mol. Cell 48, 471484. http://dx.doi.org/10.1016/j.molcel.

2012.08.031.

multiple mammalian lineages. Cell 148, 335348. http://dx.doi.org/10.1016/j.

cell.2011.11.058.

Kim, T.H., Abdullaev, Z.K., Smith, A.D., Ching, K.A., Loukinov, D.I., Green,
R.D., Zhang, M.Q., Lobanenkov, V.V., and Ren, B. (2007). Analysis of the vertebrate insulator protein CTCF-binding sites in the human genome. Cell 128,
12311245. http://dx.doi.org/10.1016/j.cell.2006.12.048.

Seitan, V.C., Hao, B., Tachibana-Konwalski, K., Lavagnolli, T., Mira-Bontenbal, H., Brown, K.E., Teng, G., Carroll, T., Terry, A., Horan, K., et al. (2011). A
role for cohesin in T-cell-receptor rearrangement and thymocyte differentiation. Nature 476, 467471. http://dx.doi.org/10.1038/nature10312.

Langmead, B., Trapnell, C., Pop, M., and Salzberg, S.L. (2009). Ultrafast and
memory-efcient alignment of short DNA sequences to the human genome.
Genome Biol. 10, R25. http://dx.doi.org/10.1186/gb-2009-10-3-r25.
Michaelis, C., Ciosk, R., and Nasmyth, K. (1997). Cohesins: chromosomal proteins that prevent premature separation of sister chromatids. Cell 91, 3545.

Seitan, V.C., Faure, A.J., Zhan, Y., McCord, R.P., Lajoie, B.R., Ing-Simmons,
E., Lenhard, B., Giorgetti, L., Heard, E., Fisher, A.G., et al. (2013). Cohesinbased chromatin interactions enable regulated gene expression within preexisting architectural compartments. Genome Res. 23, 20662077. http://dx.doi.
org/10.1101/gr.161620.113.

Mishiro, T., Ishihara, K., Hino, S., Tsutsumi, S., Aburatani, H., Shirahige, K., Kinoshita, Y., and Nakao, M. (2009). Architectural roles of multiple chromatin insulators at the human apolipoprotein gene cluster. EMBO J. 28, 12341245.
http://dx.doi.org/10.1038/emboj.2009.81.

Sexton, T., Yaffe, E., Kenigsberg, E., Bantignies, F., Leblanc, B., Hoichman,
M., Parrinello, H., Tanay, A., and Cavalli, G. (2012). Three-dimensional folding
and functional organization principles of the Drosophila genome. Cell 148,
458472. http://dx.doi.org/10.1016/j.cell.2012.01.010.

Nativio, R., Wendt, K.S., Ito, Y., Huddleston, J.E., Uribe-Lewis, S., Woodne,
K., Krueger, C., Reik, W., Peters, J.-M., and Murrell, A. (2009). Cohesin is
required for higher-order chromatin conformation at the imprinted IGF2-H19
locus. PLoS Genet. 5, e1000739. http://dx.doi.org/10.1371/journal.pgen.
1000739.

Shen, Y., Yue, F., McCleary, D.F., Ye, Z., Edsall, L., Kuan, S., Wagner, U.,
Dixon, J., Lee, L., Lobanenkov, V.V., and Ren, B. (2012). A map of the cis-regulatory sequences in the mouse genome. Nature 488, 116120. http://dx.doi.
org/10.1038/nature11243.

Nora, E.P., Lajoie, B.R., Schulz, E.G., Giorgetti, L., Okamoto, I., Servant, N.,
Piolot, T., van Berkum, N.L., Meisig, J., Sedat, J., et al. (2012). Spatial partitioning of the regulatory landscape of the X-inactivation centre. Nature 485, 381
385. http://dx.doi.org/10.1038/nature11049.
Parelho, V., Hadjur, S., Spivakov, M., Leleu, M., Sauer, S., Gregson, H.C., Jarmuz, A., Canzonetta, C., Webster, Z., Nesterova, T., et al. (2008). Cohesins
functionally associate with CTCF on mammalian chromosome arms. Cell
132, 422433. http://dx.doi.org/10.1016/j.cell.2008.01.011.
Pauli, A., Althoff, F., Oliveira, R.A., Heidmann, S., Schuldiner, O., Lehner, C.F.,
Dickson, B.J., and Nasmyth, K. (2008). Cell-type-specic TEV protease cleavage reveals cohesin functions in Drosophila neurons. Dev. Cell 14, 239251.
http://dx.doi.org/10.1016/j.devcel.2007.12.009.
Phillips-Cremins, J.E., Sauria, M.E.G., Sanyal, A., Gerasimova, T.I., Lajoie,
B.R., Bell, J.S.K., Ong, C.-T., Hookway, T.A., Guo, C., Sun, Y., et al. (2013).
Architectural protein subclasses shape 3D organization of genomes during
lineage commitment. Cell 153, 12811295. http://dx.doi.org/10.1016/j.cell.
2013.04.053.
Rollins, R.A., Morcillo, P., and Dorsett, D. (1999). Nipped-B, a Drosophila homologue of chromosomal adherins, participates in activation by remote enhancers in the cut and Ultrabithorax genes. Genetics 152, 577593.
Schmidt, D., Wilson, M.D., Ballester, B., Schwalie, P.C., Brown, G.D.,
Marshall, A., Kutter, C., Watt, S., Martinez-Jimenez, C.P., Mackay, S., et al.
(2010). Five-vertebrate ChIP-seq reveals the evolutionary dynamics of transcription factor binding. Science 328, 10361040. http://dx.doi.org/10.1126/
science.1186176.
Schmidt, D., Schwalie, P.C., Wilson, M.D., Ballester, B., Goncalves, A., Kutter,
C., Brown, G.D., Marshall, A., Flicek, P., and Odom, D.T. (2012). Waves of retrotransposon expansion remodel genome organization and CTCF binding in

Sofueva, S., Yaffe, E., Chan, W.-C., Georgopoulou, D., Vietri Rudan, M., MiraBontenbal, H., Pollard, S.M., Schroth, G.P., Tanay, A., and Hadjur, S. (2013).
Cohesin-mediated interactions organize chromosomal domain architecture.
EMBO J. 32, 31193129. http://dx.doi.org/10.1038/emboj.2013.237.
van de Werken, H.J.G., Landan, G., Holwerda, S.J.B., Hoichman, M., Klous,
P., Chachik, R., Splinter, E., Valdes-Quezada, C., Oz, Y., Bouwman, B.A.M.,
et al. (2012). Robust 4C-seq data analysis to screen for regulatory DNA interactions. Nat. Methods 9, 969972. http://dx.doi.org/10.1038/nmeth.2173.
Wendt, K.S., Yoshida, K., Itoh, T., Bando, M., Koch, B., Schirghuber, E., Tsutsumi, S., Nagae, G., Ishihara, K., Mishiro, T., et al. (2008). Cohesin mediates
transcriptional insulation by CCCTC-binding factor. Nature 451, 796801.
http://dx.doi.org/10.1038/nature06634.
Xiao, T., Wallace, J., and Felsenfeld, G. (2011). Specic sites in the C terminus
of CTCF interact with the SA2 subunit of the cohesin complex and are required
for cohesin-dependent insulation activity. Mol. Cell. Biol. 31, 21742183.
http://dx.doi.org/10.1128/MCB.05093-11.
Yaffe, E., and Tanay, A. (2011). Probabilistic modeling of Hi-C contact maps
eliminates systematic biases to characterize global chromosomal architecture. Nat. Genet. 43, 10591065. http://dx.doi.org/10.1038/ng.947.
Yaffe, E., Farkash-Amar, S., Polten, A., Yakhini, Z., Tanay, A., and Simon, I.
(2010). Comparative analysis of DNA replication timing reveals conserved
large-scale chromosomal architecture. PLoS Genet. 6, e1001011. http://dx.
doi.org/10.1371/journal.pgen.1001011.s016.
Zuin, J., Dixon, J.R., van der Reijden, M.I.J.A., Ye, Z., Kolovos, P., Brouwer,
R.W.W., van de Corput, M.P.C., van de Werken, H.J.G., Knoch, T.A., van
IJcken, W.F.J., et al. (2014). Cohesin and CTCF differentially affect chromatin
architecture and gene expression in human cells. Proc. Natl. Acad. Sci. USA
111, 9961001. http://dx.doi.org/10.1073/pnas.1317788111.

Cell Reports 10, 12971309, March 3, 2015 2015 The Authors 1309

Cell Reports

Article
Heterogeneities in Nanog Expression
Drive Stable Commitment to Pluripotency
in the Mouse Blastocyst
Panagiotis Xenopoulos,1,5 Minjung Kang,1,2,5 Alberto Puliato,3 Stefano Di Talia,4 and Anna-Katerina Hadjantonakis1,*
1Developmental

Biology Program, Sloan Kettering Institute, New York, NY 10065, USA

Cell and Molecular Biology Program, Weill Graduate School of Medical Sciences of Cornell University, New York,
NY 10065, USA
3Laboratory of Cell Migration, Candiolo Cancer Institute - FPO, IRCCS, Candiolo, Torino 10060, Italy
4Department of Cell Biology, Duke University Medical Center, Durham, NC 27710, USA
5Co-rst author
*Correspondence: [email protected]
http://dx.doi.org/10.1016/j.celrep.2015.02.010
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/).
2Biochemistry,

SUMMARY

The pluripotent epiblast (EPI) is the founder tissue of

almost all somatic cells. EPI and primitive endoderm
(PrE) progenitors arise from the inner cell mass (ICM)
of the blastocyst-stage embryo. The EPI lineage
is distinctly identied by its expression of pluripotency-associated factors. Many of these factors
have been reported to exhibit dynamic uctuations
of expression in embryonic stem cell cultures.
Whether these uctuations correlating with ICM
fate choice occur in vivo remains an open question.
Using single-cell resolution quantitative imaging of
a Nanog transcriptional reporter, we noted an irreversible commitment to EPI/PrE lineages in vivo. A
period of apoptosis occurred concomitantly with
ICM cell-fate choice, followed by a burst of EPI-specic cell proliferation. Transitions were occasionally
observed from PrE-to-EPI, but not vice versa, suggesting that they might be regulated and not stochastic. We propose that the rapid timescale of early
mammalian embryonic development prevents uctuations in cell fate.
INTRODUCTION
Pluripotency is dened as the ability of a cell to differentiate and
give rise to all somatic and germ cells (Nichols and Smith, 2012).
Although pluripotency can be induced in differentiated cells
(Takahashi and Yamanaka, 2006), how a pluripotent population
emerges in its native context, within the early mammalian embryo, remains an open question. Insight will come from elucidating the dynamic cell behaviors and molecular mechanisms
underlying the development of the mammalian blastocyst, the
embryonic stage at which a bona de pluripotent population
the epiblast (EPI)is established.
The EPI is molecularly distinct and spatially segregated
from the two extra-embryonic lineages, the primitive endoderm

(PrE) and trophectoderm (TE) of the mouse blastocyst. The

specication of these lineages occurs as two sequential binary
cell-fate decisions. The rst involves specication and segregation of TE from inner cell mass (ICM), while the second occurs
within the ICM and involves the specication of EPI and PrE
precursors, and their eventual segregation into adjacent tissue
layers (reviewed in Schrode et al., 2013). By late blastocyst
stage, the EPI and PrE lineages are dened both by their position
within the embryo and expression of lineage-specic transcription factors, such as NANOG in the EPI, and GATA6 and
GATA4 in the PrE (Xenopoulos et al., 2012). Recent studies
have illustrated that EPI/PrE allocation occurs in at least three
successive steps (Chazaud et al., 2006; Frankenberg et al.,
2011; Plusa et al., 2008). Initially, lineage-specic transcription
factors, such as NANOG and GATA6, are co-expressed by all
ICM cells, suggesting a multi-lineage priming state. Thereafter,
NANOG and PrE lineage-specic transcription factors exhibit
mutually exclusive expression, as lineage progenitors emerge
in a salt-and-pepper distribution within the ICM. At this stage,
GATA4 becomes activated in PrE progenitors, concomitant
with NANOG downregulation. Finally, lineage segregation is
achieved with the localization of PrE cells to the surface of the
ICM. At this time, other pluripotency-associated factors become
restricted to EPI cells, which have become positioned internally
within the ICM. Notably, NANOG is one of the rst markers to
be restricted within the EPI, whereas OCT4 and SOX2 become
subsequently downregulated in PrE progenitors and restricted
to EPI progenitors.
The initial specication of EPI and PrE progenitors appears to
occur in a spatially random manner (Schrode et al., 2014) and
could be achieved if a stochastic process were to underlie this
second fate decision. Indeed, an analysis of transcriptomes of
single ICM cells revealed that gene expression is highly heterogeneous at earlier stages, exhibiting no apparent lineage specicity and a hierarchical relationship of marker expression only
appearing in the late blastocyst (Guo et al., 2010; Kurimoto
et al., 2006; Ohnishi et al., 2014).
A degree of heterogeneity has been observed at both protein
and mRNA level for various pluripotency-associated factors in
embryonic stem cell (ESC) cultures. Many studies have focused

1508 Cell Reports 10, 15081520, March 10, 2015 2015 The Authors

on Nanog, a central component of the core pluripotency

transcriptional network (Chambers et al., 2007; Kalmar et al.,
2009). Experiments in ESCs have suggested that Nanog expression displays dynamic uctuations that may correlate with a
cells fate choice between self-renewal and differentiation. However, it is unclear whether uctuations in gene expression take
place in vivo in embryos where cell differentiation occurs on a
shorter timescale, nor whether they predict fate choice or fate
reversion. Notably, understanding how pluripotent cells behave
in embryos may provide information that can be reconciled
with observations made in ESCs (Smith, 2013).
To determine how the EPI emerges within the mouse
blastocyst, we generated a reporter of Nanog transcription
(Nanog:H2B-GFP). Derivation of ESCs from reporter-expressing
embryos revealed heterogeneous gene expression as an adaptation to ESC propagation. Using live imaging, we quantied the dynamics of Nanog expression in individual cells of live blastocysts,
establishing how Nanog expression inuences the fate of ICM
cells. By contrast to ESCs maintained in culture, uctuations in
Nanog expression between distinct developmental states did
not, generally, occur in vivo. However, we noted rare cases of
cells unidirectionally switching their fate, from a PrE to EPI identity. Since ICM fate changes were only observed toward the EPI,
and not toward PrE, we concluded that this change was not stochastic. Our analyses also revealed events of selective apoptosis
at the onset of ICM lineage differentiation, followed by a burst of
cell proliferation in EPI-committed cells. Collectively, these data
suggest that, although it may be dynamic in ESCs, the emergence of a pluripotent identity is sequential and linear in vivo
and not accommodating reversibility in fate. In this way, after its
allocation, the pluripotent cell population might be protected,
thereby ensuring the development of somatic lineages.
RESULTS
BAC-Based Nanog Transcriptional Reporters Mark the
Pluripotent State in ESCs and Embryos
To probe the dynamics of the pluripotent state, we developed a
BAC-based Nanog transgenic transcriptional reporter, based on
a previous design used as a readout of cellular reprogramming
during iPS cell generation (Okita et al., 2007) (Figure S1H). For
single-cell resolution readouts of Nanog expression, we generated nuclear-localized human histone H2B fusion versions of
the reporter (Figures 1A, 1C, and S1E).
To validate transgene activity, we analyzed reporter expression in transgenic ESCs under various culture conditions. These
conditions included the presence or absence of leukemia
inhibitory factor (LIF), and 2i+LIF, which promote the self-renewal
of ESCs, induce differentiation, or ground state pluripotency,
respectively (Ying et al., 2008). Immunostaining of ESCs in
2i+LIF or serum-LIF conditions revealed markedly increased or
decreased expression, respectively, of both reporter and
NANOG protein. Heterogeneous but correlated GFP and NANOG
expression was observed in Nanog:GFPTg/+ and Nanog:H2BGFPTg/+ ESCs maintained in serum+LIF conditions (Figures 1A,
1B, S1D, and S1E). Single-cell quantitative image analyses of
immunostained Nanog:H2B-GFPTg/+ ESCs maintained under
different conditions further validated reporter efcacy (Figures

1B, S1E, and S1F). Moreover, we observed an increased correlation of reporter activity with NANOG protein for the Nanog:H2BGFP transgene, compared to the targeted Nanog transcriptional
reporter in the heterozygous TNGA ESCs (Chambers et al., 2007)
(Figures S1E and S1F). These data suggest that the BAC transgenic reporters we constructed faithfully marked the pluripotent
state in ESC cultures and could be used to probe Nanog expression dynamics at single-cell resolution.
Next, we generated Nanog:H2B-GFP transgenic mice for
single-cell resolution quantitative visualization of the EPI lineage
in vivo. We used live imaging to analyze the distribution and
validate the reporter in embryos (Figure 1C). We rst observed
reporter activity in embryos at the 816 cell stage (Plusa et al.,
2008). Cells displaying high and low GFP levels were rst
observed at the mid blastocyst stage (70100 cells), as a saltand-pepper distribution of EPI and PrE precursors was established within the ICM (Chazaud et al., 2006; Plusa et al., 2008).
In late blastocysts (>100 cells), where the EPI and PrE lineages
have sorted into distinct layers, cells within the ICM forming
the EPI exhibited signicantly elevated levels of GFP compared
to PrE cells located on the surface of the ICM. Thereafter, GFP
was detected within the EPI, albeit at reduced levels in implanting E4.5 and post-implantation E5.5 embryos consistent with the
downregulation of NANOG observed at periimplantation (Chambers et al., 2003). Notably, the reporter was strongly expressed in
TE cells of early and mid blastocysts (corresponding to 3290
cell stage), possibly resulting from robust NANOG localization in
ICM and TE cells at these stages (Figure S1I) (Dietrich and Hiiragi, 2007; Messerschmidt and Kemler, 2010; Morgani et al.,
2013). However, GFP expression in the TE was signicantly
reduced in late (>100 cells) and in implanting blastocysts
(Figures 1C and S1G). TE localization was also noted for other
Nanog-based reporters analyzed at comparable embryonic
stages (Figure S1G). Collectively, these data lead us to conclude
that the Nanog:H2B-GFP reporter faithfully marks the pluripotent
state both in vitro in ESC cultures and in vivo in the emerging EPI
lineage of the mouse blastocyst.
Derivation of ESCs from Mouse Blastocysts Results in
Highly Variable Pluripotency-Associated Gene
Expression
The EPI lineage of the blastocyst represents the in vivo counterpart to ESC cultures propagated in vitro (Boroviak et al., 2014).
We therefore sought to investigate the prole of Nanog:H2BGFP reporter activity in transgenic embryo-derived ESCs. We
used an ESC derivation protocol in which an ICM outgrowth
emerges in conditions promoting ground state pluripotency in
serum-free medium containing 2i (Czechanski et al., 2014).
Indeed, under these conditions GFP was strongly expressed
by all cells of outgrowths from Nanog:H2B-GFPTg/+ blastocysts
(Figure 1D). However, both reporter and NANOG expression
became heterogeneous when the derived ESCs were propagated under standard serum+LIF conditions, either in the presence or absence of mouse embryonic feeders (MEFs). Even
in the presence of MEFs, GFP-low/NANOG-low cells were
observed within ESC colonies (Figure 1E). Fluorescence-activated cell sorting (FACS) and quantitative immunouorescence
analyses conrmed the presence of a highly heterogeneous

Cell Reports 10, 15081520, March 10, 2015 2015 The Authors 1509

Figure 1. BAC-Based Nanog Transcriptional Reporters Faithfully Mark the Pluripotent State in ESCs and Embryos
(A) Immunouorescence images of Nanog:GFPTg/+ and Nanog:H2B-GFPTg/+ ESCs grown in MEF-free conditions including serum+LIF, 2i+LIF, and serum-LIF for
three passages. Orange arrowheads identify GFP-low and NANOG-low cells within ESC colonies.
(B) Quantitative immunouorescence analysis after nuclear segmentation of Nanog:H2B-GFPTg/+ ESCs. GFP (y axis) and NANOG (x axis) uorescence values
plotted for individual ESCs propagated for ve passages in MEF-free serum+LIF conditions then grown for 4 days in various culture conditions.
(C) Reporter expression in live embryos stained with membrane marker FM4-64. GFP-hi cells, white arrowheads; GFP-low cells, orange arrowheads; TE cells
expressing GFP, yellow arrowheads. Dashed line depicts boundary between epiblast (EPI) and visceral endoderm (VE) layers of an E5.5 embryo. Cell number was
determined by staining with Hoechst.
(D) Schematic of ESC derivation. After 20 days, Nanog:H2B-GFPTg/+ ESCs were established in the presence of MEF feeders in serum+LIF conditions and then
propagated in the presence or absence of MEFs.
(E) Immunostaining and analysis of derived Nanog:H2B-GFPTg/+ ESCs. Orange arrowheads mark GFP-low/NANOG-low/OCT4+ cells.
(F) FACS analysis of derived Nanog:H2B-GFPTg/+ ESCs grown in serum+LIF conditions in the presence or absence of MEFs. Numbers in FACS histograms
indicate percentage of GFP (left) and GFP+ (right) populations. r = Pearson correlation coefcient. Scale bar represents 20 mm.

population, where both reporter and protein expression where

highly correlated (r = 0.85) (Figures 1E and 1F). From these observations, we conclude that gene expression heterogeneities
of the pluripotent state likely arise in ESCs as a result of their
in vitro propagation.

Quantitative Single-Cell Analysis of Nanog:H2B-GFP

Expression in the Emerging Pluripotent Cells of the
Blastocyst
To determine whether the transgenic reporter allowed a quantitative evaluation of Nanog expression in vivo, we analyzed the

1510 Cell Reports 10, 15081520, March 10, 2015 2015 The Authors

Figure
2. Differential
Expression
of
Nanog:H2B-GFP Reporter in Segregating
EPI and PrE Cells
(A) MINS image analysis pipeline (see Experimental Procedures). GFP and NANOG expression
plotted for individual cells after automated nuclear
segmentation and uorescence intensity normalization (best-t regression curve).
(B) Immunouorescence images of xed Nanog:
H2B-GFPTg/+ embryos. EPI cells, white arrowheads; PrE cells, orange arrowheads; TE cells expressing GFP, yellow arrowheads.
(C) Quantitative immunouorescence analyses.
(D) Reporter, GATA6, and NANOG expression
in Nanogb-geo/b-geo blastocysts carrying the
Nanog:H2B-GFP reporter. NANOG staining was
absent in Nanog mutant transgenic embryos.
r = Pearson correlation coefcient. Scale bar represents 20 mm.

pattern of reporter expression in transgenic embryos that had

been xed and stained for lineage-specic markers such as
GATA6/PrE and NANOG/EPI using single-cell quantitative image

analysis (Lou et al., 2014) (Figure 2A).

In morulae (8-16 cells) and early blastocyst (3264 cells) stages, GFP was
observed throughout the embryo, reecting a double (GATA6+ NANOG+) -positive
state. The differential levels of GFP
expression were evident in mid blastocysts (70100 cells), as embryos established a mutually exclusive distribution of
GATA6+ PrE and NANOG+ EPI progenitors (Figures 2B and S2A). In late blastocysts (>100 cells), where the EPI/PrE
sorting had occurred, reporter expression
was markedly elevated in NANOG+
EPI progenitor cells and diminished in
GATA6+ PrE progenitor cells (Figure 2B).
Quantitative uorescence analysis of
immunostained
Nanog:H2B-GFPTg/+
blastocysts at different stages conrmed
a strong correlation between GFP and
NANOG levels in ICM cells, as seen in
ESCs, thus indicating that the reporter
could be used to quantitatively infer levels
of expression of Nanog (Figures 2C and
S2B). Notably, reporter expression in TE
cells was prominent in early and mid blastocysts but displayed reduced correlation
with NANOG protein compared to the
ICM and hence was not investigated
further (Figures 2C and S2B). The H2BGFP reporter responds to changes in
Nanog expression with a characteristic
time that is determined by the stability of
the H2B-GFP fusion protein. Our control
experiments and computational analysis
indicate that we can measure changes in Nanog expression
higher than 2-fold with about 1 hr resolution (see the Supplemental Experimental Procedures). Such temporal resolution is

Cell Reports 10, 15081520, March 10, 2015 2015 The Authors 1511

Figure 3. Nanog Expression in the ICM Is

Altered by Modulation of FGF Signaling
(A) Immunouorescence images of late blastocyst
stage Nanog:H2B-GFPTg/+ embryo. Fluorescence
values for GFP and NANOG plotted for individual
cells. Values for GFP and NANOG of ICM cells
also plotted as frequency distributions obtained
by binning uorescence values in 20 logarithmically spaced categories, as described previously
(Munoz Descalzo et al., 2012).
(B) Regimen used for exogenous FGF and ERK1/2
inhibitor (ERKi) treatment experiments. Nanog:
H2B-GFPTg/+ blastocysts recovered at E2.75 and
cultured for 36 hr in KSOM medium (I), KSOM +
FGF2 (II), and ERKi (III), followed by staining for
Hoechst, NANOG, and SOX17. White arrowheads
identify NANOG+;SOX17 ICM cells exhibiting
high levels of GFP. Orange arrowheads identify
SOX17+;NANOG;GFP-low ICM cells.
(C) Reporter and NANOG distribution analysis in
cells of cultured embryos shown in (B).
Note that x and y axis scales for scatterplots vary
due to changes in reporter and NANOG expression
between embryos cultured under different conditions. r = Pearson correlation coefcient. Scale bar
represents 20 mm.

cells emerging within the ICM and provides an accurate quantitative readout of
Nanog expression.

much smaller than the timescales of cell differentiation. Finally,

by analyzing Nanog mutant embryos carrying the reporter, we
conrmed that no functional NANOG protein was produced
from the BAC transgene (Figure 2D). We therefore conclude
that the Nanog:H2B-GFP reporter faithfully marks pluripotent

Distribution of NANOG Expression

in the Blastocyst Is Altered by
Modulation of Fibroblast Growth
Factor Signaling
Next, we investigated the distribution of
the Nanog:H2B-GFP transcriptional reporter, as well as NANOG protein, as
the EPI compartment emerges within the
ICM of late blastocysts. We performed
quantitative immunouorescence analysis on mid-to-late blastocysts (90110
cells), stained with Hoechst, NANOG,
and the PrE marker SOX17 (Figure 3A).
We noted a bimodal distribution of reporter expression within the ICM, which
corresponded to bimodality in NANOG
distribution within prospective EPI and
PrE cells (Figures 3A and S3AS3D).
We investigated how these distributions could be modulated by perturbation
of broblast growth factor (FGF) signaling,
which plays a critical role in ICM lineage
choice. Pathway inhibition results in an
ICM composed exclusively of EPI precursors, while incubation
of embryos in exogenous FGF results in an all-PrE ICM
(Chazaud et al., 2006; Kang et al., 2013; Nichols et al., 2009; Yamanaka et al., 2010). Consistent with these studies, we observed
that the bimodal distribution of GFP and NANOG levels within the

1512 Cell Reports 10, 15081520, March 10, 2015 2015 The Authors

Figure 4. EPI Cells Emerge In Vivo with an Accompanying Increase in Nanog Expression
(A) Schematic of mouse blastocyst stage embryo development.
(B) Quantication of GFP intensity in individual nuclei of a living Nanog:H2B-GFPTg/+ blastocyst recovered at E3.5 (65 cells) and corresponding images of single
time points. Single-cell intensity traces were obtained by tracking single cells in time-lapse experiments. Mitotic and apoptotic cells were also tracked, resulting in
abrupt peaks in uorescence intensity. Cells belonging to the ICM are depicted in purple, while TE cells are depicted in green; cells having undergone apoptosis
are depicted as black lines. The developmental timing and the time spanned by tracks analyzed in this panel are illustrated by an arrow in (A).
(C) Top panel, quantication of relative GFP intensity in daughters/sisters versus mother cells upon division in cells of TE (green bars) or ICM (purple bars) origin.
Daughter cells retained expression levels after mitosis, with most changes within 20%.
Data were obtained from the analysis of time-lapse movies of ve or more embryos. Bottom panel, cell divisions of EPI progenitors occurring at late blastocyst
stages shown in (B). Red arrowheads identify apoptotic events, and yellow arrowheads mark cell divisions. Scale bar represents 20 mm.

ICM was lost when FGF signaling activity was modulated

(Figures 3B, 3C, and S3ES3J). We therefore conclude that a
heterogeneous distribution of NANOG expression exists in vivo.
This is represented as a stable bimodal distribution within the
ICM and reects cell fate specication toward PrE and EPI lineages, and that this distribution is altered by modulation of FGF
signaling.
Highly Variable Nanog Dynamics Are Initially Observed
Followed by the Establishment of Differential
Expression at the Onset of ICM Lineage Specication
A bimodal distribution of Nanog expression is established as the
pluripotent EPI emerges within the ICM. However, we currently
do not know whether uctuating expression of pluripotency-associated factors, such as Nanog, occurs during EPI specication,
nor, if they do exist, whether such uctuations correlate or predict
state reversions (Smith, 2013). To address this question, we
investigated the behavior of individual ICM cells in Nanog:H2BGFPTg/+ blastocysts using time-lapse imaging coupled with quantitative image analyses performed in an accurate and automated
fashion (Supplemental Experimental Procedures).

Using this methodology, we rst observed that in early blastocysts (until 60 cell stage) all cells (ICM and TE) displayed highly
heterogeneous reporter activity (0400 min in Figure 4B), consistent with previous observations (Dietrich and Hiiragi, 2007;
Ohnishi et al., 2014; Plusa et al., 2008). Following this phase,
TE cells downregulated reporter expression, whereas ICM cells
retained or increased their expression, presumably coinciding
with the establishment of a lineage bias (400 min to end of movie;
Figure 4B). We also noted apoptotic events, occurring during the
process of lineage specication in randomly positioned GFP-hi
or GFP-low ICM cells (red arrowheads; Figure 4B). As described
previously, apoptotic events occur during blastocyst development and are not a consequence of in vitro culture or phototoxicity (Artus et al., 2013; Plusa et al., 2008). Furthermore, we noted
several cell divisions occurring in GFP-hi ICM cells (yellow arrowheads; Figure 4B). Heritability of Nanog:H2B-GFP levels, and
likely cell fate, was observed in all ICM cells after division (Figure 4C). Of note, an increase in GFP uorescence was routinely
observed during mitosis, due to chromosome condensation;
thus any resulting rapid oscillations (of the order of 150 min) in reporter activity were excluded from the analysis, as they were not

Cell Reports 10, 15081520, March 10, 2015 2015 The Authors 1513

considered as reecting changes in gene expression. By using a

computational approach, we conrmed that GFP expression
was not diluted in dividing cells, suggesting that daughter cells
inherited the lineage identity of their parental cell (Figure S4).
Finally, cells within the TE displayed low, somewhat variable
and continually decreasing levels of reporter activity (Figure 4B).
Next, we focused our analyses on subsequent phases of
development, initiating from early-to-mid blastocyst stages (at
around 5060 cell) until late blastocyst stages (at around 100 cells),
where the EPI and PrE cells have sorted to their nal positions
(Figures 5A and 5B; Movie S1). A subpopulation of ICM cells
is specied to the EPI-lineage, with GFP expression maintained
or increased. By contrast, cells biased toward the PrE lineage
extinguished the reporter during differentiation (Figure 5A).
Infrequent Cell-State Reversals Occur toward, Not Away
from, a Pluripotent Identity
Notably, we observed a few instances where cells in mid-stage
blastocysts, exhibiting low levels of reporter activity, would
rapidly increase reporter activity, and concomitantly become
segregated with the EPI (Figure 5A). These could, in principle,
represent rare PrE-to-EPI conversions. Importantly, we never
observed events of EPI-to-PrE progenitor switching associated
with downregulation of Nanog expression. To obtain an independent conrmation of the absence of EPI-to-PrE transitions, we
analyzed time-lapse data from a PrE-specic single-cell resolution reporter (PdgfraH2B-GFP/+, Plusa et al., 2008). In contrast to
Nanog:H2B-GFP, expression of the PdgfraH2B-GFP reporter is
activated only in PrE-biased cells after cell-fate specication.
Therefore, PrE-to-EPI conversion would be detected as downregulation of the reporter. On the other hand, EPI-to-PrE conversion would give rise to signicantly delayed activation of the
reporter in cells lacking expression. We imaged embryos
expressing the PdgfraH2B-GFP reporter starting at the time of
cell differentiation, i.e., the time when expression of the reporter
becomes reliably detectable. We found that all positive cells at
the onset of differentiation retained their expression and no
new cells initiated expression with a signicant delay, corresponding to an EPI-to-PrE conversion (Figure 5B). Collectively,
our results with the Nanog:H2B-GFP and PdgfraH2B-GFP/+ reporters imply that cell-fate reversals are extremely rare and preferentially happen from the PrE to the EPI state.
The PrE-to-EPI unidirectionality would suggest that fate transitions are regulated and do not result from purely stochastic
uctuations in gene expression. To rule out that the inability to
detect EPI-to-PrE transition in Nanog:H2B-GFP embryos was
due to perdurance of GFP reporter, we performed the following
analysis. First, we quantied the dynamics of downregulation of
Nanog:H2B-GFP in PrE cells (Figure S6) and estimated the lifetime of H2B-GFP to be shorter than 4 hr, which is short enough
to detect cell-fate reversal events (details in the Supplemental
Experimental Procedures). Second, we analyzed time-lapse
movies of the Nanog:H2B-GFP reporter in embryos treated
with exogenous FGF (all ICM cells would downregulate NANOG)
and showed that a clear downregulation can be observed (Figure S5). In addition, we calculated the half-life of H2B-GFP
from time-lapse movies of Nanog:H2B-GFP in exogenous FGF
or cycloheximide (CHX). The decay rates of H2B-GFP in ICM

cells in the embryos under CHX or FGF treatments were around

6.5 and 5.5 hr, respectively (Figure S6). Collectively, these experiments allow us to conclude that the half-life of H2B-GFP is
shorter than 6 hr and thus not dissimilar to the half-life of
NANOG, which has been calculated to be approximately 4 hr
(Abranches et al., 2013). Altogether, these results show that
the Nanog:H2B-GFP reporter allows measuring Nanog transcriptional dynamics on timescales longer than 1 hr (see the
Supplemental Experimental Procedures).
At Late Blastocyst Stages, EPI and PrE Do Not Exhibit
Fluctuations
Next, we examined whether uctuations in Nanog:H2B-GFP,
and thus Nanog, levels were observed after ICM lineage specication. At these late blastocyst stages (>90100 cells), the active
sorting of EPI and PrE populations to adjacent tissue layers is
evident. Our analyses revealed that the EPI cells exhibited
increasing levels of reporter activity, whereas a decrease in
GFP levels was observed in cells forming the emergent PrE
epithelial layer (0 min to end of movie; Figure 5C; Movie S1). At
the end of the specication period, apoptotic events were
observed (red arrowhead; 0100 min; Figure 5C). Thereafter,
we did not observe any apoptosis. Instead, after ICM lineage
specication, several cell divisions were observed in EPI progenitors (yellow arrowheads; Figure 5C). Collectively, these observations suggest a wave of temporarily restricted apoptotic events,
occurring around the period of lineage specication, followed by
a burst of EPI lineage-specic cell proliferation. Nanog levels remained stable after ICM cell-fate divergence suggesting that
uctuations between EPI and PrE states do not occur. Based
on these data, we conclude that within the ICM the majority of
pluripotent EPI and extra-embryonic PrE progenitor cells do
not change their fate after specication. Furthermore, our data
lead us to propose that a burst of cell proliferation following
cell differentiation ensures EPI lineage-specic expansion.
Quantitative Analyses of Nanog:H2B-GFP Time-Lapse
Movies Revealed a Correlation between Cell Behaviors
and Cell-Fate Choice within the ICM
Time-lapse imaging in combination with cell tracking and quantitative analysis allowed us to determine whether fate reversals
correlate with the spatial position of individual cells within the
ICM. We simultaneously analyzed the position of a cell relative
to the blastocyst cavity, and its Nanog reporter expression levels
as a function of time. Analysis of the spatial distribution of cells
converting from PrE-to-EPI indicated that the cells migrated toward the inner region of the ICM (Figure 6A, blue cell). The cell
undergoing fate switching exhibited similar rate of GFP increase
as in cells of the embryo under conditions of ERK inhibition
(Figure 6B). This observation suggests that the fate switching
might be a result of PrE-biased cells that stop responding to
ERK signaling (due to their localization and fates of their neighbors) and, as a consequence, increase Nanog expression and
change fate. Consistently, under conditions of ERK inhibition,
all ICM cells committed to an EPI fate and displayed increasing
reporter activity, whereas GFP expression in TE cells remained
unaffected, consistent with our previous observations in xed
embryos (Figures 3B, 3C, and S3ES3J). Notably, we did not

1514 Cell Reports 10, 15081520, March 10, 2015 2015 The Authors

Figure 5. State Reversals Can Occur toward a Pluripotent Identity during Lineage Specication in Mid Blastocysts, but Cells Do Not Change
Fate in Late Embryos
(A) Quantication of GFP intensity in single nuclei of a living Nanog:H2B-GFPTg/+ at E3.5 (54 cells) and corresponding snapshots from a time-lapse movie. A
subset of tracks are detailed in the lower plot, corresponding to cells highlighted in the images in the panel on the right. Cell highlighted with an orange dot and red
outline represents a GFP-low cell that upregulated reporter expression and contributed to EPI.
(B) Quantication of GFP intensity in single nuclei of a living PdgfraH2B-GFP/+ at E3.5 (60 cells) and corresponding snapshot of uorescence channel overlapped
with bright-eld images from a time-lapse movie. Only PrE-biased cells express H2B-GFP.
(C) The same analysis applied in (A) was repeated in a later Nanog:H2B-GFPTg/+ embryo at E3.75 (85 cells). ICM cells segregated toward EPI (red) and PrE
(blue) lineages. Cells that underwent apoptosis depicted with black lines. Red arrowheads mark apoptotic events and yellow arrowheads mark cell divisions.
Tracked nuclei are highlighted by dots; the outline of each dot depicts the lineage choice; red outline depicts EPI progenitor cells; blue outline depicts PrE
progenitor cells. Scale bar represents 20 mm.

Cell Reports 10, 15081520, March 10, 2015 2015 The Authors 1515

Figure 6. Cell-State Reversals toward a Pluripotent Identity Are Associated with Changes in Position within ICM
Fluorescence levels of single ICM cells (A, left, and B, lower-left panels) and two-dimensional projections of representative space trajectories (A, central, and B,
lower-central panels) are plotted for the duration of time-lapse movies of Nanog:H2B-GFPTg/+ embryo (A) corresponding to embryo in Figure 5A and (B) in the
presence of ERKi. The distance of each cell from the barycenter of the embryo is plotted versus GFP intensity (A, right, and B, lower right).
(A) EPI cells, green and blue; PrE cell, red trajectories. Blue trajectory represents a cell that switched state acquiring an EPI identity, moving from an initial position
on the surface of ICM inward while increasing Nanog expression.
(B) Top, quantication of GFP uorescence intensities and corresponding snapshots. Bottom, all six trajectories depict the behavior of ICM cells that acquire
an EPI fate increasing reporter activity. All cells change their absolute position due to embryo growth, but there are no events of abrupt spatial change in
position of individual cells within the cohort. Lines represent cell trajectories. Displayed trajectories for mitotic cells were excluded. Unique colors identify
individual cells. The outline of the embryo is drawn for plotting relative position of nuclei. Shades of gray in the trajectory depict time: earlier (light) to later
(darker).

observe any apoptotic events during ERK inhibition, whereas a

burst of proliferation was evident in the all-EPI ICM (yellow arrowheads in Figure 6B and Movie S2). These ndings support

our previous observations, suggesting that apoptosis serves as

a selection mechanism within the ICM, while the proliferation
burst in EPI-committed cells ensures the rapid expansion of

1516 Cell Reports 10, 15081520, March 10, 2015 2015 The Authors

the pluripotent lineage after its specication. In addition, we

observed that there was no abrupt spatial position change of
EPI cells within the ICM of ERK inhibitor-treated embryos;
instead, cell motility was restrained, and only passive cell movement was observed as these embryos developed (Figure 6B).
DISCUSSION
Pluripotency-associated factors such as Nanog have been
reported to exhibit dynamic uctuations of expression in ESC
cultures. Whether uctuations in gene expression correlating
with cell-state transitions occur in vivo remains an open question. Here, we have investigated Nanog expression heterogeneities using single-cell resolution Nanog transcriptional reporters
coupled with 3D time-lapse imaging and high-resolution
automated quantitative image analyses. Our data suggest
that BAC-based Nanog:H2B-GFP transgenic reporters are expressed at physiological levels, exhibit minimal reporter perdurance, and serve as a faithful readout of Nanog expression both
in vitro in ESCs, and in vivo in mouse embryos.
For both ESCs rendered transgenic through introduction of
a Nanog:H2B-GFP construct, as well as transgenic embryoderived ESCs, heterogeneities in the levels of reporter, as well
as NANOG protein, were noted. A GFP-low cell population,
which also displayed low levels of NANOG protein, was evident
in cells that were propagated in MEF-free conditions, which
promote a differentiation bias; this population displayed OCT4
expression (Figure 1A, 1B, and S1E), suggesting that it might
consist of pluripotent cells that were actively differentiating
and/or primed for differentiation. Importantly, this variability,
and the presence of a GFP-low/NANOG-low population, in
ESC cultures have also been recently reported with another
Nanog transcriptional reporter (Abranches et al., 2013). These
observations therefore call for caution in the choice of culture
conditions for ESC propagation; hence, it was recently shown
that random monoallelic gene expression could occur stochastically as ESCs differentiated, resulting in the acquisition of
heterogeneities during the adaptation of cells to in vitro culture
(Eckersley-Maslin et al., 2014; Gendrel et al., 2014).
Furthermore, it has been suggested that reporters targeted
to the Nanog locus, which concomitantly ablate Nanog
activity, might exhibit behaviors resulting from Nanog heterozygosity (Faddah et al., 2013; Filipczyk et al., 2013). We noted
that GFP expression from the knockin/knockout TNGA
reporter was elevated compared to a BAC-based Nanog:
GFPTg/+ reporter (Figures S1AS1C). We therefore investigated
whether Nanog allele heterozygosity per se might result in
increased NANOG expression, consistent with its reported
auto-repressive activity (MacArthur et al., 2012; Navarro et al.,
2012). However, we failed to observe any noticeable difference
in reporter activity between Nanog:H2B-GFPTg/+ ESCs that
harbored a two (wild-type) or one functional Nanog alleles (Figures S1J and S1K). Thus, our data suggest that the elevated
GFP expression observed in the TNGA ESCs is not the result
of Nanog heterozygosity, and could result from allele design.
The development of a faithful EPI lineage-specic reporter
providing a single-cell resolution quantitative readout of Nanog
expression, coupled with high-resolution image data analyses

allowed us to address a central open question pertaining the

behavior of EPI cells in vivo. Critical for this type of analysis
was a single-cell resolution live imaging reporter that was sufciently bright for time-lapse image acquisition, but expressed
at physiological levels, so that reporter perdurance would not
mask downregulation in cells. Notably, our experience with destabilized uorescent protein reporters reveals reduced levels
of uorescence, not amenable to the image analyses methodologies used in this study.
Our data reveal a range of cell behaviors before and after
the pluripotent EPI population has been specied in vivo within
the ICM (Figure 7). Prior to EPI versus PrE specication has
occurred, apoptosis serves as a selective mechanism to ensure
proper segregation of lineage progenitors. Rare fate reversal
events likely occur whereby GFP-low/PrE progenitor cells
convert to a GFP-hi/EPI progenitor state as cells migrate inward
in the ICM. At this time, cell migration correlates with fate choice
and is linked to the presence of a heterogeneous population of
EPI and PrE progenitors. Progenitors could be sorting toward a
niche comprising cells with a similar lineage bias. By contrast,
in the presence of ERK inhibition, lineage choice is forced
toward one direction (all-EPI) resulting in a homogenous ICM
population and is accompanied by a lack of cell movement
(Figure 6B). Finally, after lineage specication has occurred within
the ICM, uctuations between EPI and PrE progenitors were not
observed. However, a burst of cell proliferation was observed
in EPI progenitors, as they sorted to the interior of the ICM.
Rarely cells changed their state toward, but never away from,
a pluripotent identity. By using a single-cell resolution reporter
for the PrE lineage, we previously showed that PrE progenitor
cells could downregulate reporter expression but could not
conrm the fate reversal to EPI, as cells could not be followed
after loss of the reporter (Plusa et al., 2008). Here, by using a
single-cell resolution reporter of the EPI lineage, we directly
visualized these transitions. Perhaps cells converting from
PrE-to-EPI might cease responding to an FGF signal, as in
the presence of an ERK-inhibitor, resulting in Nanog upregulation and establishment of pluripotency. These data agree with
recent studies suggesting that pulsatile FGF signaling induces
differential Nanog expression within ICM cells and drives
Nanog mRNA degradation for rapid post-transcriptional control
of pluripotency (Torres-Padilla and Chambers, 2014; Tan and
Elowitz, 2014).
Importantly, we observed no fate reversals between EPI and
PrE subsequent to their specication. This is in agreement with
a recent mathematical model accounting for the dynamics of
the regulatory network that controls ICM differentiation; simulations indicated that after specication cells would not change
identity, and thus EPI and PrE states are not interchangeable
(Bessonnard et al., 2014). In addition, we observed that after
specication a burst of cell proliferation in the pluripotent
compartment was evident. This might ensure adequate numbers
of EPI progenitors available for subsequent development.
Furthermore, our observations suggest that PrE progenitors
exhibit increased plasticity, compared to EPI progenitors,
consistent with recent studies suggesting that PrE progenitors
have a broader developmental potential than their EPI counterparts (Grabarek et al., 2012), and observations reporting that

Cell Reports 10, 15081520, March 10, 2015 2015 The Authors 1517

Figure 7. Cell Behaviors during the Emergence of the Pluripotent EPI Lineage
Schematic representation of embryo development from mid-to-late blastocyst. Prior to lineage specication (around 6480 cells): (1) apoptosis occurs randomly
in EPI or PrE progenitors as they segregate to their appropriate layers, (2) segregation is linked to spatial movements within the ICM, and (3) few GFP-low cells
might acquire a pluripotent identity and migrate inward the ICM. After lineage specication (>8090 cells): (1) cells do not uctuate between EPI and PrE states,
and (2) cell divisions occur in EPI-specied population while relocating to the interior of the ICM.

ESCs are primed toward endoderm co-express embryonic and

extra-embryonic markers (Morgani et al., 2013).
Our data suggest that once specied in wild-type embryos,
pluripotent cells do not, in general, change their fate. However,
a very limited number of state reversals may occur at this time,
but only toward a pluripotent identity. During development,
sufcient numbers of lineage progenitors must be generated,

and there is a time window during which cell-fate reversals can

occur. Moreover, mechanisms involving apoptotic events and
symmetric cell divisions may ensure that a pluripotent identity
is protected and maintained in vivo. These observations within
the embryonic environment appear contrasting with studies in
ESCs. A possible explanation for this apparent disparity could
be that ESCs in culture do not receive the appropriate inputs

1518 Cell Reports 10, 15081520, March 10, 2015 2015 The Authors

from a niche, namely, neighboring cells and extra-cellular components. Another could be due to the differential timescales.
ICM cells commit to PrE and EPI fates in less than 24 hr, and,
since development proceeds unidirectionally, even though
there may be non-differentiating (or symmetric) cell divisions,
neither cell population arising within the ICM self-renews. By
contrast, ESCs can be maintained indenitely in vitro under
conditions of self-renewal, a timescale that presents an unrestricted period for conversion between alternative states.

minimum in the distance between the cell centroid and the centroids of all cells
in the subsequent frame. If such minimum does not exist, the algorithm uses
cell segmentation to look for maximum overlap between cells at the subsequent time point and the cell being tracked. As additional criteria, changes
in nuclear shape and overall distance are required to lie within control ranges.
Such criteria generate unambiguous cell assignments and provide reliable
tracking of a large fraction of cells (60%70% in this study). The algorithm
was validated by visual inspection, and each track was obtained by matching
forward (prospective) and reverse (retrospective) tracking data (details provided in the Supplemental Experimental Procedures).

EXPERIMENTAL PROCEDURES

SUPPLEMENTAL INFORMATION

Mouse Husbandry
All animal experiments were approved by the Institutional Animal Care and
Use Committee at the Memorial Sloan Kettering Cancer Center. Mice were
maintained under a 12-hr light-dark cycle. Mouse lines used in this study
were Nanog:H2B-GFPTg/+, Nanogb-geo/+ (Mitsui et al., 2003), and NanogGFP/+
(Hatano et al., 2005). Alleles are schematized in Figure S1H.

Supplemental Information includes Supplemental Experimental Procedures,

six gures, and two movies and can be found with this article online at
http://dx.doi.org/10.1016/j.celrep.2015.02.010.

Live Embryo Imaging

For live imaging, embryos were cultured in glass-bottomed dishes (MatTek)
in an environmental chamber as done previously (Kang et al., 2013). Live imaging conditions used were compatible with normal development as shown
previously (Plusa et al., 2008). For incubation experiments, an ERK1/ERK2
inhibitor, 1 mM PD0325901 (StemGent), was added to medium 23 hr prior
to initiation of 3D time-lapse imaging. GFP was excited using a 488-nm Argon
laser. Live image data were acquired using four laser scanning confocal
imaging systems: Zeiss LSM510META, LSM710, LSM780, and Leica SP8.
Images were acquired using 203/0.75, 403/1.3, or 633/1.4 objectives.
2030 xy planes separated by 2 mm were acquired per z stack, every
15 min. Movies of 3D time-lapse sequences were compiled and annotated
using QuickTime Pro (Apple).
Immunostaining of ESCs and Embryos
Immunostaining of ESCs and embryos was performed as previously
(Kalmar et al., 2009; Kang et al., 2013; Munoz Descalzo et al., 2012). Primary
antibodies used were CDX2 (1:100, Biogenex), GATA6 (1:100, R&D Systems),
NANOG (1:500, Cosmo Bio), OCT4 (1:100, Santa Cruz Biotechnology), and
SOX17 (1:100, R&D Systems). Secondary Alexa Fluor (Invitrogen) conjugated
antibodies were used at 1:500. DNA was visualized with Hoechst 33342
(5 mg/ml; Invitrogen).
Quantitative Fluorescence Image Analysis
Quantitative uorescence measurements from images of xed ESCs and xed
or live embryos were performed using an automated image processing workow comprising the segmentation of multiple nuclei in 3D data (Figure 2A). The
front-end software, MINS, is a MATLAB (MathWorks)-based graphic user
interface described previously (Lou et al., 2014; http://katlab-tools.org). We
noted that for normalizing uorescence values of each channel the Hoechst
channel per cell was not optimal to compensate for loss of uorescence intensity throughout the sample depth (differences in Hoechst uorescence intensity throughout z stack indicated by orange arrowheads in Figure 2A). We
thus developed an algorithm to generate a regression curve across values of
the Hoechst channel for individual cells (details in the Supplemental Experimental Procedures). Fluorescence values for other channels were normalized
using this curve (Figure 2A). To compare uorescence values between ICM
and TE cells, TE cells being analyzed needed to be in the same focal plane
as ICM cells. Thus, TE cells positioned at the beginning and end of the z stack
were excluded from the analysis. Only accurately segmented nuclei were
included in analyses.
Nuclear Segmentation and Cell Tracking
Using the segmentation output of the MINS software, we developed an
algorithm for cell tracking. To determine the position of any given cell at a subsequent time point, the algorithm rst seeks to identify a clear, well-dened

AUTHOR CONTRIBUTIONS
P.X. and A.-K.H. conceived the project and designed the experiments. P.X.
generated the Nanog:H2B-GFP mouse line and performed ESC and xed
embryo imaging experiments. M.K. performed time-lapse embryo imaging experiments. A.P. and S.D.T. carried out quantitative analyses of experimental
data sets. P.X., M.K., and A.-K.H. wrote the manuscript with input from A.P.
and S.D.T.
ACKNOWLEDGMENTS
We thank X. Lou for developing the algorithm used to calculate regression curves for uorescence intensity normalizations; J. Nichols for the
Nanogb-geo/+ mouse strain; A. Martinez-Arias for TNGA ESCs; S. Nowotschin,
N. Saiz, and N. Schrode for discussions and comments on the manuscript;
the Memorial Sloan Kettering Molecular Cytology and Rockefeller University
Bio-Imaging Core Facilities for use of their instruments for live embryo imaging. Work in A.-K.Hs. laboratory is supported by the NIH (R01-HD052115
and R01-DK084391) and NYSTEM (N13G-236). S.D.T. is supported by NIH
(R00-HD074670). A.P. is supported by Finalized Research and Founding for
Investments in Basic Research (RBAP11BYNP-Newton).
Received: July 14, 2014
Revised: January 4, 2015
Accepted: January 31, 2015
Published: March 5, 2015
REFERENCES
Abranches, E., Bekman, E., and Henrique, D. (2013). Generation and characterization of a novel mouse embryonic stem cell line with a dynamic reporter of
Nanog expression. PLoS ONE 8, e59928.
Artus, J., Kang, M., Cohen-Tannoudji, M., and Hadjantonakis, A.K. (2013).
PDGF signaling is required for primitive endoderm cell survival in the inner
cell mass of the mouse blastocyst. Stem Cells 31, 19321941.
Bessonnard, S., De Mot, L., Gonze, D., Barriol, M., Dennis, C., Goldbeter, A.,
Dupont, G., and Chazaud, C. (2014). Gata6, Nanog and Erk signaling control
cell fate in the inner cell mass through a tristable regulatory network. Development 141, 36373648.
Boroviak, T., Loos, R., Bertone, P., Smith, A., and Nichols, J. (2014). The ability
of inner-cell-mass cells to self-renew as embryonic stem cells is acquired
following epiblast specication. Nat. Cell Biol. 16, 516528.
Chambers, I., Colby, D., Robertson, M., Nichols, J., Lee, S., Tweedie, S., and
Smith, A. (2003). Functional expression cloning of Nanog, a pluripotency
sustaining factor in embryonic stem cells. Cell 113, 643655.
Chambers, I., Silva, J., Colby, D., Nichols, J., Nijmeijer, B., Robertson, M.,
Vrana, J., Jones, K., Grotewold, L., and Smith, A. (2007). Nanog safeguards
pluripotency and mediates germline development. Nature 450, 12301234.

Cell Reports 10, 15081520, March 10, 2015 2015 The Authors 1519

Chazaud, C., Yamanaka, Y., Pawson, T., and Rossant, J. (2006). Early lineage
segregation between epiblast and primitive endoderm in mouse blastocysts
through the Grb2-MAPK pathway. Dev. Cell 10, 615624.
Czechanski, A., Byers, C., Greenstein, I., Schrode, N., Donahue, L.R., Hadjantonakis, A.K., and Reinholdt, L.G. (2014). Derivation and characterization of
mouse embryonic stem cells from permissive and nonpermissive strains.
Nat. Protoc. 9, 559574.
Dietrich, J.E., and Hiiragi, T. (2007). Stochastic patterning in the mouse preimplantation embryo. Development 134, 42194231.
Eckersley-Maslin, M.A., Thybert, D., Bergmann, J.H., Marioni, J.C., Flicek, P.,
and Spector, D.L. (2014). Random monoallelic gene expression increases
upon embryonic stem cell differentiation. Dev. Cell 28, 351365.
Faddah, D.A., Wang, H., Cheng, A.W., Katz, Y., Buganim, Y., and Jaenisch, R.
(2013). Single-cell analysis reveals that expression of nanog is biallelic and
equally variable as that of other pluripotency factors in mouse ESCs. Cell
Stem Cell 13, 2329.
Filipczyk, A., Gkatzis, K., Fu, J., Hoppe, P.S., Lickert, H., Anastassiadis, K., and
Schroeder, T. (2013). Biallelic expression of nanog protein in mouse embryonic
stem cells. Cell Stem Cell 13, 1213.
Frankenberg, S., Gerbe, F., Bessonnard, S., Belville, C., Pouchin, P., Bardot,
O., and Chazaud, C. (2011). Primitive endoderm differentiates via a threestep mechanism involving Nanog and RTK signaling. Dev. Cell 21, 10051013.
Gendrel, A.V., Attia, M., Chen, C.J., Diabangouaya, P., Servant, N., Barillot, E.,
and Heard, E. (2014). Developmental dynamics and disease potential of
random monoallelic gene expression. Dev. Cell 28, 366380.
ska, K., Saiz, N., Piliszek, A., Frankenberg, S., Nichols,
Grabarek, J.B., Zyzyn
J., Hadjantonakis, A.K., and Plusa, B. (2012). Differential plasticity of epiblast
and primitive endoderm precursors within the ICM of the early mouse embryo.
Development 139, 129139.
Guo, G., Huss, M., Tong, G.Q., Wang, C., Li Sun, L., Clarke, N.D., and Robson,
P. (2010). Resolution of cell fate decisions revealed by single-cell gene expression analysis from zygote to blastocyst. Dev. Cell 18, 675685.
Hatano, S.Y., Tada, M., Kimura, H., Yamaguchi, S., Kono, T., Nakano, T.,
Suemori, H., Nakatsuji, N., and Tada, T. (2005). Pluripotential competence of
cells associated with Nanog activity. Mech. Dev. 122, 6779.
Kalmar, T., Lim, C., Hayward, P., Munoz-Descalzo, S., Nichols, J., GarciaOjalvo, J., and Martinez Arias, A. (2009). Regulated uctuations in nanog
expression mediate cell fate decisions in embryonic stem cells. PLoS Biol. 7,
e1000149.
Kang, M., Piliszek, A., Artus, J., and Hadjantonakis, A.K. (2013). FGF4 is
required for lineage restriction and salt-and-pepper distribution of primitive
endoderm factors but not their initial expression in the mouse. Development
140, 267279.
Kurimoto, K., Yabuta, Y., Ohinata, Y., Ono, Y., Uno, K.D., Yamada, R.G., Ueda,
H.R., and Saitou, M. (2006). An improved single-cell cDNA amplication
method for efcient high-density oligonucleotide microarray analysis. Nucleic
Acids Res. 34, e42.
Lou, X., Kang, M., Xenopoulos, P., Munoz-Descalzo, S., and Hadjantonakis,
A.K. (2014). A rapid and efcient 2D/3D nuclear segmentation method for analysis of early mouse embryo and stem cell image data. Stem Cell Reports 2,
382397.

Mitsui, K., Tokuzawa, Y., Itoh, H., Segawa, K., Murakami, M., Takahashi, K.,
Maruyama, M., Maeda, M., and Yamanaka, S. (2003). The homeoprotein
Nanog is required for maintenance of pluripotency in mouse epiblast and ES
cells. Cell 113, 631642.
Morgani, S.M., Canham, M.A., Nichols, J., Sharov, A.A., Migueles, R.P., Ko,
M.S., and Brickman, J.M. (2013). Totipotent embryonic stem cells arise in
ground-state culture conditions. Cell Rep. 3, 19451957.
Munoz Descalzo, S., Rue, P., Garcia-Ojalvo, J., and Martinez Arias, A. (2012).
Correlations between the levels of Oct4 and Nanog as a signature for nave pluripotency in mouse embryonic stem cells. Stem Cells 30, 26832691.
Navarro, P., Festuccia, N., Colby, D., Gagliardi, A., Mullin, N.P., Zhang, W.,
Karwacki-Neisius, V., Osorno, R., Kelly, D., Robertson, M., and Chambers, I.
(2012). OCT4/SOX2-independent Nanog autorepression modulates heterogeneous Nanog gene expression in mouse ES cells. EMBO J. 31, 45474562.
Nichols, J., and Smith, A. (2012). Pluripotency in the embryo and in culture.
Cold Spring Harb. Perspect. Biol. 4, a008128.
Nichols, J., Silva, J., Roode, M., and Smith, A. (2009). Suppression of Erk signalling promotes ground state pluripotency in the mouse embryo. Development 136, 32153222.
Ohnishi, Y., Huber, W., Tsumura, A., Kang, M., Xenopoulos, P., Kurimoto, K.,
Oles, A.K., Arauzo-Bravo, M.J., Saitou, M., Hadjantonakis, A.K., and Hiiragi, T.
(2014). Cell-to-cell expression variability followed by signal reinforcement progressively segregates early mouse lineages. Nat. Cell Biol. 16, 2737.
Okita, K., Ichisaka, T., and Yamanaka, S. (2007). Generation of germlinecompetent induced pluripotent stem cells. Nature 448, 313317.
Plusa, B., Piliszek, A., Frankenberg, S., Artus, J., and Hadjantonakis, A.K.
(2008). Distinct sequential cell behaviours direct primitive endoderm formation
in the mouse blastocyst. Development 135, 30813091.
Schrode, N., Xenopoulos, P., Piliszek, A., Frankenberg, S., Plusa, B., and Hadjantonakis, A.K. (2013). Anatomy of a blastocyst: cell behaviors driving cell fate
choice and morphogenesis in the early mouse embryo. Genesis 51, 219233.
Schrode, N., Saiz, N., Di Talia, S., and Hadjantonakis, A.K. (2014). GATA6
levels modulate primitive endoderm cell fate choice and timing in the mouse
blastocyst. Dev. Cell 29, 454467.
Smith, A. (2013). Nanog heterogeneity: tilting at windmills? Cell Stem Cell
13, 67.
Takahashi, K., and Yamanaka, S. (2006). Induction of pluripotent stem cells
from mouse embryonic and adult broblast cultures by dened factors. Cell
126, 663676.
Tan, F.E., and Elowitz, M.B. (2014). Brf1 posttranscriptionally regulates pluripotency and differentiation responses downstream of Erk MAP kinase. Proc.
Natl. Acad. Sci. USA 111, E1740E1748.
Torres-Padilla, M.E., and Chambers, I. (2014). Transcription factor heterogeneity in pluripotent stem cells: a stochastic advantage. Development 141,
21732181.
Xenopoulos, P., Kang, M., and Hadjantonakis, A.K. (2012). Cell lineage allocation within the inner cell mass of the mouse blastocyst. Results Probl. Cell
Differ. 55, 185202.

MacArthur, B.D., Sevilla, A., Lenz, M., Muller, F.J., Schuldt, B.M., Schuppert,
A.A., Ridden, S.J., Stumpf, P.S., Fidalgo, M., Maayan, A., et al. (2012). Nanogdependent feedback loops regulate murine embryonic stem cell heterogeneity. Nat. Cell Biol. 14, 11391147.

Yamanaka, Y., Lanner, F., and Rossant, J. (2010). FGF signal-dependent

segregation of primitive endoderm and epiblast in the mouse blastocyst.
Development 137, 715724.

Messerschmidt, D.M., and Kemler, R. (2010). Nanog is required for primitive

endoderm formation through a non-cell autonomous mechanism. Dev. Biol.
344, 129137.

Ying, Q.L., Wray, J., Nichols, J., Batlle-Morera, L., Doble, B., Woodgett, J.,
Cohen, P., and Smith, A. (2008). The ground state of embryonic stem cell
self-renewal. Nature 453, 519523.

1520 Cell Reports 10, 15081520, March 10, 2015 2015 The Authors

7\a^W`Sg]c`^SS`a
eWbV1SZZ@S^]`ba

0`SOYW\U\SeU`]c\R-4`][[]ZSQcZO`US\SbWQab]RSdSZ]^[S\bOZ
\Sc`]PW]Z]Ug1SZZ@S^]`ba^cPZWaVSabV]cUVb^`]d]YW\UQcbbW\USRUS
`SaSO`QVa^O\\W\UbVSa^SQb`c[]TZWTSaQWS\QSa
1SZZ@S^]`baWaO\]^S\OQQSaaX]c`\OZbVOb]TTS`abVS_cOZWbg`WU]`O\R
dWaWPWZWbgg]cdSQ][Sb]Sf^SQbT`][1SZZ>`Saa
2]g]cVOdSO\SePW]Z]UWQOZW\aWUVb-AcP[Wbg]c`^O^S`b]1SZZ@S^]`ba
4]`[]`SW\T]`[ObW]\dWaWbeeeQSZZQ][QSZZ`S^]`ba

/\]^S\OQQSaaX]c`\OZeWbVW[^OQb
4`][1SZZ>`Saa

Cell Reports

Article
Direct Activation of STING in the Tumor
Microenvironment Leads to Potent
and Systemic Tumor Regression and Immunity
Leticia Corrales,1,3 Laura Hix Glickman,2,3 Sarah M. McWhirter,2 David B. Kanne,2 Kelsey E. Sivick,2 George E. Katibah,2
Seng-Ryong Woo,1 Edward Lemmens,2 Tamara Banda,2 Justin J. Leong,2 Ken Metchette,2 Thomas W. Dubensky, Jr.,2,4,*
and Thomas F. Gajewski1,4,*
1Department

of Pathology, The University of Chicago, 929 E57th Street GCIS 3H, Chicago, IL 60637, USA
BioTech, Inc., 626 Bancroft Way, 3C, Berkeley, CA 94710, USA
3Co-rst author
4Co-senior author
*Correspondence: [email protected] (T.W.D.), [email protected] (T.F.G.)
http://dx.doi.org/10.1016/j.celrep.2015.04.031
This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
2Aduro

SUMMARY

Spontaneous tumor-initiated T cell priming is dependent on IFN-b production by tumor-resident dendritic cells. On the basis of recent observations indicating that IFN-b expression was dependent upon
activation of the host STING pathway, we hypothesized that direct engagement of STING through intratumoral (IT) administration of specic agonists would
result in effective anti-tumor therapy. After proof-ofprinciple studies using the mouse STING agonist
DMXAA showed a potent therapeutic effect, we
generated synthetic cyclic dinucleotide (CDN) derivatives that activated all human STING alleles as
well as murine STING. IT injection of STING agonists
induced profound regression of established tumors
in mice and generated substantial systemic immune
responses capable of rejecting distant metastases
and providing long-lived immunologic memory. Synthetic CDNs have high translational potential as a
cancer therapeutic.
INTRODUCTION
The responsiveness of tumors to immunotherapy depends, at
least in part, on the immunophenotype of the tumor microenvironment (TME) (Gajewski et al., 2013). Substantial evidence indicates that tumor-inltrating lymphocytes (TILs) are correlated
with favorable prognosis in diverse malignancies (Galon et al.,
2012) and predicts a positive clinical outcome in response to
several immunotherapy strategies (Postow et al., 2012; Wolchok
et al., 2013). Understanding the underlying mechanisms that
promote spontaneous T cell inltration is critical toward developing new therapeutic strategies that can be used to effectively
promote an immune-responsive TME.
Innate immune sensing in the TME is a critical step in promoting spontaneous tumor-initiated T cell priming and subsequent
1018 Cell Reports 11, 10181030, May 19, 2015 2015 The Authors

TIL inltration (Fuertes et al., 2011). Transcriptional proling analyses of melanoma patients have revealed that tumors containing inltrating activated T cells are characterized by a type I IFN
transcriptional signature (Harlin et al., 2009). Studies in mice
have demonstrated that type I IFN signaling plays a critical role
in tumor-initiated T cell priming (Diamond et al., 2011; Fuertes
et al., 2011). Mice lacking the IFN-a/b receptor in DCs cannot
reject immunogenic tumors, and CD8a+ DCs from these mice
are defective in antigen cross-presentation to CD8+ T cells.
Furthermore, Baft3/ mice that lack the CD8a+ DC lineage
lose the capacity to spontaneously prime tumor-specic CD8+
T cells (Fuertes et al., 2011; Hildner et al., 2008). These ndings
in humans and in mice indicate that the tumor-resident antigenpresenting cell (APC) compartment is defective in non-T-cell-inamed tumors. Thus, strategies to induce type I IFN signaling
and APC activation in the TME to bridge the innate and adaptive
immune responses may have therapeutic utility.
Recent work has demonstrated that activation of the STING
pathway in tumor-resident host APCs is required for induction
of a spontaneous CD8+ T cell response against tumor-derived
antigens in vivo (Woo et al., 2014). In addition, activation of this
pathway and the subsequent production of IFN-b contributes
to the anti-tumor effect of radiation (Deng et al., 2014), which
can be potentiated with co-administration of a natural STING
agonist. STING (stimulator of interferon genes, also known as
TMEM173, MITA, ERIS, and MPYS) is a transmembrane protein
localized to the ER that undergoes a conformational change in
response to direct binding of cyclic dinucleotides (CDNs), resulting in a downstream signaling cascade involving TBK1 activation, IRF-3 phosphorylation, and production of IFN-b and other
cytokines (Burdette et al., 2011; Burdette and Vance, 2013; Ishikawa and Barber, 2008). IFN-b is the signature cytokine induced
in response to activating STING, by either exogenous CDNs produced by bacterial infection or through binding of a structurally
distinct endogenous CDN produced by a host cyclic GMPAMP synthetase (cGAS) in response to sensing cytosolic double-stranded DNA (dsDNA) (Ablasser et al., 2013; Diner et al.,
2013; McWhirter et al., 2009; Sun et al., 2013; Woodward
et al., 2010; Zhang et al., 2013). These observations suggested

that direct activation of the STING pathway in the TME by intratumoral (IT) injection of specic agonists might be an effective
therapeutic strategy to promote broad tumor-initiated T cell
priming against an individuals tumor antigen repertoire.
To test this therapeutic approach, we began with 5,6-dimethylxanthenone-4-acetic acid (DMXAA), a dened avonoid
compound known as a vascular disrupting agent that was shown
to have anti-tumor activity in mouse models (Baguley and Ching,
1997). This drug ultimately failed in humans when combined with
standard-of-care chemotherapy in a phase 3 efcacy trial in nonsmall-cell lung cancer (Lara et al., 2011). Interestingly, recent
structure-function studies of mouse STING (mSTING) and human STING (hSTING) demonstrated that DMXAA is a direct
ligand for mSTING (Conlon et al., 2013; Gao et al., 2013a; Kim
et al., 2013; Prantner et al., 2012). However, detailed analysis revealed that polymorphisms in hSTING rendered it unable to bind
DMXAA, therefore abrogating its activity in human cells. These
ndings provide a mechanistic insight for the lack of DMXAA efcacy in humans as well as the rationale for the development of
new pharmacologic compounds that potently activate hSTING.
Whereas STING agonists are being developed as vaccine adjuvants (Dubensky et al., 2013; Ebensen et al., 2011; Gray et al.,
2012), whether STING agonists could have direct anti-tumor
therapeutic effects has been under-explored and the lack of
dened agonists that could activate all known hSTING alleles
has been lacking.
In the current report, we conrm that DMXAA is a strong
agonist of the mSTING pathway in vitro and in vivo. We
show that IT injection of DMXAA effectively primes CD8+ T cell
responses to promote rejection of established tumors in a
STING-dependent fashion. Based on these proof-of-concept results, we synthesized a large panel of CDNs and selected compounds capable of activating all known hSTING alleles. Unlike
DMXAA, selected compounds indeed stimulated human PBMCs
to produce IFN-b. Like DMXAA, these STING agonists exhibit
signicant anti-tumor efcacy in several mouse tumor models,
without signicant local or systemic toxicity. Strikingly, direct
IT injection of selected CDNs into established B16 melanoma,
CT26 colon, and 4T1 breast carcinomas resulted in rapid and
profound tumor regression and promoted lasting systemic antigen-specic T cell immunity. These effects were entirely STINGdependent and resulted in regression of non-injected tumors in
the same hosts. We selected dithio-(RP, RP)-[cyclic[A(20 ,50 )
pA(30 ,50 )p]], (ML RR-S2 CDA) as the lead molecule for continued
development. This agent has high translational potential as a
therapeutic intervention strategy to induce activation of the
TME in multiple tumor types, with the mechanistic goal of generating effective tumor-initiated CD8+ T cell priming and lasting
anti-tumor efcacy.
RESULTS
DMXAA Stimulates the STING Pathway In Vitro
We rst conrmed that DMXAA was a functional agonist of the
STING pathway using mouse macrophages in vitro. STING aggregation was assessed using STING/ macrophages expressing mSTING-HA. Control macrophages presented a diffuse
pattern of STING in the cytoplasm, but after 1 hr of incubation

with DMXAA, approximately 60% of cells displayed aggregates

of STING in perinuclear sites (Figure 1A). Downstream phosphorylation of TBK1 and IRF3 was observed, which was abolished in
STING/ cells (Figure 1B). This correlated with an increase in
the apparent molecular weight of STING, which has been reported to be due to its phosphorylation (Konno et al., 2013).
STING/ macrophages reconstituted with mSTING-HA showed
restored phosphorylation of TBK1 and IRF3. IFN-b secretion was
detected from wild-type (WT), but not from STING/, macrophages in response to DMXAA (Figure 1C). Similar results were
observed with bone-marrow-derived DCs (BM-DC) from WT
versus STING/ mice (Figures S1A and S1B). We also used
BM-DCs cells to study the expression of additional immunoregulatory molecules. IFN-b, IFN-a, TNF-a, IL-1b, IL-6, and IL12p40
were induced after stimulation with DMXAA in WT cells, but not
STING/ BM-DCs (Figure S1C). Whereas LPS induced expression of CD40, PD-L1, CD86, and MHC class II in both WT and
STING-decient DCs, induction with DMXAA was observed
only in WT cells (Figures S1D and S1E). Together, these data
along with previous studies (Conlon et al., 2013; Gao et al.,
2013a; Kim et al., 2013; Prantner et al., 2012) conrm that
DMXAA is a strong agonist of mSTING, resulting in the production of IFN-b and other innate cytokines and activation of DCs.
DMXAA Induces Strong Anti-tumor Immunity In Vivo
In order to evaluate whether stimulation of STING could augment
anti-tumor immunity in vivo, we chose an IT route of administration to focus activation on those APCs acquiring tumor antigens.
To assess an antigen-specic immune response, we utilized the
B16 melanoma cell line transduced to express the model antigen
SIYRYYGL (B16.SIY) (Blank et al., 2004). B16.SIY tumor cells
were inoculated into the ank of WT or STING-decient mice
and injected IT with DMXAA at day 7. The dose of 500 mg of
DMXAA was chosen after examining single doses ranging from
150 to 625 mg, with the highest dose of 625 mg showing unacceptable toxicity (data not shown). In WT animals, the selected
dosage induced potent tumor regression and complete tumor
rejection in the majority of mice; however, no reduction in tumor
growth was observed in response to DMXAA in the absence of
host STING (Figure 2A). Analysis of splenocytes 5 days after
treatment showed a marked increase in the frequency of SIYspecic IFN-g-producing T cells in WT mice, but not in STING/
mice (Figure 2B). Similarly, treatment with DMXAA caused a high
frequency of SIY-specic CD8+ T cells detected by SIY/Kb pentamer staining in WT animals, but not in STING-decient mice
(Figure 2C). Next, we examined whether DMXAA treatment had
any effect in animals decient in the type I IFN receptor (IFNAR).
A signicant portion of the anti-tumor effect of DMXAA was lost
in IFNAR-decient mice, and none of the decient animals
showed complete tumor rejection (Figure 2D).
To determine whether immunologic memory was induced, WT
mice that had rejected B16.SIY tumors were re-challenged
60 days after the initial inoculation with the same tumor cells.
None of the re-challenged animals developed tumors (Figure 2E).
We then investigated whether the anti-tumor immune response
induced following DMXAA administration could be potent
enough to reject non-injected secondary tumors. B16.SIY cells
were injected in both anks of mice, but only one tumor was
Cell Reports 11, 10181030, May 19, 2015 2015 The Authors 1019

Figure 1. DMXAA Activates the STING Pathway and Promotes the Activation of APCs
(A) STING/ mouse bone-marrow-derived macrophages (BMM) transduced to express STING-HA tag were stimulated for 1 hr with 50 mg/ml DMXAA and stained
with specic antibodies against HA-tag, CD11b, and DAPI. Single cell images were acquired in the ImageStream, and data were analyzed with the IDEAS
software (Amnis; Millipore). The data in the graph represent average of percentage of cells with aggregates from three independent experiments.
(B) WT or STING/ BMM were stimulated with 50 mg/ml of DMXAA for the indicated time points. The amount of pTBK1, total TBK1, pIRF3, total IRF3, STING, and
GAPDH was measured by western blot.
(C) WT or STING/ BMM were stimulated with 50 mg/ml of DMXAA for 12 hr. The amount of secreted IFN-b was measured by ELISA.

treated with DMXAA. Tumor regression was observed in both

sites (Figure 2F), suggesting that IT DMXAA administration can
have a therapeutic effect on distant tumors. This effect was unlikely secondary to systemic distribution of the drug, because
deliberate systemic administration of DMXAA via intraperitoneal
administration had an inferior therapeutic effect (data not
shown). These results demonstrate that a STING agonist can
activate tumor-specic immune response capable of eliminating
distal tumors and protecting from tumor challenge.
To assess whether the potent anti-tumor efcacy resulting
from IT administration of DMXAA could be broadly applied, we
tested several additional syngeneic tumor models. Treatment
with DMXAA signicantly reduced the growth of B16.F10
(without expression of SIY) and TRAMPC2 tumors in C57BL/6
mice, 4T1 tumors in BALB/c mice, and Ag104L tumors in C3H
mice, indicating that the therapeutic effect of DMXAA is not
restricted to a specic tumor histology or mouse genetic background (Figures S2AS2D).
To determine whether the adaptive immune response was
required for tumor control, B16.SIY cells were inoculated into
RAG2/ mice that lack mature T and B cells. DMXAA treatment
lost most of its therapeutic effect in RAG2/ hosts, although
there was a partial control of tumor growth (Figure S2E). A similar
loss of therapeutic effect was observed in TCRa/ mice (Figure S2F) and in mice depleted of CD8+ T cells (Figure S2G),
but not in mice depleted of CD4+ T cells or NK cells (Figures
S2H and S2I). These results indicate that a signicant compo1020 Cell Reports 11, 10181030, May 19, 2015 2015 The Authors

nent of the therapeutic effect of DMXAA is mediated by CD8+

T cells.
Identication of Synthetic hSTING-Activating Molecules
Having shown that the STING pathway could be harnessed to
promote tumor antigen-specic CD8+ T cell priming, leading to
signicant therapeutic efcacy, we sought to identify compounds that could potently activate hSTING and therefore be
considered for clinical translation. CDNs have been studied as
small-molecule second messengers synthesized by bacteria,
which regulate diverse processes including motility and formation of biolms (Romling et al., 2013). The immunogenicity of recombinant protein antigens can be augmented with CDNs used
as an adjuvant, giving CDNs a potential application toward vaccine development (Dubensky et al., 2013; Ebensen et al., 2007a,
2007b, 2011; Gray et al., 2012). We sought to develop synthetic
CDN compounds with increased activity in human cells as well
as the ability to engage all known polymorphic hSTING molecules. The availability of CDN-STING crystal structures, along
with recent results describing hSTING allele/CDN-dependent
signaling relationships, facilitated structure-based studies to
design CDN compounds with increased activity. We synthesized
compounds that varied in purine nucleotide base, structure of
the phosphate bridge linkage, and substitution of the nonbridging oxygen atoms at the phosphate bridge with sulfur
atoms. Native CDN molecules are sensitive to degradation by
phosphodiesterases that are present in host cells or in the

Figure 2. Rejection of Tumors in Response to DMXAA Is STING Dependent

(A) WT or STING/ C57BL/6 mice were inoculated with 106 B16.SIY cells in the left ank. When tumor volumes were 100200 mm3, they received a single IT dose
of 500 mg of DMXAA or saline. Tumor volume was measured at the indicated time points (n = 5).
(B and C) WT or STING/ C57BL/6 mice (n = 5) were treated as in (A), and 5 days later, splenocytes were harvested and re-stimulated in vitro in the presence of
culture medium or soluble SIY peptide for 16 hr. The frequency of tumor-specic IFN-g-producing cells was assessed by ELISPOT (B), and the percentage of SIYspecic CD8+ T cells was assessed by staining splenocytes with antibodies against TCRb, CD4, CD8, and SIY pentamer (C). Cells were acquired in the LSRII-Blue
cytometer and analyzed with FlowJo software. Results are shown as mean SEM. *p < 0.05; **p < 0.01.
(D) WT or IFNAR/ C57BL/6 mice were inoculated with 106 B16.SIY cells in the left ank (n = 5). When tumor volumes were 100200 mm3, they received a single
IT dose of 500 mg of DMXAA or saline. Tumor volume was measured at the indicated time points.
(E) WT mice that had rejected B16.SIY tumors were re-challenged with 106 B16.SIY in the contralateral ank. Naive mice were used as controls. Tumor size was
measured at the indicated time points.
(F) WT mice were inoculated with 106 B16.SIY cells in the left and the right anks (n = 5). When tumor volumes were 100200 mm3, 500 mg of DMXAA or saline was
injected IT in the right ank only and tumor volume was measured at the indicated time points. Data are representative of at least three independent experiments
or two independent experiments for the contralateral tumor model. Results are shown as mean tumor volume SEM. *p < 0.5; **p < 0.01; ***p < 0.001. ns, not
signicant.

systemic circulation (Yan et al., 2008). We found that Rp, Rp (R,R)

dithio-substituted diastereomer CDNs were both resistant to
digestion with snake venom phosphodiesterase and induced
higher expression of type I IFNs in human THP-1 cells compared
to the Rp, Sp (R,S) dithio-substituted diastereomers or unmodied CDNs (data not shown).
To increase their afnity for STING, CDNs were also synthesized with a phosphate bridge conguration containing both
20 -50 and 30 -50 linkages, termed mixed linkage (ML), as found
in endogenous human CDNs produced by cGAS (Ablasser
et al., 2013; Diner et al., 2013; Gao et al., 2013b; Wolchok
et al., 2013). The synthesis of dithio mixed-linkage CDNs, via
modications of literature procedures (Gaffney et al., 2010), resulted in both R,R- and R,S dithio diastereomers, which were puried and separated by a combination of silica gel and C18
reverse-phase prep-HPLC chromatography, affording CDNs
with R95% purity as shown for ML RR-S2 CDA (Figure S3A, upper panel). The spectra for both 1H NMR (data not shown) and
the 31P NMR (y axis of Figure S3A, upper panel) were consistent

with ML RR-S2 CDA. Direct evidence for the regiochemistry of

the phosphodiester linkages was obtained by 1H-1H COSY (correlation spectroscopy for assignment of ribose protons shown
on x axis of Figure S3A, lower panel), in combination with a
1
H-31P HMBC (heteronuclear multiple-bond correlation spectroscopy) 2D NMR (Figure S3A, lower panel). The 3D X-ray crystal structure of ML RR-S2 CDA conrms the presence of the
20 -50 , 30 -50 mixed phosphodiester linkage and a dithio [RP, RP]
diastereomer conguration (Figure S3B).
Synthetic CDNs Have Enhanced Binding Afnity to
STING and Activate All Known hSTING Alleles
We puried recombinant hSTING and evaluated the relative
binding afnity to various modied CDNs using differential scanning uorimetry (DSF). DSF measures the stability of complex
formation as a function of temperature as an indirect readout
of protein-ligand association (Cavlar et al., 2013; Niesen et al.,
2007). The increased shift in thermal stability of hSTING bound
to ML RR-S2 CDA or ML RR-S2 cGAMP relative to unmodied
Cell Reports 11, 10181030, May 19, 2015 2015 The Authors 1021

(legend on next page)

1022 Cell Reports 11, 10181030, May 19, 2015 2015 The Authors

CDNs indicates that R,R dithio and ML modications enhance

the binding afnity to STING (Figure 3A). Similar results were obtained using puried mSTING (Figure S3C). It has been shown
recently that the bisphosphothionate analog of endogenous
cGAMP (ML cGAMP) is resistant to hydrolysis by ENPP1 phosphodiesterase and thus is more potent at inducing IFN-b secretion in human THP1 cells (Li et al., 2014). Similarly, we found that
R,R dithio-modied CDA compounds (ML RR-S2 CDA and RRS2 CDA) showed enhanced type I IFN production over CDA in
THP-1 human monocytes (Figure S3D).
SNPs in the hSTING gene have been shown to affect the
responsiveness to bacterial-derived canonical CDNs (Diner
et al., 2013; Gao et al., 2013b). Five haplotypes of hSTING
have been identied (WT, REF, HAQ, AQ, and Q alleles), which
vary at amino acid positions 71, 230, 232, and 293 (Figure 3B)
(Jin et al., 2011; Yi et al., 2013). To test the responsiveness of
the ve hSTING variants to synthetic CDNs, we created stable
HEK293T cell lines (defective in endogenous STING signaling)
expressing each of the full-length hSTING variants. Similar levels
of STING protein were expressed in each of the cell lines (Figure 3C). DMXAA potently activated mSTING and failed to activate any of the ve hSTING alleles (Figure S3E), consistent
with previous studies evaluating one of the hSTING alleles (Conlon et al., 2013). Cells expressing hSTINGREF responded poorly
to stimulation with the bacterial CDN compounds cGAMP,
CDA, and CDG but were responsive to the endogenously produced cGAS product, ML cGAMP (Diner et al., 2013). Interestingly, the hSTINGQ allele was also refractory to the bacterial
CDNs. Cells expressing mSTING were responsive to all of the
CDNs tested (Figure 3D). Cells transformed with either an empty
vector or expressing a non-functional mutant (I199N) STING protein (Goldenticket; Sauer et al., 2011) were not responsive to any
of the compounds (data not shown). In contrast, the dithio,
mixed-linkage CDN derivatives (ML RR-CDA, ML RR-S2 CDG,
and ML RR-S2 cGAMP) potently activated all ve hSTING
alleles, including the refractory hSTINGREF and hSTINGQ alleles
(Figure 3D).
CDN Derivatives Potently Induce STING-Dependent
Signaling in Murine and Human Immune Cells
To determine whether modied CDNs activated downstream
STING signaling, we assessed murine bone marrow macrophages (BMMs) isolated from WT C57BL/6 and STING/ (Goldenticket) mice (Sauer et al., 2011) for induction of IFN-b and other
cytokines. Synthetic dithio mixed-linkage CDNs (ML RR-S2 CDA

and ML RR-S2 CDG) induced the highest expression of IFN-b

and the pro-inammatory cytokines TNF-a, IL-6, and MCP-1
on a molar equivalent basis, as compared to endogenous ML
cGAMP and the TLR3 agonist poly I:C (Figure S4A). The modied
CDNs did not induce signaling in STING/ BMMs, whereas, as
expected, poly I:C agonists were still active. ML RR-S2 CDA was
also found to induce aggregation of STING and induce phosphorylation of TBK1 and IRF3 in mouse BMM (Figures S4B and
S4C). All of the modied CDNs tested also enhanced MHC class
II and expression of co-stimulatory markers in a STING-dependent manner (Figure S4D).
To examine activation of STING signaling in primary human
cells, we stimulated PBMCs from a panel of human donors
harboring different STING alleles (hSTINGWT/WT, hSTINGWT/REF,
and hSTINGWT/HAQ) and measured induction of IFN-b. In
contrast to a lack of activation by DMXAA, dithio-modied ML
CDNs induced IFN-b expression across these donors (Figure 3E).
Importantly, dithio-modied ML CDNs, as well as the non-canonical cGAS product ML cGAMP, induced IFN-b expression
in a donor homozygous for the hSTINGREF allele, which was refractory to stimulation with canonical CDNs in HEK293T cells
(Figures 3D and 3F). Looking at protein secretion by multiple donors that are homozygous for the hSTINGWT allele, ML RR-S2
CDA induced signicantly higher levels of IFN-a when compared
to ML cGAMP (Figure S4E). Thus, ML RR-S2 CDNs are viable
clinical candidates capable of activating STING pathway in human cells harboring different STING alleles and genotypes.
Intratumoral Delivery of Synthetic CDN Derivatives
Results in Profound Anti-tumor Efcacy in Established
B16 Melanoma
To evaluate whether modied dithio ML CDN compounds
conferred increased anti-tumor activity, mice bearing established B16.F10 tumors were treated with three IT injections of
CDN derivatives over a 1-week period. Whereas treatment with
ML CDA and ML CDG had modestly reduced tumor growth,
the R,R dithio derivatives profoundly inhibited tumor growth (Figure 4A). However, ML RR-S2 CDG was reactogenic, and these
mice developed open wounds in the treated tumor that did not
heal (data not shown). Lower dose levels of ML RR-S2 CDG
were not efcacious (data not shown), indicating that this molecule had a narrow therapeutic index. In contrast, no injection site
reactogenicity was observed with ML RR-S2 CDA and several
mice developed vitiligo upon fur regrowth following complete
eradication of the treated tumor (data not shown). Importantly,

Figure 3. Modied CDNs Potently Activate STING and Signal through All Human STING Allelic Variants
(A) Puried human STING binding to CDNs were analyzed by thermal shift assay. Temperature curves are the average from a representative experiment of three
independent experiments performed in duplicate. Tm shift values are mean values SEM.
(B) Domain structure of hSTING is shown with the positions of the amino acid variations (bottom). The allelic frequencies of the hSTING isoforms shown on the left
hand column were obtained from the 1000 Genome Project database as previously described (Yi et al., 2013).
(C) HEK293T cells were stably transfected with the indicated STING alleles. Whole-cell lysates from HEK293T cells stably expressing the indicated full-length
STING-HA proteins were analyzed by western blot with anti-HA antibodies.
(D) HEK293T cells expressing the indicated STING alleles were transfected with an IFN-b-luciferase reporter construct. After 24 hr, cells were stimulated for 6 hr
with the indicated CDN compound (10 mM) and assessed for IFN-b-reporter activity.
(E and F) Human PBMCs from donors with the indicated STING alleles were stimulated with 10 mM of the indicated CDN or 100 mg/ml DMXAA (E), or human
PBMCs from a donor homozygous for the reference variant (STINGREF/REF) were stimulated with 10 mM and 50 mM of the indicated CDN or 100 mg/ml DMXAA (F).
After a 6-hr stimulation, fold induction of IFN-b was measured by qRT-PCR and relative normalized expression was determined by comparison with untreated
controls. Results are representative of at least two independent experiments.

Cell Reports 11, 10181030, May 19, 2015 2015 The Authors 1023

Figure 4. Synthetic CDN Modications Signicantly Improve Anti-tumor Efcacy in

Established B16 Tumors
(A and B) WT C57BL/6 mice were inoculated with
5 3 104 B16.F10 cells in the left ank (n = 8). When
tumor volumes were 100 mm3, they received three
25 mg IT doses of ML CDA, ML CDG, ML RR-S2
CDG, ML RR-S2 CDA, or HBSS as control (A) or
either three 50 mg doses of the endogenous cGAS
product ML cGAMP, ML RR-S2 CDA, ML RR-S2
cGAMP, or HBSS as control (n = 8; B).
(C) WT C57BL/6 mice or STING/ mice were
treated with three IT doses of CDN ML RR-S2 CDA
(50 mg), murine type B CpG ODN 1668 (50 mg), or
HBSS vehicle control. Treatments were administered on the days indicated by the arrows, and
tumor measurements were taken twice weekly.
Data are representative of at least two independent experiments. Results are shown as mean
tumor volume SEM. **p < 0.01; ***p < 0.001. ns,
not signicant.

these mice demonstrated signicantly higher overall survival

compared to mice treated with ML RR-S2 CDG (Figure S5A).
ML RR-S2 CDA also showed higher anti-tumor control than the
endogenous ML cGAMP (Figure 4B). Thus, ML RR-S2 CDA
was selected for advancement to clinical development.
To determine whether the CDN-induced anti-tumor efcacy
was STING-dependent, we compared activity in B16 tumorbearing WT (C57BL/6) and STING/ mice. CDN therapeutic efcacy was completely lost in STING/ mice. In contrast, a CpG
TLR9 agonist (Kawarada et al., 2001) did modestly reduce tumor
growth in STING/ mice, demonstrating that this mouse strain
is capable of mounting an immune-mediated antitumor response (Figure 4C). A dose response of the ML RR-S2 CDA compound was performed in B16 tumor-bearing mice, which
identied an optimal anti-tumor dose level that also elicited
maximum tumor antigen-specic CD8+ T cell responses (data
not shown) and improved long-term survival to 50% (Figures
S5B and S5C).
To evaluate which cell types are targeted by STING agonists in
the TME, we analyzed the expression of IFN-b after ML RR-S2
CDA or DMXAA stimulation in BM-DC, bone marrow-derived
macrophages, puried T cells from naive mice, B16 tumor cells,
mouse embryonic broblasts (MEFs), and mouse primary dermal
broblasts. Except for tumor cells, all cell types tested were
found to express IFN-b. However, expression in WT BM-DCs
was ten times higher than in the other cell types (Figure S5D).
To determine whether this was also the case in the TME, we
sorted four different cell populations from pre-established B16
tumors: DCs (CD45+ CD11c+ MHCII+), macrophages (CD45+
CD11b+ F4/80+ MHC-II+), T cells (CD45+ CD3+), and endothelial
1024 Cell Reports 11, 10181030, May 19, 2015 2015 The Authors

cells (CD45 CD31+). Sorted cells were

then stimulated ex vivo with ML RR-S2
CDA or DMXAA. All subsets expressed
IFN-b upon stimulation with the STING
agonists. Expression in macrophages
was highest, followed by DCs, which
were both higher as compared with lymphocytes and endothelial cells (Figure S5E). These data demonstrate that the main source of type I IFN in the TME is likely APCs,
although stromal cells and T cells might also contribute. Interestingly and in agreement with these data, we observed that the
CD8a+/CD103+ DCs play a critical role in vivo, as the therapeutic
effect of DMXAA was signicantly diminished in Batf3/ mice
(Figure S5F). Altogether, these ndings suggest that CD8a+/
CD103+ DCs are critical for curative anti-tumor therapy by
STING agonists.
ML RR-S2 CDA Induces Lasting Immune-Mediated
Tumor Rejection in Multiple Tumor Types
To test anti-tumor efcacy in diverse tumor models, BALB/c
mice bearing established 4T1 colon or CT26 mammary carcinomas were treated with ML RR-S2 CDA. All treated animals
showed signicant and durable tumor regression. Mice that
were cured of their primary tumor were completely resistant to
re-challenge when the same tumor cell line was used (Figures
5A and 5B). However, animals cured from CT26 tumors after
ML RR-S2 CDA treatment showed no protection when they
were re-challenged with the 4T1 tumor cells, demonstrating
specicity (Figure 5B). Increased T cell responses were
observed against the endogenous CT26 rejection antigen AH1
(Slansky et al., 2000) (Figure 5C). IT injection of ML RR-S2 CDA
into one tumor in BALB/c mice bearing bilateral CT26 or 4T1 tumors also demonstrated signicant regression of the contralateral, untreated tumor (Figures 5D and S5G). Using a different
model to study the distal effect of ML RR-S2 CDA, we implanted
B16.F10 melanoma in C57BL/6 mice and 7 days later intravenously infused B16.F10 melanoma cells to generate lung

metastases. The 2-week-old established ank tumors were

treated with ML RR-S2 CDA, DMXAA, or HBSS control, and
2 weeks later, lung metastases were enumerated. Mice treated
in the ank tumor with ML RR-S2 CDA showed substantial control of distant lung metastases (Figure 5E). Together, these
results demonstrate that IT injection with ML RR-S2 CDA eradicates multiple tumor types and primes an effective systemic
CD8+ T cell immune response that signicantly inhibits the
growth of distal, untreated lesions.
DISCUSSION
Our results indicate that IT administration of STING agonists
generates a potent anti-tumor immune T cell response and striking durable disease regression in multiple mouse tumor models.
Although DMXAA has a potent therapeutic effect in mice, it lacks
the ability to activate hSTING. The chemically modied CDN
compounds described here have the capacity to activate all human polymorphic STING molecules while retaining the ability to
engage mSTING and demonstrate a similarly impressive anti-tumor effect. The participation of the adaptive immune response
for the observed anti-tumor responses is supported by the secondary rejection of non-injected tumors, the clearance of lung
metastases, and long-term immunologic memory observed
against autologous tumor re-challenge.
The mechanism of the therapeutic effect observed with the
compounds tested in this study was absolutely dependent on
host STING, and the majority of the anti-tumor effect was dependent upon T cells, specically CD8+ T cells, as described (Jassar
et al., 2005; Wallace et al., 2007). However, a partial therapeutic
effect was observed in RAG/ and TCRa/ mice, indicating
that innate immune cells are an important contributor to anti-tumor efcacy. Our in vitro data show that the various synthetic
STING agonists tested induced IFN-b production by APCs via a
mechanism that depended on the classical STING-TBK1-IRF3
signaling pathway (Ishikawa et al., 2009). In addition, STING agonists induced production of other cytokines, DC maturation, and
also chemokine production in vitro. Previous work characterizing
the vascular disrupting property of DMXAA demonstrated induction of TNF-a by stromal cells in vivo (Joseph et al., 1999) and
attenuation of the therapeutic effect in TNFR/ mice (Zhao
et al., 2002). Thus, the T-cell-independent component of the therapeutic effect of STING agonists in vivo may be mediated through
early TNF-a-mediated tumor vascular destruction. In addition, it
seems likely that the induction of chemokines by STING agonists
may also contribute to effective migration of activated T cells into
the TME within the injected tumor site. This multi-faceted mechanism of action may explain the therapeutic potency of STING
agonists against a range of cancers in vivo.
We developed synthetic CDN-derivative molecules based on
rationally designed STING structure-function relationships. The
lead molecule ML RR-S2 CDA has several features that improve
both stability and lipophilicity, promoting signicantly increased
STING signaling as compared to endogenous and pathogenderived CDNs. Whereas canonical CDNs have been evaluated
as vaccine adjuvants and were recently shown to inhibit growth
of 4T-1 tumors when given by intraperitoneal injection (Chandra
et al., 2014), those investigations used canonical CDNs that may

not be appropriate for clinical development, because there are

hSTING alleles at signicant frequencies in the population that
are refractory to these structures. In the present study, we
show the profound anti-tumor efcacy resulting from activation
of the STING pathway with synthetic CDNs that not only activate
all known hSTING alleles but have signicantly higher potency
than the natural STING ligands generated by cGAS. Although
some human donors showed higher responsiveness to ML
cGAMP, such as the donor bearing the refractory reference allele
represented in Figure 3F, overall ML RR-S2 CDA activated all
known hSTING allelic variants. This compound will therefore
be attractive for clinical development.
A possible limitation of the treatment approach described
herein is the necessity for IT injection to achieve maximal therapeutic effect. However, a practical advantage of this strategy is
that it has the potential to generate T cell responses against
tumor-specic antigens expressed by a patients individual cancer. The attractiveness of IT injection approaches has been rekindled based on several recent clinical trial observations. IT
injection of the oncolytic virus T-VEC has been shown in a randomized trial to provide improved clinical activity in melanoma
patients compared with control (Goins et al., 2014). In addition,
Levy and colleagues have shown that IT injection of the TLR9
agonist CpG along with local low-dose radiation therapy had
clinical activity in patients with non-Hodgkins lymphoma (Brody
et al., 2010). Both of these studies demonstrated regression of
non-injected lesions, consistent with the induction of tumor-specic T cells that could promote regression of tumors at distant
sites. These observations, along with the impressive potency
of STING agonists preclinically, support the development of
clinical strategies for IT injection of STING agonists as a cancer
therapeutic in patients.

EXPERIMENTAL PROCEDURES
Cells and Cell Isolations
The cells used for the in vivo experiments were: the C57BL/6-derived
melanoma cell lines B16.F10 and B16.F10.SIY (henceforth referred to as
B16.SIY), the breast cancer 4T1 cell line, and the colon cancer CT26 cell lines,
all originally purchased from ATCC. All cells were maintained at 37 C with 5%
CO2 in DMEM supplemented with 10% heat-inactivated FCS, penicillin,
streptomycin, L-arginine, L-glutamine, folic acid, and L-asparagine.
Immortalized WT and STING/ macrophages were obtained as described
in Roberson and Walker (1988). The WT macrophages were obtained from Dr.
K. Fitzgerald (University of Massachusetts). Non-immortalized macrophages
were derived from the bone marrow of WT (C57BL/6) or STING/ mice and
cultured in BMM media (RPMI media with 5% CSF, 5% FBS, 13 L-glutamine,
and 13 pen/strep) for 7 days prior to use. Human PBMCs were isolated by
density-gradient centrifugation using Ficoll-Paque Plus (GE Healthcare).
For stable overexpression of HA-STING in STING/ macrophages, the fulllength mSting-HA DNA sequence was generated by PCR. Sequence encoding
full-length mSTING was amplied from pUNOI-mSTING plasmid (Invivogen)
using a 50 primer containing an EcoRI site, gcagacGAATTCATGCCATACTC
CAACCTGCATCCAGCCATCCCACGGCCCAGAGGTCACCGCTCCAAATAT
GTAGCCCTCATCTTTCTGGTGGCCAG, and a 30 primer containing the HAtag nucleotide sequence, followed by a TGA stop codon and a NotI site, tca
catGCGGCCGCTCAGGCGTAGTCAGGCACGTCGTAAGGATAGATGAGGTC
AGTGCGGAGTGGGAGAGGCTGATCC. The mSTING-HA PCR product was
gel puried and double digested with EcoRI and NotI and then cloned into
the multiple cloning site of pMXS-IRES-GFP with Quick ligation kit (New
England Biolabs).

Cell Reports 11, 10181030, May 19, 2015 2015 The Authors 1025

Figure 5. ML RR-S2 CDA Promotes Immune-Mediated Tumor Rejection

(A and B) WT BALB/c mice were inoculated with 105 4T-1 breast cancer (A) or CT26 (B) colon carcinoma cells in the left ank. When tumor volumes were 100 mm3,
they received three 50 mg doses IT of ML RR-S2 CDA or HBSS vehicle control (left graph). Mice were re-implanted with 105 4T1 (A and B) or CT26 (B) tumor cells on
the opposite ank on day 55 post-initial tumor implantation. Naive mice were used as controls (right graph; n = 8).
(C) WT BALB/c mice were inoculated with 105 CT26 colon carcinoma cells in the left ank and treated on days 11, 14, and 18 with IT injections of ML RR-S2 CDA
(25 mg each) or HBSS vehicle control (n = 4). 21 days post-implantation of CT26 tumors, PBMCs were stimulated with AH1 (gp70423-431) and assessed by IFN-g
ELISPOT assay.

(legend continued on next page)

1026 Cell Reports 11, 10181030, May 19, 2015 2015 The Authors

Stable HEK293T STING-expressing cell lines were generated with MSCV2.2

retroviral plasmids that contain STING cDNA cloned upstream of an IRES
in frame with GFP. hSTING(REF)-HA, hSTING(WT)-HA, hSTING(HAQ)-HA,
hSTING(Q)-HA, and mSTING(WT)-HA retroviral plasmids were obtained from
the Vance Laboratory at UC Berkeley. hSTING(AQ)-HA was derived from
hSTING(Q)-HA using a QuickChange Site-Directed Mutagenesis kit (Stratagene). Retroviral vectors were transfected into the amphotropic Phoenix packaging cell line using Lipofectamine (Invitrogen). After 2 days, viral supernatants
were harvested and used for transduction of STING/ macrophages or
HEK293T cells. GFP+ cells were sorted in FACSAria (BD Bioscience) or
MoFlow cell sorters.
Protein Expression and Purication
STING ligand-binding domains (human amino acids [aas] 140379; mouse aas
139378) were cloned via ligation-independent cloning into a custom pETbased vector containing a 6xHIS-SUMO-tobacco etch virus protease (TEV)
site. All plasmids were conrmed by sequencing. Proteins were expressed in
BL21 DE3 Rosetta 2 cells (EMD Millipore). Cells were grown in LB media at
37 C until an OD600 of about 0.6. Cells were then shifted to 18 C, induced
with 0.25 mM isopropyl-beta-D-thiogalactopyranoside, and grown for 18
20 hr. Fusion proteins were puried on Ni-NTA agarose (QIAGEN). The
6xHIS-SUMO tag was removed by digestion with TEV protease (Sigma Aldrich)
overnight during dialysis against buffer containing 20 mM Tris-HCl, 150 mM
NaCl, 5 mM imidazole, 10% glycerol, and 0.5 mM Tris (2-carboxyethyl) phosphine (pH 7.5). After dialysis, the 6xHIS-SUMO tag and TEV protease were
removed by Ni-NTA agarose. Proteins were concentrated to between 9 and
13 mg/ml. Aliquots were ash frozen in liquid nitrogen and stored at 80 C.
ImageStream Analysis of STING Aggregation in Cells
STING/ macrophages overexpressing STING-HA tag were stimulated for
1 hr with 50 mg/ml of DMXAA resuspended in 7.5% of NaHCO3, 50 mM of
ML RR-S2 CDA resuspended in HBSS, or only the vehicles as control. After
the incubation, cells were stained with anti-CD11b-APC (M1/70; BioLegend),
rabbit anti-HA-tag (C29F4; Cell Signaling) and anti-rabbit IgG-PE (Invitrogen),
and DAPI (Invitrogen). Single cell images were acquired in the ImageStreamx
Mark II (Amnis), and data were analyzed using IDEAS software.
Western Blot Analysis
WT, STING/ macrophages, and STING/ macrophages overexpressing
STING-HA or an empty vector were stimulated with 50 mg/ml DMXAA for 0,
15, 60, or 180 min; BM-DCs from WT or STING/ mice were stimulated
with 25 mg/ml DMXAA for the same time points. Proteins were extracted
with Triton-X100 buffer (150 mM sodium chloride, 50 mM Tris, 1% TritonX100 [pH 8.0]) with proteinase inhibitors (Thermo Scientic) and phosphatase
inhibitors (Sigma). 30 mg of protein was electrophoresed in 10% SDS-PAGE
gels and transferred onto Immobilon-FL membranes (Millipore). Blots were
incubated with antibodies specic for phosphorylated TBK1 (Ser172), phosphorylated IRF3 (Ser396), total TBK1, STING, and GAPDH (Cell Signaling) or
total IRF3 (Invitrogen). Proteins from HEK293T lines stably expressing STING
were extracted with M-PER (Thermo Scientic). 6 mg of protein was loaded
onto a 412% MES NuPAGE gel (Life Technologies), transferred to nitrocellulose, and probed with anti-HA antibody (Santa Cruz Biotechnology). Anti-rabbit IRDye 680RD label secondary antibody was used for visualization of bands
with the Odyssey Imaging system (LI-COR Biosciences).
Differential Scanning Fluorimetry
Thermal shift assays were performed as in Cavlar et al. (2013). Assays were
conducted with STING ligand binding domain at 1 mg/ml with or without

various CDNs at 1 mM in 20 mM Tris-HCl, 150 mM NaCl (pH 7.5), and 1:500

dilution of SYPRO Orange Dye (Life Technologies). The uorescence as a
function of temperature was recorded in a CFX 96 real-time PCR machine
(Bio-Rad) reading on the HEX channel EX 450490 EM 560580 nm. The temperature gradient was from 15 C to 80 C, ramping 0.5 C per 15 s. Curves were
t to a Boltzmann sigmoidal (Graph Pad Prism) to establish the midpoint of
thermal unfolding (Tm).
Murine IFN-b ELISA
WT or STING/ macrophages and BM-DCs from WT or STING/ mice were
stimulated with 50 mg/ml DMXAA. Conditioned media were collected after 4 hr.
IFN-b concentration was assessed using VeriKine Mouse Interferon Beta
ELISA Kit (PBL interferon source).
qRT-PCR Analysis of Cytokines
BM-DCs from WT or STING/ mice were stimulated with 25 mg/ml DMXAA or
100 ng/ml LPS for 4 hr. Total RNA was isolated using the RNeasy kit (QIAGEN)
and incubated with DNase I, Amplication Grade (Invitrogen). cDNA was synthesized using High Capacity cDNA Reverse Transcription Kit (Applied Biosystems), expression of cytokines was measured by real-time qRT-PCR using
specic primers/probes for mouse INF-b, TNF-a, IL-6, and IL12p40, and panspecic primers were to quantify expression of the IFN-a family. Primer sequences are listed in Table S1 in Supplemental Information. PCR reactions
were performed in the 7300 Real Time PCR system (Applied Biosystems).
The results are expressed as 2DCt using 18S as endogenous control.
WT BMM was stimulated with CDN at 5 mM in HBSS with the addition of Effectene (QIAGEN) transfection reagent (per kit protocol). Human PBMCs were
stimulated as indicated. Stimulated cells were assessed by real-time qRTPCR for gene expression of IFN-b1, MCP-1, TNF-a, and IL-6 using the
PrimePCR RNA purication, and cDNA analysis system, and run on the
CFX96 gene cycler (Bio-Rad). Relative normalized expression was determined
by comparing induced target gene expression to unstimulated controls, using
the reference genes Gapdh and Ywhaz (mouse) and GUSB and PGK1 (human),
genes conrmed to have a coefcient variable (CV) below 0.5 and M value
below 1, and thus did not vary with different treatment conditions.
Mice
All animals were used according to protocols approved by Institutional Animal
Use Committee of the University of Chicago and Aduro Biotech and maintained in pathogen-free conditions in a barrier facility. C57BL/6, BALB/c,
C3H/He, and TCRa/ mice were obtained from Jackson and Charles River.
RAG2/ mice were obtained from Taconic. Tmem173/ (STING-decient)
mice were provided by Dr. G. Barber (University of Miami), and STING/
(Goldenticket; Sauer et al., 2011) and IFNAR/ mice were purchased from
Jackson Laboratories. Batf3/ mice were provided by Dr. Kenneth M. Murphy
(Washington University School of Medicine).
In Vivo Tumor Experiments
106 of B16-SIY tumor cells, 5 3 104 B16.F10 tumor cells, 105 4T1 and CT26, or
106 other tumor cells were injected s.c. in 100 ml DPBS or HBSS on the right
ank of mice. Following tumor implantation, mice were randomized into treatment groups. When tumors were 100200 mm3 in volume (57 mm wide),
either one single or three doses of DMXAA resuspended in 7.5% of NaHCO3
or CDNs formulated in HBSS or vehicle control were injected IT. Measurements of tumors were performed twice per week using calipers, and the tumor
volume was calculated with the formula V = (length 3 width2)/2. In some experiments, tumor-free survivors were re-challenged with tumor cells on the opposite ank several weeks after the injection of the primary tumor. Naive mice

(D) WT BALB/c mice were implanted with 105 of CT26 tumor cells on both anks. On the days indicated, mice were treated in one ank only with ML RR-S2 CDA
(50 mg) or HBSS vehicle control (n = 8).
(E) WT C57BL/6 were inoculated with 5 3 104 B16.F10 melanoma cells on the right ank at day 0 and implanted i.v. with 105 cells on day 7. Naive mice were
implanted with cells i.v. only as a control. Flank tumors were treated on the days indicated with ML RR-S2 CDA (50 mg), DMXAA (150 mg), or HBSS control (n = 8).
On day 28, lungs were harvested and lung tumor nodules counted. The histogram depicts total numbers of nodules in the ML RR-S2 CDA, DMXAA, or HBSScontrol-treated mice, compared to the untreated i.v.-only tumor implanted mice. The images depict the ML RR-S2 CDA and HBSS-control-treated mice. Data are
representative of at least two independent experiments. Results are shown as mean SEM. **p < 0.01; ***p < 0.001.

Cell Reports 11, 10181030, May 19, 2015 2015 The Authors 1027

were used as controls. For the contralateral experiments, mice were implanted
on both anks and only one tumor was treated. For the B16 melanoma lung
metastasis experiments, mice were implanted on the ank with 5 3 104 cells
B16.F10 on day 0 and then injected intravenously with 1 3 105 cells on
day 7. Lungs were harvested on day 28. Administration of compounds, measurements of tumors, and counting of lung tumors were performed in a blinded
fashion.
IFN-g ELISPOT and SIY-Pentamer Staining
Splenocytes were analyzed 5 days after the rst IT injection of DMXAA or ML
RR-S2 CDA. For the ELISPOTs, 106 splenocytes were plated per well and stimulated overnight with SIY peptide (160 nM) or AH1 (1 mM) peptide, with PMA
(50 ng/ml) plus ionomycin (0.5 mM) as a positive control or medium as negative
control. Spots were developed using the BD mouse IFN-g kit according to the
manufacturers instructions, and the number of spots was measured using an
Immunospot Series 3 Analyzer and analyzed using ImmunoSpot software
(Cellular Technology). For SIY-pentamer staining, splenocytes were preincubated for 15 min with anti-CD16/32 monoclonal antibody (93) to block potential
nonspecic binding and labeled with PE-MHC class I pentamer (Proimmune)
consisting of murine H-2Kb complexed to SIYRYYGL (SIY) peptide, antiTCRb-AF700 (H57-597), anti-CD8-Pacic Blue (53-6.7), anti-CD4-Pacic
Orange (RM4-5) (all antibodies from BioLegend), and the Fixable Viability
Dye eFluor 450 (eBioscience). Stained cells were analyzed using LSR II cytometer with FACSDiva software (BD Biosciences). Data analysis was conducted
with FlowJo software (Tree Star).
Luciferase Assay
104 HEK293T cells were seeded in 96-well plates and transiently transfected
(Lipofectamine 2000) with human IFN-b rey reporter plasmid (Fitzgerald
et al., 2003) and TK-Renilla luciferase reporter for normalization. The following
day, cells were stimulated with 10 mM of each CDN or 100 mg/ml DMXAA using
digitonin permeabilization (50 mM HEPES, 100 mM KCL, 3 mM MgCl2, 0.1 mM
DTT, 85 mM sucrose, 0.2% BSA, 1 mM ATP, 0.1 mM GTP, and 10 mg/ml digitonin) to ensure uniform uptake. After 20 min, stimulation mixtures were
removed and normal media was added. After a total of 6 hr, cell lysates
were prepared and reporter gene activity measured using the Dual Luciferase
Assay System (Promega) on a Spectramax M3 luminometer.
Statistical Analysis
Students paired t test was used to calculate two-tailed p values to estimate
statistical signicance of differences between two treatment groups using
Prism 6 software. The number of mice per group and the statistically signicant
p values are labeled in the gures and/or legends with asterisks.

E.L, T.B., J.J.L., K.M., and T.W.D. are all paid employees of Aduro BioTech,
hold stock in the company, and may be inventors on patent applications
that apply to the CDN molecules described in the manuscript.
Received: November 26, 2014
Revised: February 24, 2015
Accepted: April 14, 2015
Published: May 7, 2015
REFERENCES
Ablasser, A., Goldeck, M., Cavlar, T., Deimling, T., Witte, G., Rohl, I., Hopfner,
K.P., Ludwig, J., and Hornung, V. (2013). cGAS produces a 20 -50 -linked cyclic
dinucleotide second messenger that activates STING. Nature 498, 380384.
Baguley, B.C., and Ching, L.-M. (1997). Immunomodulatory actions of xanthenone anticancer agents. BioDrugs 8, 119127.
Blank, C., Brown, I., Peterson, A.C., Spiotto, M., Iwai, Y., Honjo, T., and Gajewski, T.F. (2004). PD-L1/B7H-1 inhibits the effector phase of tumor rejection by
T cell receptor (TCR) transgenic CD8+ T cells. Cancer Res. 64, 11401145.
Brody, J.D., Ai, W.Z., Czerwinski, D.K., Torchia, J.A., Levy, M., Advani, R.H.,
Kim, Y.H., Hoppe, R.T., Knox, S.J., Shin, L.K., et al. (2010). In situ vaccination
with a TLR9 agonist induces systemic lymphoma regression: a phase I/II study.
J. Clin. Oncol. 28, 43244332.
Burdette, D.L., and Vance, R.E. (2013). STING and the innate immune
response to nucleic acids in the cytosol. Nat. Immunol. 14, 1926.
Burdette, D.L., Monroe, K.M., Sotelo-Troha, K., Iwig, J.S., Eckert, B., Hyodo,
M., Hayakawa, Y., and Vance, R.E. (2011). STING is a direct innate immune
sensor of cyclic di-GMP. Nature 478, 515518.
Cavlar, T., Deimling, T., Ablasser, A., Hopfner, K.P., and Hornung, V. (2013).
Species-specic detection of the antiviral small-molecule compound CMA
by STING. EMBO J. 32, 14401450.
Chandra, D., Quispe-Tintaya, W., Jahangir, A., Asafu-Adjei, D., Ramos, I., Sintim, H.O., Zhou, J., Hayakawa, Y., Karaolis, D.K., and Gravekamp, C. (2014).
STING ligand c-di-GMP improves cancer vaccination against metastatic
breast cancer. Cancer Immunol Res 2, 901910.
Conlon, J., Burdette, D.L., Sharma, S., Bhat, N., Thompson, M., Jiang, Z., Rathinam, V.A.K., Monks, B., Jin, T., Xiao, T.S., et al. (2013). Mouse, but not human STING, binds and signals in response to the vascular disrupting agent
5,6-dimethylxanthenone-4-acetic acid. J. Immunol. 190, 52165225.

SUPPLEMENTAL INFORMATION

Deng, L., Liang, H., Xu, M., Yang, X., Burnette, B., Arina, A., Li, X.-D., Mauceri,
H., Beckett, M., Darga, T., et al. (2014). STING-dependent cytosolic DNA
sensing promotes radiation-induced type I interferon-dependent antitumor immunity in immunogenic tumors. Immunity 41, 843852.

Supplemental Information includes Supplemental Experimental Procedures,

ve gures, and one table and can be found with this article online at http://
dx.doi.org/10.1016/j.celrep.2015.04.031.

Diamond, M.S., Kinder, M., Matsushita, H., Mashayekhi, M., Dunn, G.P., Archambault, J.M., Lee, H., Arthur, C.D., White, J.M., Kalinke, U., et al. (2011). Type
I interferon is selectively required by dendritic cells for immune rejection of tumors. J. Exp. Med. 208, 19892003.

AUTHOR CONTRIBUTIONS
L.C., L.H.G., T.W.D., and T.F.G. conceived the research and conducted the
experiments with S.M.M., D.B.K., K.E.S., E.L., J.J.L., G.E.K., T.B., S.-R.W.,
and K.M. T.W.D., and T.F.G. conceived and supervised the entire project
and wrote and revised the manuscript with L.C., L.H.G., K.E.S., and S.M.M.
ACKNOWLEDGMENTS
We thank Meredith Leong, Pete Lauer, and Russell Vance for discussions;
Jake Bruml, Ryan Duggan, Michael Y.K. Leung, and Diana Ranoa for technical
assistance; and Elie Diner for retroviral constructs. We thank Hector Nolla at
the UC Berkeley Cancer Research Laboratory (CRL) for help with cell sorting.
L.C. was supported by a post-doctoral fellowship from the Ramon Areces
Foundation. This work was supported by P01 CA97296 and R01CA181160
from the National Cancer Institute. L.H.G., S.M.M., D.B.K., K.E.S., G.E.K.,

1028 Cell Reports 11, 10181030, May 19, 2015 2015 The Authors

Diner, E.J., Burdette, D.L., Wilson, S.C., Monroe, K.M., Kellenberger, C.A., Hyodo, M., Hayakawa, Y., Hammond, M.C., and Vance, R.E. (2013). The innate
immune DNA sensor cGAS produces a noncanonical cyclic dinucleotide that
activates human STING. Cell Rep. 3, 13551361.
Dubensky, T.W., Jr., Kanne, D.B., and Leong, M.L. (2013). Rationale, progress
and development of vaccines utilizing STING-activating cyclic dinucleotide
adjuvants. Ther Adv Vaccines 1, 131143.
Ebensen, T., Schulze, K., Riese, P., Link, C., Morr, M., and Guzman, C.A.
(2007a). The bacterial second messenger cyclic diGMP exhibits potent adjuvant properties. Vaccine 25, 14641469.
Ebensen, T., Schulze, K., Riese, P., Morr, M., and Guzman, C.A. (2007b). The
bacterial second messenger cdiGMP exhibits promising activity as a mucosal
adjuvant. Clin. Vaccine Immunol. 14, 952958.
Ebensen, T., Libanova, R., Schulze, K., Yevsa, T., Morr, M., and Guzman, C.A.
(2011). Bis-(30 ,50 )-cyclic dimeric adenosine monophosphate: strong Th1/Th2/
Th17 promoting mucosal adjuvant. Vaccine 29, 52105220.

Fitzgerald, K.A., McWhirter, S.M., Faia, K.L., Rowe, D.C., Latz, E., Golenbock,
D.T., Coyle, A.J., Liao, S.M., and Maniatis, T. (2003). IKKepsilon and TBK1 are
essential components of the IRF3 signaling pathway. Nat. Immunol. 4,
491496.
Fuertes, M.B., Kacha, A.K., Kline, J., Woo, S.R., Kranz, D.M., Murphy, K.M.,
and Gajewski, T.F. (2011). Host type I IFN signals are required for antitumor
CD8+ T cell responses through CD8alpha+ dendritic cells. J. Exp. Med. 208,
20052016.
Gaffney, B.L., Veliath, E., Zhao, J., and Jones, R.A. (2010). One-ask syntheses of c-di-GMP and the [Rp,Rp] and [Rp,Sp] thiophosphate analogues. Org.
Lett. 12, 32693271.
Gajewski, T.F., Woo, S.R., Zha, Y., Spaapen, R., Zheng, Y., Corrales, L., and
Spranger, S. (2013). Cancer immunotherapy strategies based on overcoming
barriers within the tumor microenvironment. Curr. Opin. Immunol. 25, 268
276.
Galon, J., Page`s, F., Marincola, F.M., Angell, H.K., Thurin, M., Lugli, A., Zlobec,
I., Berger, A., Bifulco, C., Botti, G., et al. (2012). Cancer classication using the
Immunoscore: a worldwide task force. J. Transl. Med. 10, 205.
Gao, P., Ascano, M., Zillinger, T., Wang, W., Dai, P., Serganov, A.A., Gaffney,
B.L., Shuman, S., Jones, R.A., Deng, L., et al. (2013a). Structure-function analysis of STING activation by c[G(20 ,50 )pA(30 ,50 )p] and targeting by antiviral
DMXAA. Cell 154, 748762.
Gao, P., Ascano, M., Wu, Y., Barchet, W., Gaffney, B.L., Zillinger, T., Serganov,
A.A., Liu, Y., Jones, R.A., Hartmann, G., et al. (2013b). Cyclic [G(20 ,50 )pA(30 ,50 )
p] is the metazoan second messenger produced by DNA-activated cyclic
GMP-AMP synthase. Cell 153, 10941107.
Goins, W.F., Huang, S., Cohen, J.B., and Glorioso, J.C. (2014). Engineering
HSV-1 vectors for gene therapy. Methods Mol. Biol. 1144, 6379.
Gray, P.M., Forrest, G., Wisniewski, T., Porter, G., Freed, D.C., DeMartino,
J.A., Zaller, D.M., Guo, Z., Leone, J., Fu, T.M., and Vora, K.A. (2012). Evidence
for cyclic diguanylate as a vaccine adjuvant with novel immunostimulatory activities. Cell. Immunol. 278, 113119.
Harlin, H., Meng, Y., Peterson, A.C., Zha, Y., Tretiakova, M., Slingluff, C.,
McKee, M., and Gajewski, T.F. (2009). Chemokine expression in melanoma
metastases associated with CD8+ T-cell recruitment. Cancer Res. 69, 3077
3085.
Hildner, K., Edelson, B.T., Purtha, W.E., Diamond, M., Matsushita, H., Kohyama, M., Calderon, B., Schraml, B.U., Unanue, E.R., Diamond, M.S., et al.
(2008). Batf3 deciency reveals a critical role for CD8alpha+ dendritic cells
in cytotoxic T cell immunity. Science 322, 10971100.
Ishikawa, H., and Barber, G.N. (2008). STING is an endoplasmic reticulum
adaptor that facilitates innate immune signalling. Nature 455, 674678.
Ishikawa, H., Ma, Z., and Barber, G.N. (2009). STING regulates intracellular
DNA-mediated, type I interferon-dependent innate immunity. Nature 461,
788792.
Jassar, A.S., Suzuki, E., Kapoor, V., Sun, J., Silverberg, M.B., Cheung, L., Burdick, M.D., Strieter, R.M., Ching, L.M., Kaiser, L.R., and Albelda, S.M. (2005).
Activation of tumor-associated macrophages by the vascular disrupting agent
5,6-dimethylxanthenone-4-acetic acid induces an effective CD8+ T-cell-mediated antitumor immune response in murine models of lung cancer and mesothelioma. Cancer Res. 65, 1175211761.
Jin, L., Xu, L.G., Yang, I.V., Davidson, E.J., Schwartz, D.A., Wurfel, M.M., and
Cambier, J.C. (2011). Identication and characterization of a loss-of-function
human MPYS variant. Genes Immun. 12, 263269.
Joseph, W.R., Cao, Z., Mountjoy, K.G., Marshall, E.S., Baguley, B.C., and
Ching, L.M. (1999). Stimulation of tumors to synthesize tumor necrosis factor-alpha in situ using 5,6-dimethylxanthenone-4-acetic acid: a novel
approach to cancer therapy. Cancer Res. 59, 633638.
Kawarada, Y., Ganss, R., Garbi, N., Sacher, T., Arnold, B., and Hammerling, G.J.
(2001). NK- and CD8(+) T cell-mediated eradication of established tumors by
peritumoral injection of CpG-containing oligodeoxynucleotides. J. Immunol.
167, 52475253.

Kim, S., Li, L., Maliga, Z., Yin, Q., Wu, H., and Mitchison, T.J. (2013). Anticancer
avonoids are mouse-selective STING agonists. ACS Chem. Biol. 8, 1396
1401.
Konno, H., Konno, K., and Barber, G.N. (2013). Cyclic dinucleotides trigger
ULK1 (ATG1) phosphorylation of STING to prevent sustained innate immune
signaling. Cell 155, 688698.
Lara, P.N., Jr., Douillard, J.Y., Nakagawa, K., von Pawel, J., McKeage, M.J.,
Albert, I., Losonczy, G., Reck, M., Heo, D.S., Fan, X., et al. (2011). Randomized
phase III placebo-controlled trial of carboplatin and paclitaxel with or without
the vascular disrupting agent vadimezan (ASA404) in advanced non-small-cell
lung cancer. J. Clin. Oncol. 29, 29652971.
Li, L., Yin, Q., Kuss, P., Maliga, Z., Millan, J.L., Wu, H., and Mitchison, T.J.
(2014). Hydrolysis of 230 -cGAMP by ENPP1 and design of nonhydrolyzable
analogs. Nat. Chem. Biol. 10, 10431048.
McWhirter, S.M., Barbalat, R., Monroe, K.M., Fontana, M.F., Hyodo, M.,
Joncker, N.T., Ishii, K.J., Akira, S., Colonna, M., Chen, Z.J., et al. (2009). A
host type I interferon response is induced by cytosolic sensing of the bacterial
second messenger cyclic-di-GMP. J. Exp. Med. 206, 18991911.
Niesen, F.H., Berglund, H., and Vedadi, M. (2007). The use of differential scanning uorimetry to detect ligand interactions that promote protein stability.
Nat. Protoc. 2, 22122221.
Postow, M.A., Harding, J., and Wolchok, J.D. (2012). Targeting immune
checkpoints: releasing the restraints on anti-tumor immunity for patients
with melanoma. Cancer J. 18, 153159.
Prantner, D., Perkins, D.J., Lai, W., Williams, M.S., Sharma, S., Fitzgerald, K.A.,
and Vogel, S.N. (2012). 5,6-Dimethylxanthenone-4-acetic acid (DMXAA) activates stimulator of interferon gene (STING)-dependent innate immune pathways and is regulated by mitochondrial membrane potential. J. Biol. Chem.
287, 3977639788.
Roberson, S.M., and Walker, W.S. (1988). Immortalization of cloned mouse
splenic macrophages with a retrovirus containing the v-raf/mil and v-myc oncogenes. Cell. Immunol. 116, 341351.
Romling, U., Galperin, M.Y., and Gomelsky, M. (2013). Cyclic di-GMP: the rst
25 years of a universal bacterial second messenger. Microbiol. Mol. Biol. Rev.
77, 152.
Sauer, J.D., Sotelo-Troha, K., von Moltke, J., Monroe, K.M., Rae, C.S., Brubaker, S.W., Hyodo, M., Hayakawa, Y., Woodward, J.J., Portnoy, D.A., and
Vance, R.E. (2011). The N-ethyl-N-nitrosourea-induced Goldenticket mouse
mutant reveals an essential function of Sting in the in vivo interferon response
to Listeria monocytogenes and cyclic dinucleotides. Infect. Immun. 79, 688
694.
Slansky, J.E., Rattis, F.M., Boyd, L.F., Fahmy, T., Jaffee, E.M., Schneck, J.P.,
Margulies, D.H., and Pardoll, D.M. (2000). Enhanced antigen-specic antitumor immunity with altered peptide ligands that stabilize the MHC-peptideTCR complex. Immunity 13, 529538.
Sun, L., Wu, J., Du, F., Chen, X., and Chen, Z.J. (2013). Cyclic GMP-AMP synthase is a cytosolic DNA sensor that activates the type I interferon pathway.
Science 339, 786791.
Wallace, A., LaRosa, D.F., Kapoor, V., Sun, J., Cheng, G., Jassar, A., Blouin,
A., Ching, L.M., and Albelda, S.M. (2007). The vascular disrupting agent,
DMXAA, directly activates dendritic cells through a MyD88-independent
mechanism and generates antitumor cytotoxic T lymphocytes. Cancer Res.
67, 70117019.
Wolchok, J.D., Kluger, H., Callahan, M.K., Postow, M.A., Rizvi, N.A., Lesokhin,
A.M., Segal, N.H., Ariyan, C.E., Gordon, R.A., Reed, K., et al. (2013). Nivolumab plus ipilimumab in advanced melanoma. N. Engl. J. Med. 369, 122133.
Woo, S.R., Fuertes, M.B., Corrales, L., Spranger, S., Furdyna, M.J., Leung,
M.Y., Duggan, R., Wang, Y., Barber, G.N., Fitzgerald, K.A., et al. (2014).
STING-dependent cytosolic DNA sensing mediates innate immune recognition of immunogenic tumors. Immunity 41, 830842.
Woodward, J.J., Iavarone, A.T., and Portnoy, D.A. (2010). c-di-AMP secreted
by intracellular Listeria monocytogenes activates a host type I interferon
response. Science 328, 17031705.

Cell Reports 11, 10181030, May 19, 2015 2015 The Authors 1029

Yan, H., Wang, X., KuoLee, R., and Chen, W. (2008). Synthesis and immunostimulatory properties of the phosphorothioate analogues of cdiGMP. Bioorg.
Med. Chem. Lett. 18, 56315634.

Zhang, X., Shi, H., Wu, J., Zhang, X., Sun, L., Chen, C., and Chen, Z.J. (2013).
Cyclic GMP-AMP containing mixed phosphodiester linkages is an endogenous high-afnity ligand for STING. Mol. Cell 51, 226235.

Yi, G., Brendel, V.P., Shu, C., Li, P., Palanathan, S., and Cheng Kao, C. (2013).
Single nucleotide polymorphisms of human STING can affect innate immune
response to cyclic dinucleotides. PLoS ONE 8, e77846.

Zhao, L., Ching, L.M., Kestell, P., and Baguley, B.C. (2002). The antitumour activity of 5,6-dimethylxanthenone-4-acetic acid (DMXAA) in TNF receptor-1
knockout mice. Br. J. Cancer 87, 465470.

1030 Cell Reports 11, 10181030, May 19, 2015 2015 The Authors

Cell Reports

Article
Manipulation of the Quorum Sensing Signal AI-2
Affects the Antibiotic-Treated Gut Microbiota
Jessica Ann Thompson,1,5 Rita Almeida Oliveira,1,5 Ana Djukovic,2 Carles Ubeda,2,3 and Karina Bivar Xavier1,4,*

^ncia, 2780-156 Oeiras, Portugal

Gulbenkian de Cie
de Genomica y Salud, Centro Superior de Investigacion en Salud Publica, FISABIO, Valencia 46020, Spain
3Centers of Biomedical Research Network (CIBER) in Epidemiology and Public Health, Madrid 28029, Spain
4Instituto de Tecnologia Qumica e Biolo
gica, Universidade Nova de Lisboa, 2780-157 Oeiras, Portugal
5Co-rst author
*Correspondence: [email protected]
http://dx.doi.org/10.1016/j.celrep.2015.02.049
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
1Instituto

2Departamento

SUMMARY

The mammalian gut microbiota harbors a diverse

ecosystem where hundreds of bacterial species
interact with each other and their host. Given that bacteria use signals to communicate and regulate group
behaviors (quorum sensing), we asked whether such
communication between different commensal species can inuence the interactions occurring in this
environment. We engineered the enteric bacterium,
Escherichia coli, to manipulate the levels of the interspecies quorum sensing signal, autoinducer-2 (AI-2),
in the mouse intestine and investigated the effect
upon antibiotic-induced gut microbiota dysbiosis.
E. coli that increased intestinal AI-2 levels altered
the composition of the antibiotic-treated gut microbiota, favoring the expansion of the Firmicutes
phylum. This signicantly increased the Firmicutes/
Bacteroidetes ratio, to oppose the strong effect of
the antibiotic, which had almost cleared the Firmicutes. This demonstrates that AI-2 levels inuence
the abundance of the major phyla of the gut microbiota, the balance of which is known to inuence
human health.
INTRODUCTION
The mammalian intestinal tract is home to approximately 1014
bacteria, encoding over 100-fold more genes than are within
the human genome (Savage, 1977). This complement to the
hosts coding capacity provides a repertoire of additional
metabolic functions including the digestion of complex polysaccharides, production of fatty acids, and vitamin biosynthesis
(Arumugam et al., 2011; Louis et al., 2014). Interactions between
commensal species and the host fulll immensely important
physiological roles, promoting the development of the intestinal
tract, maturation of the immune system, and immunological
tolerance to antigens (Berg, 1996; Hooper et al., 2012; RakoffNahoum et al., 2004). These bacteria also provide a major protective barrier against pathogens, through a phenomenon known

as colonization resistance (Bohnhoff and Miller, 1962; Lawley

and Walker, 2013).
Though the gut microbiota is clearly benecial in many ways,
imbalances in this community (dysbiosis), including those
induced by diet or antimicrobial usage, can pose a threat to
host health. Antibiotics such as clindamycin or ampicillin that
alter the commensal bacterial community also increase host
susceptibility to opportunists such as Clostridium difcile and
vancomycin-resistant enterococci (Bufe et al., 2012; Ubeda
et al., 2010). Furthermore, shifts in the composition of the microbiota, particularly those involving the two predominant phyla in
the mammalian gut, the Bacteroidetes and the Firmicutes, are
associated with the pathogenesis of obesity, diabetes, chronic
inammatory bowel diseases, and gastrointestinal cancer, as
well as autism and stress (Finegold et al., 2010; Frank et al.,
2007; Galley et al., 2014; Ley et al., 2006; Qin et al., 2012; Turnbaugh et al., 2006; Wang et al., 2012). Consequently, the ability
to drive this community from disease-associated to healthy
states, by manipulating the native signals and interactions that
occur between its members, to restore colonization resistance,
for example, offers great potential for therapeutic benet.
The mechanisms through which the resident microbial community inhibits the growth of invading microbes remain largely
unknown, but there is increasing evidence that direct microbemicrobe interactions play a critical role in this process (Bufe
et al., 2015; Hsiao et al., 2014; Kamada et al., 2012, 2013; Ng
et al., 2013; Reeves et al., 2012). Cross-feeding and metabolic
interactions clearly inuence the composition of the microbiota
(Bufe et al., 2015; Fabich et al., 2008; Flint et al., 2007; Kamada
et al., 2012; Leatham et al., 2009; Ng et al., 2013). Bacteria also
harbor vast arrays of mechanisms to sense and respond to
features of their environment, including the presence of other
bacteria. Small diffusible molecules known as autoinducers are
synthesized and released in accordance with cell number; their
subsequent detection enables bacteria to synchronously regulate behaviors at the population level in a process known as
quorum sensing (Rutherford and Bassler, 2012). Attachment,
biolm formation, motility, and virulence are among the many
phenotypes controlled in this manner. As quorum sensing
and the behaviors it regulates are important in many bacteriabacteria interactions in the host context, both symbiotic and
pathogenic (Ruby, 2008; Rutherford and Bassler, 2012), this

Cell Reports 10, 18611871, March 24, 2015 2015 The Authors 1861

phenomenon is likely to also contribute to the interactions between bacteria inhabiting the mammalian gut.
Many quorum sensing signals are species specic; however,
production of and responses to one molecule, autoinducer-2
(AI-2), are observed throughout the bacterial kingdom (Chen
et al., 2002; Miller et al., 2004; Pereira et al., 2013). As AI-2 produced by one species can inuence gene expression in another,
this signal can foster interspecies communication and enable
bacteria to modify behaviors such as virulence, luminescence,
and biolm formation across different species (Armbruster
et al., 2010; Cuadra-Saenz et al., 2012; Pereira et al., 2008; Xavier and Bassler, 2005a). This feature makes AI-2 an excellent
candidate for mediating cell-cell interactions in the mammalian
gut, where hundreds of bacterial species co-exist and interact.
Multiple gut-associated bacteria that encode the AI-2 synthase,
LuxS, or produce AI-2 have already been identied (Antunes
et al., 2005; Hsiao et al., 2014; Lukas et al., 2008; Schauder
et al., 2001). Furthermore, the human commensal bacterium
Ruminococcus obeum was recently reported to inhibit colonization of the mouse gut by V. cholerae, partially through AI-2
signaling, highlighting how AI-2 produced by commensal bacteria can affect invading pathogens (Hsiao et al., 2014). We hypothesized that signaling through AI-2 might also occur between
commensal members of the microbiota and asked whether
AI-2 can shape the species composition of this community under
conditions of dysbiosis.
To investigate this hypothesis, we made use of the natural
signal production and depletion capabilities of the enteric bacterium, Escherichia coli. This organism secretes large amounts of
AI-2 into the environment; it also harbors a highly efcient mechanism for signal uptake and degradation, known as the Lsr transport system (Pereira et al., 2012; Xavier and Bassler, 2005b). By
internalizing and processing AI-2 produced by itself as well as
from other species, Lsr system-expressing bacteria can disrupt
the ability of neighboring species to correctly determine population density and regulate AI-2-dependent behavior appropriately, as shown in vitro in mixed cultures of E. coli and Vibrio
spp. (Xavier and Bassler, 2005a). As a result, this system has
been explored as a potential means for AI-2 interspecies quorum
quenching (Roy et al., 2010). Given that E. coli can also stably
colonize the mouse gut following streptomycin treatment (Conway et al., 2004), we used this bacterium as a tool to manipulate
AI-2 signaling in vivo and demonstrated the accumulation and
depletion of AI-2 in the intestinal tract of gnotobiotic mice. Streptomycin induces gut dysbiosis and has been used extensively to
study Salmonella pathogenesis and E. coli physiology following
the disruption of colonization resistance (Barthel et al., 2003;
Spees et al., 2013). We characterized the changes induced by
streptomycin upon the composition of the microbiota and then
determined the effect of E. coli-mediated AI-2 manipulation
upon this antibiotic-treated community.
RESULTS
Engineered E. coli Mutants Manipulate AI-2 Availability
in the Mouse Gut
To manipulate AI-2 levels in the mouse gut, we constructed a
combination of E. coli mutants affected in the Lsr system that

accumulate different levels of AI-2 in vitro. To obtain high levels

of AI-2, we constructed a mutant in lsrK, which encodes the
signal kinase required for phosphorylation and intracellular
retention of the signal. In the absence of LsrK, E. coli cannot
sequester nor degrade AI-2 intracellularly, so the molecule
accumulates extracellularly. To deplete environmental AI-2, we
deleted lsrR, as it encodes a repressor of the Lsr transport
system. In this mutant, the Lsr transporter is constitutively expressed at high levels and this strain is highly efcient at internalizing and scavenging environmental AI-2. The signal synthase,
LuxS, was also deleted to produce mutants that do not produce
AI-2 and thus do not contribute to the extracellular pool of AI-2.
These mutations were introduced into a YFP-expressing strain of
E. coli to ease identication of bacteria recovered from mice.
In vitro validation of these strains conrmed that the DlsrK
mutant strain accumulated and maintained high extracellular
AI-2 levels (Figure S1A). Bacteria affected in lsrR rapidly internalized the signal, whereas no AI-2 could be detected in cultures
containing mutants in the AI-2 synthase gene, luxS. We also
conrmed that the DlsrRDluxS mutant strain internalized exogenously supplied AI-2 and could be used as a tool to deplete environmental signal (Figure S1B). No uptake occurred in cultures of
DlsrKDluxS mutant bacteria provided with the signal, conrming
that this strain does not deplete AI-2 in vitro (Figure S1B).
Germ-free C57BL/6J mice were then mono-colonized with
either wild-type (WT), DlsrK, or DlsrRDluxS mutant E. coli to
determine the effect of each strain on AI-2 accumulation, without
interference from other bacteria resident in the intestinal tract
that might also produce or import AI-2. An additional group of
germ-free mice was gavaged with PBS to provide a negative
control. All E. coli strains colonized the mice to the same level,
with approximately 109 colony-forming units (CFU)/g feces recovered 5 days after inoculation (Figure S1C). To determine
the levels of AI-2 in the intestinal tract at this time point, extracts
of the cecal contents were analyzed using a Vibrio harveyi
biosensor that produces light in response to AI-2. Extracts
from WT- and DlsrK-gavaged mice induced almost 25-fold
more light production than cecal extracts from the control
PBS-gavaged germ-free mice (Figure 1), demonstrating that
these strains produce AI-2 in vivo, which accumulates in the
gastrointestinal tract. The similar levels of AI-2 observed in these
extracts, from WT- and DlsrK-colonized mice, suggest that Lsr
transporter expression in the WT may be repressed in the
cecum. This could be the result of inhibition by metabolites present in the gut, as glucose and other compounds such as glycerol are known to negatively regulate expression of this system
(Pereira et al., 2012; Xavier and Bassler, 2005b). No signal was
detected in extracts from mice colonized with DlsrRDluxS, a
non-AI-2-producing mutant E. coli (Figure 1).
To determine whether DlsrRDluxS mutant E. coli could scavenge AI-2 present in the gut, germ-free mice were gavaged
with a 1:1 mix of DlsrK E. coli (to provide AI-2) in combination
with either the DlsrRDluxS mutant (which removes but does
not produce signal in vitro) or the control strain, DlsrKDluxS
(which neither produces nor internalizes signal). As predicted
from our in vitro results, no AI-2 was detected in cecal contents
from mice colonized with the mixture of DlsrK and DlsrRDluxS
mutants, whereas those from mice co-colonized with the DlsrK

1862 Cell Reports 10, 18611871, March 24, 2015 2015 The Authors

Figure 1. E. coli Accumulate and Deplete AI-2 in the Gut of Monocolonized Mice
AI-2 activity in cecal extracts harvested from germ-free mice 5 days after
gavage with PBS, WT, DlsrK or DlsrRDluxS; or a 1:1 mix of DlsrK and
DlsrRDluxS; or DlsrK and DlsrKDluxS E. coli, as measured by Vibrio harveyi
bioluminescence. Data shown are the mean, and the error bars correspond to
SD; n = 3. See also Figure S1.

and DlsrKDluxS mixture clearly contained signal (Figure 1). Total

bacterial loads were similar in both groups of mice (Figure S1C),
and the ratio between mutant bacterial strains remained close to
one (Figure S1D). These data showed that there were no differences in the numbers of AI-2-producing DlsrK bacteria between
the two groups, which could have otherwise explained the
different levels of AI-2 detected. This demonstrates that the
DlsrRDluxS strain can efciently internalize AI-2 in the gut and
thus that deletion of lsrR relieved repression of the Lsr
transporter. In summary, these results show that DlsrK and
DlsrRDluxS mutant E. coli can be used to manipulate AI-2
signaling, either accumulating or scavenging AI-2 in the mouse
gut, respectively, and validate the use of the DlsrKDluxS strain
as a control that does not inuence signal levels.
Streptomycin Induces Major Changes in the Fecal
Microbiota
Streptomycin disrupts the microbiota, inducing a breakdown in
colonization resistance, which enables E. coli to colonize the
gut to high levels (Conway et al., 2004). We reasoned that this
would provide a good system to determine the effect of AI-2
manipulation mediated by E. coli on the composition of the microbiota. In the absence of a detailed metagenomic description
of the dysbiosis caused by prolonged exposure to streptomycin,
we characterized the effect of this antibiotic upon the gut bacterial community during 28 days of treatment without manipulation
of AI-2 levels. Antibiotic treatment decreased bacterial load:
4 days after the initial administration of antibiotic, density had
dropped almost 20-fold from 1.61 3 1010 bacterial 16S rRNA
gene copies/g feces in untreated mice to 8.36 3 108 copies/g

feces (Figure 2A). Despite continued exposure to streptomycin,

the total bacterial load gradually increased after day 4 and
stabilized 5- to 10-fold lower than that observed in untreated
mice.
Streptomycin also induced major changes to the composition
of the microbiota, as shown by high-throughput sequencing of
the 16S rRNA genes amplied from DNA recovered from feces
and analysis of the most abundant taxa prior to and during
treatment (Figure 2B). Before treatment, the microbiota was
composed of many different phylotypes, the majority of which
belonged to the two phyla that commonly predominate among
the gut bacterial community, the Firmicutes and the Bacteroidetes. These constituted 43% and 48%, respectively, of the
bacteria detected (Figure 2C), with other phyla such as the
Proteobacteria, Actinobacteria, and Deferribacteres present at
low abundances, as previously observed in the mammalian gut
(Stecher et al., 2007; Turnbaugh et al., 2010). Many of these
taxa could no longer be detected 28 days into streptomycin
treatment (Figures 2B and 2C), indicating an antibiotic-induced
decrease in diversity of the gut bacterial community. Concurrently, only a few populations expanded. This trend of multiple
losses and few increases was also visible at the lower taxonomic
level when the relative frequencies of the 100 most-abundant
operational taxonomic units (OTUs) (dened with 97% sequence
similarity) were analyzed over time (Figure S2A). Many OTUs had
already decreased in abundance below the level of detection
only 2 days after the onset of treatment and did not subsequently
recover (white boxes, Figure S2A). By day 28, just three OTUs,
belonging to the Bacteroidetes phylum (OTU1, 2, and 4), constituted more than 60% of the bacteria detected in each mouse
analyzed (Figure S3). This apparent decrease in community
diversity was conrmed upon quantication of both the Chao
richness index and the Shannon diversity index. These indexes
both decreased by day 2 of treatment and remained low in streptomycin-treated microbiota when compared to those calculated
from the untreated samples (Figures S2B and S2C). Thus, neither
richness nor diversity of the microbiota recovered in the presence of antibiotic.
At the higher phylogenetic level, these changes drove a major
shift in the balance between the Bacteroidetes and Firmicutes.
From day 2, the Bacteroidetes had reached a relative abundance
of approximately 90% and remained at this level during the treatment (Figure 2D). In contrast, the Firmicutes decreased hugely
and only represented 0.7% of the bacterial community by the
end of the experiment (Figures 2C and 2D). Despite uctuations
in the relative abundance of some phylotypes of the Proteobacteria (Figure S3, yellow), no signicant difference was seen in the
prevalence of this phylum.
Though streptomycin induced consistent changes in the ratio
of Firmicutes to Bacteroidetes in all mice, its effect varied greatly
at the individual OTU level during the early stages of antibiotic
treatment. This is shown by the distinct changes in presence
and abundance of specic OTUs in the different animals (Figures
S2A and S3), which appeared to stabilize by day 28. The Jaccard
distance, which provides a measure of dissimilarity taking into
account the presence and absence of OTUs in the microbiota
of each mouse, conrmed the apparent increase in inter-individual variation between the communities on days 4, 7, and 12 of

Cell Reports 10, 18611871, March 24, 2015 2015 The Authors 1863

Figure 2. Streptomycin Changes Intestinal Microbiota Load and Composition

Streptomycin was given to four non-littermate mice in drinking water for 28 days. Animals were housed in separate cages, and fecal samples were collected for
microbiota analysis prior to and 2, 4, 7, 12, and 28 days into streptomycin treatment.
(A) Total microbiota load measured from DNA by qPCR of the 16S rRNA gene copy number/g feces.
(B) Intestinal microbiota composition at the time points indicated. Each stacked bar represents the mean of the most abundant bacterial taxa in all the mice. The
colored segments represent the relative fraction of each bacterial taxon.
(C) Relative abundance of the major phyla found in the gut microbiota before and after 28 days of treatment. Data shown are the mean the SD.
(D) Relative abundance of the Bacteroidetes (red circles) and Firmicutes (blue circles).
(E) Phylogenetic dissimilarities between microbial communities on each day determined by the mean unweighted Unifrac distance of the bacterial communities of
each mouse versus all other mice.
Data shown in (A), (D), and (E) are the median, and the error bars show the interquartile range. Data were analyzed with the paired Students t test (*p < 0.05). n = 4
except for day 2, where n = 3. See also Figures S2 and S3.

antibiotic treatment, followed by a decrease on day 28 (Figure S2D). This variability also affected the phylogenetic distance
of the communities, as the unweighted UniFrac distance similarly increased after exposure to streptomycin and then
lowered in the bacterial communities present on day 28 (Fig-

ure 2E). Distance measurements were in fact smaller for the

microbiota after 28 days of streptomycin treatment than when
untreated, showing a reduction in mouse-mouse variability
upon prolonged antibiotic treatment (Figures 2E and S2D). These
results demonstrate a highly variable effect of streptomycin upon

1864 Cell Reports 10, 18611871, March 24, 2015 2015 The Authors

Figure 3. E. coli Colonization Levels and Total Microbiota Load in

Streptomycin-Treated Mice
Individually housed mice were given streptomycin in drinking water and 2 days
into treatment were colonized with the different E. coli mutants as specied.
Fecal samples were collected at time points indicated.
(A) Loads of E. coli mutant strains as CFU/g feces from mice gavaged with the
different E. coli mutants.
(B) Total microbiota (gray bars) and E. coli mutants (white bars) loads from day
28, determined by qPCR of 16S rRNA or yfp gene copy number/g feces from
DNA extracted from fecal samples from mice from each of the different groups.
Data shown are the median, and the error bars show the interquartile range;
n = 9 per group.

the gut-associated bacteria within different mice during the early

phases of antibiotic treatment that became more consistent
across individuals by day 28. As the overall effect of the antibiotic
was reproducible at this stage, this time point was selected for
subsequent analysis of the inuence of AI-2 on the bacterial
community that emerges as a consequence of streptomycininduced dysbiosis.
AI-2 Produced by E. coli Increases the Ratio of
Firmicutes to Bacteroidetes during
Streptomycin-Induced Dysbiosis
To determine whether AI-2 inuences the species composition
of the gut microbial community established during long-term

streptomycin treatment, individually housed mice were gavaged

2 days after the start of antibiotic treatment with either DlsrK,
DlsrRDluxS, or DlsrKDluxS YFP-expressing streptomycin-resistant E. coli. All three strains colonized the mice to similar levels,
with approximately 108 CFU/g feces recovered throughout the
experiment (Figure 3A). qPCR conrmed that E. coli colonization
levels, and also the total bacterial load of the microbiota, were
the same in all the three groups of mice 28 days after the start
of antibiotic treatment (which corresponded to 26 days of
E. coli colonization; Figure 3B).
Metagenomic analysis of the microbiota in fecal samples from
day 28 of treatment using PCoA of the Jaccard index (Figure 4A)
revealed that the structure of the microbiota in mice colonized
with DlsrK mutant bacteria was signicantly different from those
in mice colonized with either of the other two groups (AMOVA
test; p = 0.007, DlsrK versus DlsrRDluxS; p = 0.006, DlsrK versus
DlsrKDluxS). No signicant difference was observed between
the communities of DlsrRDluxS- and DlsrKDluxS-colonized
mice (p = 0.572). PCoA of the unweighted Unifrac showed the
same trend, as the structure of the microbiota was also signicantly different in terms of phylogeny between mice colonized
by DlsrK and DlsrRDluxS or DlsrKDluxS mutant E. coli (AMOVA;
p = 0.021 and p = 0.019, respectively; Figure 4B). This was due to
the changes in relative abundance of multiple OTUs in DlsrK
mutant-colonized mice compared to those in the mice containing either of the other two E. coli strains. The abundances of
six OTUs were signicantly different when compared between
mice colonized by DlsrK and those containing DlsrKDluxS or
the DlsrRDluxS bacteria (shown in Figures 4C4H); other OTUs
showed a similar trend for both comparisons but signicant differences for only one of these cases (Figure S4). Some of the
observed changes were considerable: OTU2, a member of the
Bacteroidales order present at very high abundance on day 28,
more than halved in frequency from 20.5% to 7.3% in
the DlsrKDluxS- and DlsrK-colonized mice, respectively (Figure S4A). No signicant differences were observed when the
abundances of the OTUs detected in mice colonized with either
DlsrKDluxS or DlsrRDluxS were compared.
Most OTUs that changed in abundance were less prevalent
in the presence of DlsrK mutant E. coli than when in the presence of either of the other two mutants (red colors, Figure 4I).
These OTUs were all members of the Bacteroidetes phylum
and included two from the Bacteroidales order and a member
of the Porphyromonadaceae family. In contrast, one OTU classied as a Lachnospiraceae (a family within the Clostridiales
order of Firmicutes) was increased in the presence of DlsrK
mutant bacteria compared to both of the other groups (Figure 4C). Another member of the Lachnospiraceae family and
an OTU correlating to the Parasutterella genus (classied within
Burkholderiales order of Proteobacteria) were also found
at signicantly higher frequencies in the fecal microbiota of
DlsrK-colonized mice compared to those harboring DlsrKDluxS
mutant bacteria (Figures S4B and S4D). These changes in
abundance of multiple OTUs when in the presence of AI-2-producing (DlsrK) bacteria combined to give a signicant effect
at the phylum level (Figure 5). The Bacteroidetes, which again
dominated the microbiota, were signicantly lower in abundance in the DlsrK mutant-containing mice than in those

Cell Reports 10, 18611871, March 24, 2015 2015 The Authors 1865

Figure 4. Microbiota Composition of Streptomycin-Treated Mice Differs in the Presence of DlsrK Mutant E. coli
Intestinal microbiota composition was analyzed in samples collected from the same mice as in Figure 3 after 28 days of antibiotic exposure (corresponding to
26 days of E. coli colonization; n = 9 per group).
(A and B) Analysis of the overall microbiota of the different E. coli groups by PCoA plots of Jaccard and unweighted Unifrac distances, respectively. The rst two
coordinates are shown. Each group is labeled with a different color, as indicated. Ellipses centered on the categorical averages of the metric distances with a 95%
condence interval for the rst two coordinates of each group were drawn on the associated PCoA.
(CH) Relative abundance of individual OTUs that exhibited a signicant difference between the group colonized with DlsrK and both the other two groups are
shown. Data shown are the median, and error bars show the interquartile range. Data were analyzed with the Wilcoxon test using the Benjamini-Hochberg
correction (*q < 0.1; **q < 0.05; ns, not signicant).
(I) Heatmap showing the fold change in relative abundance of OTUs in mice colonized with DlsrK mutant bacteria divided by the mean of the same OTU in mice
colonized with either the DlsrRDluxS or the control group, DlsrKDluxS, E. coli. All the OTUs that exhibited a signicant difference between the different groups are
shown.
See also Figure S4.

colonized with either DlsrRDluxS or DlsrKDluxS bacteria (Figure 5A). In contrast, the Firmicutes were positively affected
by the presence of AI-2-producing DlsrK mutant bacteria, un-

dergoing 3- to 6-fold increase to 0.8% from a relative abundance of 0.24% and 0.13% in mice colonized with DlsrKDluxS
or DlsrRDluxS mutant bacteria, respectively (Figure 5B). This

1866 Cell Reports 10, 18611871, March 24, 2015 2015 The Authors

corresponding to these two phyla, respectively, encode homologs to the AI-2 synthase (Figure 6).
In conclusion, the presence of DlsrK mutant bacteria, which
accumulate high levels of AI-2, changed the abundance of the
Bacteroidetes and Firmicutes. The latter group were favored
despite the continued presence of streptomycin, providing
evidence that the bacterial communication molecule, AI-2, can
modulate the composition of gut microbiota.

DISCUSSION

Figure 5. Colonization with DlsrK Mutant Bacteria Changes the

Relative Abundance of the Major Phyla
(A and B) Relative abundances of the (A) Bacteroidetes and (B) Firmicutes
phyla in streptomycin-treated mice colonized with either DlsrKDluxS, DlsrK, or
DlsrRDluxS mutant E. coli (samples were collected 28 days into streptomycin
treatment).
(C) The corresponding ratio of the relative abundances of Firmicutes to
Bacteroidetes from the data described above.
Data shown are the median, and the error bars show the interquartile range
of n = 9 per group. Data were analyzed with the Wilcoxon test using the
Benjamini-Hochberg correction (**q < 0.05; ***q < 0.01; ns, not signicant).

resulted in an increase in the ratio of Firmicutes to Bacteroidetes. Though this increase was signicant (Figure 5C), the ratio in these DlsrK mutant-colonized mice remained much lower
than that observed in the microbiota of untreated animals
(0.897). Interestingly, the Bacteroidetes and Firmicutes are predicted to have very different AI-2 production capabilities: 17%
and 83% of currently sequenced genomes (KEGG database)

There is increasing evidence that microbe-microbe interactions

inuence the composition of the gut microbiota and affect the
balance of this community. Our data support this notion by
revealing that the administration of AI-2-producing bacteria
into antibiotic-treated gut microbiota affected the abundance
of multiple phylotypes and changed the overall structure of the
emerging microbial community. Streptomycin caused a drop in
bacterial load consistent with those seen in previous studies
that analyzed single-dose or short-term treatments (Garner
et al., 2009; Sekirov et al., 2008; Stecher et al., 2007). We also
characterized the effect of streptomycin treatment using metagenomics and observed major changes to the bacterial community. Richness and diversity decreased due to a decrease
in abundance, often to below the level of detection, of many
OTUs: 80% of those present in untreated mice were undetectable by day 28 of streptomycin treatment. Such losses are likely
to free niches for colonization and expansion by other members
of this community that are better adapted to growth in the
presence of antibiotics that can take advantage of the lower
abundance of competitors.
Streptomycin is generally thought to create an environment
favorable for growth of the Proteobacteria (Stecher et al.,
2007). Though dose-dependent increases in prevalence of a
bacterial group that included the Bacteroidetes were also previously reported (Sekirov et al., 2008), we had expected to see the
expansion of the resident members of Proteobacteria. However,
we observed only transient expansions of this phylum whereas
the Bacteroidetes dominated, perhaps as a result of our longer
antibiotic treatment (Stecher et al., 2007). Streptomycin resistance is easily acquired by spontaneous mutations: it is possible
that the prevalence of Bacteroidetes was due to a major competitive advantage conferred upon bacteria that were resistant
to streptomycin. Whether pre-existing or acquired during the
course of treatment, this resistance could therefore be a major
cause for the expansion of the OTUs most abundant at the end
of treatment. These OTUs, all members of the Bacteroidetes
phylum, were consistently present across all mice tested
(OTUs 1, 2, and 4; Figure S3). As acquisition of resistance by independent spontaneous mutation events in the same OTUs in
each of the mice during this treatment seems unlikely, this suggests that streptomycin-resistant members were already present in untreated mice and were subsequently selected for on
exposure to antibiotic. Additionally, our data suggest that either
the prevalence of streptomycin resistance or the propensity to
gain it is higher among the Bacteroidetes than the Firmicutes.
Though selection of resistant strains is likely to be a considerable force in shaping the community that emerges during

Cell Reports 10, 18611871, March 24, 2015 2015 The Authors 1867

Figure 6. Higher Prevalence of LuxS Orthologs in the Complete Genomes of Bacteria Belonging to the Firmicutes
Percentage of full genome sequences corresponding to members of the Bacteroidetes and Firmicutes containing protein orthologs of LuxS.

antibiotic treatment, marked differences in relative abundance of

the different phylotypes present at the end of our treatment were
observed. This nding suggests that individual species have
different abilities to exploit the space freed upon antibiotic treatment and that other antagonistic or synergistic interactions such
as competition between species with similar metabolic demands
and functions must also inuence the growth of these microbes
(Stecher et al., 2010). Given the high bacterial densities found
within the multispecies gut environment, we asked whether
the interspecies quorum sensing signal AI-2 could inuence
the presence and prevalence of species interacting within the
microbiota during the emergence of the streptomycin-treated
community.
Using E. coli strains engineered to either accumulate or
deplete AI-2 in vivo, we demonstrated that manipulation of
signal levels had no effect upon the numbers of E. coli itself (validating the use of this bacterium as a tool) nor upon the overall
density of the microbiota upon colonization of streptomycintreated mice. However, analysis of the species composition revealed differences in the microbiota of mice colonized with the
DlsrK mutant E. coli, compared with those of the mice colonized
with the other two mutants. As this strain accumulates AI-2
extracellularly, our data suggest that increased AI-2 availability
leads to changes in the gut-associated bacteria that, at the
phylum level, increase the ratio of Firmicutes to Bacteroidetes.
This opposes the effect of streptomycin, which massively favors
the Bacteroidetes. Such changes in the abundance of the major
phyla have the potential to further inuence AI-2 levels among
the microbiota: signal production capabilities vary greatly between the major phyla, with a greater proportion of Firmicutes
than Bacteroidetes encoding LuxS orthologs. By clearing
most Firmicutes, streptomycin treatment is likely to create an
environment containing relatively little AI-2, due to a paucity of
AI-2 producers among the resulting gut community. With this
in mind, it is perhaps unsurprising that the E. coli mutant strain
that scavenges AI-2 had no detectable effect upon the streptomycin-treated microbiota. However, increased AI-2 availability,
in favoring the Firmicutes, promotes the group of bacteria with a
higher frequency of AI-2 producers. This positive feedback
might be necessary to provide a context to achieve quorum
and to potentiate further AI-2-dependent responses among
the microbiota.
Members of both the Firmicutes and Bacteroidetes fulll
many functions important to host physiology, including the

fermentation of diverse dietary polysaccharides too complex

to be digested by the host. This process produces short-chain
fatty acids (SCFAs) such as propionate, butyrate, and acetate
(Louis et al., 2014). These molecules, particularly butyrate, provide approximately 10% of the hosts calorie intake (McNeil,
1984) and also inuence host gene expression, proinammatory
cytokine secretion, and Treg induction (Arpaia et al., 2013;
Chang et al., 2014; Furusawa et al., 2013; Smith et al., 2013).
As a result, signaling through butyrate and other SCFAs modulates inammation within the intestinal tract (Maslowski et al.,
2009). Individual species of the Bacteroidetes and Firmicutes
are thought to make distinct contributions to the pools of
each SCFA: Firmicutes are proposed to include the major butyrate producers whereas increased prevalence of Bacteroidetes
has been correlated with increased proportions of propionate in
the total SCFA pool (Salonen et al., 2014). Thus, changes in
abundance of these bacteria (such as those observed here
in response to streptomycin or AI-2 availability) could alter the
concentration of these metabolites within the gut, with consequent downstream effects upon host physiology, and potentially explain the altered concentrations of SCFAs and increase
in inammatory tone previously observed in the ceca of streptomycin-treated mice (Garner et al., 2009; Spees et al., 2013).
Moreover, the increase in abundance of Firmicutes observed
in the presence of the AI-2-producing E. coli mutant offers the
exciting possibility that AI-2 signaling might have ameliorated
the effect of streptomycin on the microbiota-derived functions:
it favored the group of bacteria most detrimentally affected
by antibiotic treatment and may have inuenced their functions
via signaling responses. Heightening the impact of signal
manipulation among the microbiota in this way can be used to
aid the identication of AI-2-regulated functions in this important bacterial community and explore the possibility of using
AI-2 signaling to restore the protective functions of the gut microbiota or inuence microbiota-induced host responses. The
results presented here will facilitate the identication of candidate bacteria that are more likely to be sensitive to this signal
molecule (those that favor signal producers, for example),
knowledge which can be used to design models that further
potentiate AI-2-dependent effects. This work highlights the potential gain from understanding and manipulating the bacterial
chemical repertoire operating within the bacterial community
inhabiting the gut, towards the aim of tailoring the composition
of the microbiota to our benet.

1868 Cell Reports 10, 18611871, March 24, 2015 2015 The Authors

EXPERIMENTAL PROCEDURES
Bacterial Strains and Culture Conditions
All E. coli strains and primers used in this study are listed in supplemental
tables. These are all E. coli K-12 MG1655 derivatives with a streptomycinresistant mutation in rpsL-1(K43N) that express CFP or YFP constitutively.
For details of culture conditions, genetic manipulation, and strain construction,
see Supplemental Experimental Procedures.
Animal Studies
All experiments involving mice were approved by the Institutional Ethics
^ncia and the Portuguese National
Committee at the Instituto Gulbenkian de Cie
Entity (Direcao Geral de Alimentacao e Veterinaria; ref. no. 008957, approval
date 19/03/2013) following the Portuguese legislation (PORT 1005/92), which
complies with the European Directive 86/609/EEC of the European Council.
To demonstrate AI-2 production in the gut, germ-free mice were gavaged
with sterile PBS or 105 CFUs of wild-type, DlsrK, or DlsrRDluxS E. coli strains.
To demonstrate removal of AI-2 from the gut by the E. coli DlsrRDluxS mutant
strain, germ-free mice were gavaged with a 1:1 mix of 105 CFUs of DlsrK and
DlsrKDluxS or DlsrK and DlsrRDluxS mutant bacteria labeled with either CFP
or YFP. Fecal samples were collected 5 days after colonization and plated
to determine bacterial load; cecal contents were harvested for analysis of
AI-2 levels.
6- to 8-week-old male C57BL/6J mice conventionally raised under specic
pathogen-free (SPF) conditions were used to analyze the effects of streptomycin and colonization by E. coli mutant strains upon the intestinal microbiota.
To determine the effects of streptomycin treatment on the gut microbiota
composition, four non-littermate mice were housed individually and maintained
under 5 g/l streptomycin ad libitum in the drinking water (Conway et al. 2004).
Fresh fecal samples were collected prior to antibiotic administration (day 0)
and 2, 4, 7, 12, and 28 days during treatment for subsequent DNA extraction.
To assess the effects of different E. coli mutants upon the gastrointestinal ora,
ve groups of mice (n = 6) were treated and maintained under streptomycin, as
described above. On day 2 of treatment, two mice per group were gavaged with
100 ml PBS containing 108 colony-forming units (CFU) of either ARO071 (DlsrK),
ARO093 (DlsrKDluxS), or ARO081 (DlsrRDluxS) and individually caged (so that
n = 10 per treatment). Fecal samples were collected, part was homogenized in
1 ml sterile PBS and plated to determine colonization levels (CFU/g feces), and
the remainder of each sample was used for subsequent DNA extraction.
Detection of AI-2 Activity
AI-2 activity was measured as previously described (Taga and Xavier, 2011)
using the V. harveyi AI-2 reporter strain TL26 (DluxN DluxS DcqsS; Long
et al., 2009). To determine AI-2 activity in mouse cecal extracts, the cecal contents were homogenized at a 10% weight/volume concentration in 0.1 M
MOPS (pH 7). Samples were centrifuged and ltered, and then an equal volume of methanol was added to precipitate further debris. Supernatants were
vacuum-dried, resuspended at 50% weight/volume in sterile water, and
analyzed by bioassay as above. Enumeration of V. harveyi CFUs demonstrated
no signicant differences in growth of the reporter strain across the samples.
Sample Collection and DNA Extraction
DNA was extracted from fecal samples using the QIAamp DNA Stool Mini Kit
(QIAGEN) following the manufacturers instructions plus an additional membrane disruption step using 0.1-mm glass beads and high-speed shaking.
Samples were stored at 20 C. Total DNA obtained was quantied with Qubit
dsDNA BR Assay Kit (Invitrogen); samples with more than 10 ng/ml were
sequenced and further analyzed (n = 9 per treatment).
Quantication of Bacterial Load by qPCR
Quantitative PCR (qPCR) was performed on DNA extracted from fecal samples
using 16S rRNA universal primers and YFP (CFP)-specic primers to determine total bacterial and YFP-expressing E. coli loads/g feces, respectively.
16S rRNA Gene Amplication, Pyrosequencing, and Analysis
For each sample, the V1V3 region of the 16S rRNA gene was amplied by
PCR and sequenced using a 454 GS FLX Titanium platform following Roche

recommendations. Sequences were processed using mothur (Schloss et al.,

2009) as previously described (Ubeda et al., 2013), with some modications.
See Supplemental Experimental Procedures for further details of all methods
and statistical analyses.
ACCESSION NUMBERS
The NCBI Sequence Read Archive accession number for the 16s rRNA
sequences reported in this paper is SRP051373.
SUPPLEMENTAL INFORMATION
Supplemental Information includes Supplemental Experimental Procedures,
four gures, and two tables and can be found with this article online at
http://dx.doi.org/10.1016/j.celrep.2015.02.049.
AUTHOR CONTRIBUTIONS
J.A.T., R.A.O., C.U., and K.B.X. designed the experiments. J.A.T. and R.A.O.
performed all animal and in vitro experiments and most analyses. R.A.O.,
A.D., and C.U. performed sample preparation for 16S rRNA DNA sequencing
and analysis. J.A.T. and K.B.X. wrote the manuscript with input from co-authors.
ACKNOWLEDGMENTS
We thank Jocelyne Demengeot, Isabel Gordo, Miguel P. Soares, Sandrine
Isaac, and Joao B. Xavier for helpful discussion and critically reading the
manuscript. We also acknowledge Joana Amaro for technical assistance.
This work was funded by grants from the Howard Hughes Medical Institute (International Early Career Scientist; HHMI 55007436) and the Fundacao para a
^ncia e Tecnologia (PTDC/BIA-EVF/118075/2010 and RECI/IMI-IMU/0038/
Cie
2012). C.U. was supported by grants from Ministerio de Ciencia e Innovacion
(MICINN; SAF2011-29458) and a Marie-Curie Career Integration Grant
(PCIG09-GA-2011-293894).
Received: December 18, 2014
Revised: February 18, 2015
Accepted: February 20, 2015
Published: March 19, 2015
REFERENCES
Antunes, L.C., Ferreira, L.Q., Ferreira, E.O., Miranda, K.R., Avelar, K.E., Domingues, R.M., and Ferreira, M.C. (2005). Bacteroides species produce Vibrio
harveyi autoinducer 2-related molecules. Anaerobe 11, 295301.
Armbruster, C.E., Hong, W., Pang, B., Weimer, K.E.D., Juneau, R.A., Turner,
J., and Swords, W.E. (2010). Indirect pathogenicity of Haemophilus inuenzae
and Moraxella catarrhalis in polymicrobial otitis media occurs via interspecies
quorum signaling. mBio. 1, e00102-10.
Arpaia, N., Campbell, C., Fan, X., Dikiy, S., van der Veeken, J., deRoos, P., Liu,
H., Cross, J.R., Pfeffer, K., Coffer, P.J., and Rudensky, A.Y. (2013). Metabolites
produced by commensal bacteria promote peripheral regulatory T-cell generation. Nature 504, 451455.
Arumugam, M., Raes, J., Pelletier, E., Le Paslier, D., Yamada, T., Mende, D.R.,
Fernandes, G.R., Tap, J., Bruls, T., Batto, J.-M., et al.; MetaHIT Consortium
(2011). Enterotypes of the human gut microbiome. Nature 473, 174180.
Barthel, M., Hapfelmeier, S., Quintanilla-Martnez, L., Kremer, M., Rohde, M.,
Hogardt, M., Pfeffer, K., Russmann, H., and Hardt, W.-D. (2003). Pretreatment
of mice with streptomycin provides a Salmonella enterica serovar Typhimurium colitis model that allows analysis of both pathogen and host. Infect.
Immun. 71, 28392858.
Berg, R.D. (1996). The indigenous gastrointestinal microora. Trends Microbiol. 4, 430435.
Bohnhoff, M., and Miller, C.P. (1962). Enhanced susceptibility to Salmonella
infection in streptomycin-treated mice. J. Infect. Dis. 111, 117127.

Cell Reports 10, 18611871, March 24, 2015 2015 The Authors 1869

Bufe, C.G., Jarchum, I., Equinda, M., Lipuma, L., Gobourne, A., Viale, A.,
Ubeda, C., Xavier, J., and Pamer, E.G. (2012). Profound alterations of intestinal
microbiota following a single dose of clindamycin results in sustained susceptibility to Clostridium difcile-induced colitis. Infect. Immun. 80, 6273.

Leatham, M.P., Banerjee, S., Autieri, S.M., Mercado-Lubo, R., Conway, T., and
Cohen, P.S. (2009). Precolonized human commensal Escherichia coli strains
serve as a barrier to E. coli O157:H7 growth in the streptomycin-treated mouse
intestine. Infect. Immun. 77, 28762886.

Bufe, C.G., Bucci, V., Stein, R.R., McKenney, P.T., Ling, L., Gobourne, A., No,
D., Liu, H., Kinnebrew, M., Viale, A., et al. (2015). Precision microbiome reconstitution restores bile acid mediated resistance to Clostridium difcile. Nature
517, 205208.

Ley, R.E., Turnbaugh, P.J., Klein, S., and Gordon, J.I. (2006). Microbial ecology: human gut microbes associated with obesity. Nature 444, 10221023.

Chang, P.V., Hao, L., Offermanns, S., and Medzhitov, R. (2014). The microbial
metabolite butyrate regulates intestinal macrophage function via histone
deacetylase inhibition. Proc. Natl. Acad. Sci. USA 111, 22472252.
Chen, X., Schauder, S., Potier, N., Van Dorsselaer, A., Pelczer, I., Bassler, B.L.,
and Hughson, F.M. (2002). Structural identication of a bacterial quorumsensing signal containing boron. Nature 415, 545549.
Conway, T., Krogfelt, K., and Cohen, P. (2004). The life of commensal Escherichia coli in the mammalian intestine. EcoSal Plus 2004 http://dx.doi.org/10.
1128/ecosalplus.8.3.1.2.
Cuadra-Saenz, G., Rao, D.L., Underwood, A.J., Belapure, S.A., Campagna,
S.R., Sun, Z., Tammariello, S., and Rickard, A.H. (2012). Autoinducer-2 inuences interactions amongst pioneer colonizing streptococci in oral biolms.
Microbiology 158, 17831795.

Long, T., Tu, K.C., Wang, Y., Mehta, P., Ong, N.P., Bassler, B.L., and Wingreen, N.S. (2009). Quantifying the integration of quorum-sensing signals with
single-cell resolution. PLoS Biol. 7, e68.
Louis, P., Hold, G.L., and Flint, H.J. (2014). The gut microbiota, bacterial
metabolites and colorectal cancer. Nat. Rev. Microbiol. 12, 661672.
Lukas, F., Gorenc, G., and Kopecny, J. (2008). Detection of possible AI-2mediated quorum sensing system in commensal intestinal bacteria. Folia
Microbiol. (Praha) 53, 221224.
Maslowski, K.M., Vieira, A.T., Ng, A., Kranich, J., Sierro, F., Yu, D., Schilter,
H.C., Rolph, M.S., Mackay, F., Artis, D., et al. (2009). Regulation of inammatory responses by gut microbiota and chemoattractant receptor GPR43.
Nature 461, 12821286.
McNeil, N.I. (1984). The contribution of the large intestine to energy supplies
in man. Am. J. Clin. Nutr. 39, 338342.

Fabich, A.J., Jones, S.A., Chowdhury, F.Z., Cernosek, A., Anderson, A.,
Smalley, D., McHargue, J.W., Hightower, G.A., Smith, J.T., Autieri, S.M.,
et al. (2008). Comparison of carbon nutrition for pathogenic and commensal
Escherichia coli strains in the mouse intestine. Infect. Immun. 76, 11431152.

Miller, S.T., Xavier, K.B., Campagna, S.R., Taga, M.E., Semmelhack, M.F.,
Bassler, B.L., and Hughson, F.M. (2004). Salmonella typhimurium recognizes
a chemically distinct form of the bacterial quorum-sensing signal AI-2. Mol.
Cell 15, 677687.

Finegold, S.M., Dowd, S.E., Gontcharova, V., Liu, C., Henley, K.E., Wolcott,
R.D., Youn, E., Summanen, P.H., Granpeesheh, D., Dixon, D., et al. (2010).
Pyrosequencing study of fecal microora of autistic and control children.
Anaerobe 16, 444453.

Ng, K.M., Ferreyra, J.A., Higginbottom, S.K., Lynch, J.B., Kashyap, P.C.,
Gopinath, S., Naidu, N., Choudhury, B., Weimer, B.C., Monack, D.M., and
Sonnenburg, J.L. (2013). Microbiota-liberated host sugars facilitate post-antibiotic expansion of enteric pathogens. Nature 502, 9699.

Flint, H.J., Duncan, S.H., Scott, K.P., and Louis, P. (2007). Interactions and
competition within the microbial community of the human colon: links between
diet and health. Environ. Microbiol. 9, 11011111.

Pereira, C.S., McAuley, J.R., Taga, M.E., Xavier, K.B., and Miller, S.T. (2008).
Sinorhizobium meliloti, a bacterium lacking the autoinducer-2 (AI-2) synthase,
responds to AI-2 supplied by other bacteria. Mol. Microbiol. 70, 12231235.

Frank, D.N., St Amand, A.L., Feldman, R.A., Boedeker, E.C., Harpaz, N., and
Pace, N.R. (2007). Molecular-phylogenetic characterization of microbial
community imbalances in human inammatory bowel diseases. Proc. Natl.
Acad. Sci. USA 104, 1378013785.

Pereira, C.S., Santos, A.J., Bejerano-Sagie, M., Correia, P.B., Marques, J.C.,
and Xavier, K.B. (2012). Phosphoenolpyruvate phosphotransferase system
regulates detection and processing of the quorum sensing signal autoinducer-2. Mol. Microbiol. 84, 93104.

Furusawa, Y., Obata, Y., Fukuda, S., Endo, T.A., Nakato, G., Takahashi, D.,
Nakanishi, Y., Uetake, C., Kato, K., Kato, T., et al. (2013). Commensal
microbe-derived butyrate induces the differentiation of colonic regulatory
T cells. Nature 504, 446450.

Pereira, C.S., Thompson, J.A., and Xavier, K.B. (2013). AI-2-mediated signalling in bacteria. FEMS Microbiol. Rev. 37, 156181.

Galley, J.D., Nelson, M.C., Yu, Z., Dowd, S.E., Walter, J., Kumar, P.S., Lyte, M.,
and Bailey, M.T. (2014). Exposure to a social stressor disrupts the community
structure of the colonic mucosa-associated microbiota. BMC Microbiol. 14,
189.

Qin, J., Li, Y., Cai, Z., Li, S., Zhu, J., Zhang, F., Liang, S., Zhang, W., Guan, Y.,
Shen, D., et al. (2012). A metagenome-wide association study of gut microbiota in type 2 diabetes. Nature 490, 5560.
Rakoff-Nahoum, S., Paglino, J., Eslami-Varzaneh, F., Edberg, S., and Medzhitov, R. (2004). Recognition of commensal microora by toll-like receptors is
required for intestinal homeostasis. Cell 118, 229241.

Garner, C.D., Antonopoulos, D.A., Wagner, B., Duhamel, G.E., Keresztes, I.,
Ross, D.A., Young, V.B., and Altier, C. (2009). Perturbation of the small
intestine microbial ecology by streptomycin alters pathology in a Salmonella
enterica serovar typhimurium murine model of infection. Infect. Immun. 77,
26912702.

Reeves, A.E., Koenigsknecht, M.J., Bergin, I.L., and Young, V.B. (2012).
Suppression of Clostridium difcile in the gastrointestinal tracts of germfree
mice inoculated with a murine isolate from the family Lachnospiraceae. Infect.
Immun. 80, 37863794.

Hooper, L.V., Littman, D.R., and Macpherson, A.J. (2012). Interactions

between the microbiota and the immune system. Science 336, 12681273.

Roy, V., Fernandes, R., Tsao, C.-Y., and Bentley, W.E. (2010). Cross species
quorum quenching using a native AI-2 processing enzyme. ACS Chem. Biol.
5, 223232.

Hsiao, A., Ahmed, A.M.S., Subramanian, S., Grifn, N.W., Drewry, L.L., Petri,
W.A., Jr., Haque, R., Ahmed, T., and Gordon, J.I. (2014). Members of the human gut microbiota involved in recovery from Vibrio cholerae infection. Nature
515, 423426.
Kamada, N., Kim, Y.-G., Sham, H.P., Vallance, B.A., Puente, J.L., Martens,
E.C., and Nunez, G. (2012). Regulated virulence controls the ability of a pathogen to compete with the gut microbiota. Science 336, 13251329.

Ruby, E.G. (2008). Symbiotic conversations are revealed under genetic interrogation. Nat. Rev. Microbiol. 6, 752762.
Rutherford, S.T., and Bassler, B.L. (2012). Bacterial quorum sensing: its role in
virulence and possibilities for its control. Cold Spring Harb. Perspect. Med. 2,
pii: a012427.

Kamada, N., Seo, S.-U., Chen, G.Y., and Nunez, G. (2013). Role of the
gut microbiota in immunity and inammatory disease. Nat. Rev. Immunol.
13, 321335.

Salonen, A., Lahti, L., Salojarvi, J., Holtrop, G., Korpela, K., Duncan, S.H., Date,
P., Farquharson, F., Johnstone, A.M., Lobley, G.E., et al. (2014). Impact of diet
and individual variation on intestinal microbiota composition and fermentation
products in obese men. ISME J. 8, 22182230.

Lawley, T.D., and Walker, A.W. (2013). Intestinal colonization resistance.

Immunology 138, 111.

Savage, D.C. (1977). Microbial ecology of the gastrointestinal tract. Annu. Rev.
Microbiol. 31, 107133.

1870 Cell Reports 10, 18611871, March 24, 2015 2015 The Authors

Schauder, S., Shokat, K., Surette, M.G., and Bassler, B.L. (2001). The LuxS
family of bacterial autoinducers: biosynthesis of a novel quorum-sensing
signal molecule. Mol. Microbiol. 41, 463476.
Schloss, P.D., Westcott, S.L., Ryabin, T., Hall, J.R., Hartmann, M., Hollister,
E.B., Lesniewski, R.A., Oakley, B.B., Parks, D.H., Robinson, C.J., et al.
(2009). Introducing mothur: open-source, platform-independent, communitysupported software for describing and comparing microbial communities.
Appl. Environ. Microbiol. 75, 75377541.
Sekirov, I., Tam, N.M., Jogova, M., Robertson, M.L., Li, Y., Lupp, C., and Finlay, B.B. (2008). Antibiotic-induced perturbations of the intestinal microbiota
alter host susceptibility to enteric infection. Infect. Immun. 76, 47264736.

Taga, M.E., and Xavier, K.B. (2011). Methods for analysis of bacterial autoinducer-2 production. Curr. Protoc. Microbiol. Chapter 1, Unit 1C.1.
Turnbaugh, P.J., Ley, R.E., Mahowald, M.A., Magrini, V., Mardis, E.R., and
Gordon, J.I. (2006). An obesity-associated gut microbiome with increased
capacity for energy harvest. Nature 444, 10271031.
Turnbaugh, P.J., Quince, C., Faith, J.J., McHardy, A.C., Yatsunenko, T., Niazi,
F., Affourtit, J., Egholm, M., Henrissat, B., Knight, R., and Gordon, J.I. (2010).
Organismal, genetic, and transcriptional variation in the deeply sequenced gut
microbiomes of identical twins. Proc. Natl. Acad. Sci. USA 107, 75037508.

Smith, P.M., Howitt, M.R., Panikov, N., Michaud, M., Gallini, C.A., Bohlooly-Y,
M., Glickman, J.N., and Garrett, W.S. (2013). The microbial metabolites,
short-chain fatty acids, regulate colonic Treg cell homeostasis. Science 341,
569573.

Ubeda, C., Taur, Y., Jenq, R.R., Equinda, M.J., Son, T., Samstein, M., Viale, A.,
Socci, N.D., van den Brink, M.R.M., Kamboj, M., and Pamer, E.G. (2010).
Vancomycin-resistant Enterococcus domination of intestinal microbiota is
enabled by antibiotic treatment in mice and precedes bloodstream invasion
in humans. J. Clin. Invest. 120, 43324341.

Spees, A.M., Wangdi, T., Lopez, C.A., Kingsbury, D.D., Xavier, M.N., Winter,
S.E., Tsolis, R.M., and Baumler, A.J. (2013). Streptomycin-induced inammation enhances Escherichia coli gut colonization through nitrate respiration.
mBio. 4, e00430-13.

Ubeda, C., Bucci, V., Caballero, S., Djukovic, A., Toussaint, N.C., Equinda, M.,
Lipuma, L., Ling, L., Gobourne, A., No, D., et al. (2013). Intestinal microbiota
containing Barnesiella species cures vancomycin-resistant Enterococcus
faecium colonization. Infect. Immun. 81, 965973.

Stecher, B., Robbiani, R., Walker, A.W., Westendorf, A.M., Barthel, M.,
Kremer, M., Chaffron, S., Macpherson, A.J., Buer, J., Parkhill, J., et al.
(2007). Salmonella enterica serovar typhimurium exploits inammation to
compete with the intestinal microbiota. PLoS Biol. 5, 21772189.

Wang, T., Cai, G., Qiu, Y., Fei, N., Zhang, M., Pang, X., Jia, W., Cai, S., and
Zhao, L. (2012). Structural segregation of gut microbiota between colorectal
cancer patients and healthy volunteers. ISME J. 6, 320329.

Stecher, B., Chaffron, S., Kappeli, R., Hapfelmeier, S., Freedrich, S., Weber,
T.C., Kirundi, J., Suar, M., McCoy, K.D., von Mering, C., et al. (2010). Like
will to like: abundances of closely related species can predict susceptibility
to intestinal colonization by pathogenic and commensal bacteria. PLoS
Pathog. 6, e1000711.

Xavier, K.B., and Bassler, B.L. (2005a). Interference with AI-2-mediated bacterial cell-cell communication. Nature 437, 750753.
Xavier, K.B., and Bassler, B.L. (2005b). Regulation of uptake and processing
of the quorum-sensing autoinducer AI-2 in Escherichia coli. J. Bacteriol.
187, 238248.

Cell Reports 10, 18611871, March 24, 2015 2015 The Authors 1871

Cell Reports

Resource
Mapping Social Behavior-Induced Brain
Activation at Cellular Resolution in the Mouse
Yongsoo Kim,1 Kannan Umadevi Venkataraju,1 Kith Pradhan,1 Carolin Mende,1 Julian Taranda,1 Srinivas C. Turaga,2
Ignacio Arganda-Carreras,2 Lydia Ng,3 Michael J. Hawrylycz,3 Kathleen S. Rockland,1,4 H. Sebastian Seung,2
and Pavel Osten1,*
1Cold

Spring Harbor Laboratory, Cold Spring Harbor, NY 11724, USA

of Brain and Cognitive Sciences, Massachusetts Institute of Technology, Boston, MA 02139, USA
3Allen Institute for Brain Science, Seattle, WA 98103, USA
4Boston University School of Medicine, Boston, MA 02118, USA
*Correspondence: [email protected]
http://dx.doi.org/10.1016/j.celrep.2014.12.014
This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/3.0/).
2Department

SUMMARY

Understanding how brain activation mediates behaviors is a central goal of systems neuroscience. Here,
we apply an automated method for mapping brain
activation in the mouse in order to probe how sexspecic social behaviors are represented in the male
brain. Our method uses the immediate-early-gene
c-fos, a marker of neuronal activation, visualized by
serial two-photon tomography: the c-fos-GFP+ neurons are computationally detected, their distribution
is registered to a reference brain and a brain atlas,
and their numbers are analyzed by statistical tests.
Our results reveal distinct and shared female and
male interaction-evoked patterns of male brain activation representing sex discrimination and social
recognition. We also identify brain regions whose degree of activity correlates to specic features of social
behaviors and estimate the total numbers and the
densities of activated neurons per brain areas. Our
study opens the door to automated screening of
behavior-evoked brain activation in the mouse.
INTRODUCTION
Central to the understanding of brain functions is insight into the
distribution of neuronal activity that drives behavior. Local measurements of brain activity in behaving mice can be made with
electrodes and uorescent calcium indicators (Buzsaki, 2004;
Grewe and Helmchen, 2009), but such approaches provide information regarding only a very small fraction of the 70 million
neurons that comprise the mouse brain. The detection of
elevated levels of the immediate-early genes (IEGs) linked to
recent neuronal activity (Clayton, 2000; Guzowski et al., 2005)
is a more spatially comprehensive technique. While it lacks the
time resolution of electrophysiological recordings or calcium imaging, it does have the potential of providing a complete view of
recent whole-brain activity. Once determined, the whole-brain
IEG-based map can be used to generate structure-function hy292 Cell Reports 10, 292305, January 13, 2015 2015 The Authors

potheses to be probed by high-resolution recordings as well as

optogenetic and chemogenetic methods (Fenno et al., 2011;
Lee et al., 2014).
Here, we use a pipeline of computational methods that permits
automated unbiased mapping of c-fos induction in mouse brains
at single-cell resolution, in a similar way as recently described for
mapping the induction of the IEG Arc (Vousden et al., 2014). Specically, we use serial two-photon (STP) tomography (Ragan
et al., 2012) to image the expression of c-fos-GFP, a transgenic
c-fos green uorescent protein reporter (Reijmers et al., 2007),
across the entire mouse brain. The activated c-fos-GFP+ cells
are computationally detected, their location is mapped at stereotaxic coordinates within a reference brain, and their numbers and
densities per anatomical brain areas are determined within the
Allen Mouse Brain Atlas. Finally, region of interest (ROI)-based
and voxel-based statistical tests are applied to identify brain
areas with behaviorally evoked c-fos-GFP activation.
To demonstrate the application of the computational pipeline
to the mapping of behavior-evoked brain activation, we focus
on mouse social behavior and generate activation maps representing sex-specic social behaviors in the male brain. Rodent
social behavior is an area of intense research, and c-fos mapping, lesion studies, and other functional approaches have
been used to identify brain regions that are activated and
contribute to male and female sexual behaviors as well as
male-male aggressive behaviors (Anderson, 2012; Bia1y and
Kaczmarek, 1996; Brennan and Zufall, 2006; Coolen et al.,
1996; Pfaus and Heeb, 1997; Veening et al., 2005; Yang and
Shah, 2014). Much less is known, on the other hand, about the
brain areas activated during the initial period of sex discrimination and social recognition before the manifestation of the correct behavioral response.
Here, we explore the question of sex discrimination and social
recognition by limiting the male-female and male-male interactions to a brief 90 s period, during which the behavioral repertoire
comprises only social exploratory activity, such as anogenital
snifng, close following, and nose-to-nose snifng, without mating or aggression. A side-by-side comparison of the female
and male interaction-evoked whole-brain activation revealed (1)
a broad activation of areas downstream of both the main and
accessory olfactory bulb (MOB and AOB) in the male-female

Figure 1. STP Tomography and Computational Detection of c-fos-GFP+ Cells

(A) Imaging and data processing pipeline for
mapping whole-brain activation in c-fos-GFP
mice.
(B) A sample 280-serial section data set of a c-fosGFP mouse brain imaged by STP tomography.
(CH) Registration of CN-detected c-fos-GFP+
cells in the RSTP brain. (C) A coronal section
shows the autouorescence signal, which is used
for registering the 3D reconstructed sample brain
(D) onto the RSTP brain (E). (F) A total of 2,177
c-fos-GFP+ cells were detected in the same
coronal section; scale bar, 1 mm. (G) A total of
360,183 c-fos-GFP+ cells were detected in the
whole brain, reconstructed in 3D, and (H) registered onto the RSTP brain using the image
registration parameters established in the (D) and
(E) step.

interaction and a bias toward structures downstream of the MOB

in the male-male interaction; (2) activation of structures related to
behavioral motivation during the male-female, but not male-male,
interaction; and (3) sex-specic as well as shared hypothalamic
activation. Taking advantage of the cellular resolution of the
whole-brain data, we then identied brain regions whose level
of activation was correlated to specic features of the social behaviors, including regions linked to anogenital snifng that lie
downstream of the pheromone-activated AOB and regions linked
to close following that belong to the striatopallidothalamocortical
circuitry. Finally, we calculated the total numbers and the densities of c-fos-GFP+ cells per activated brain region of the femaleand male-specic brain data sets, providing a quantitative
estimate of whole-brain activation evoked by social behaviors.
RESULTS
Whole-Brain Detection of c-fos-GFP+ Cells in STP
Tomography Data Sets
We have established an automated and quantitative wholebrain method for mapping behaviorally evoked c-fos induction

in transgenic reporter mice expressing

c-fos-GFP from a recombinant c-fos
promoter (Reijmers et al., 2007) (Experimental Procedures). This necessitated
the development and optimization of (1)
computational detection of c-fos-GFP+
cells in the mouse brain imaged by
STP tomography (Ragan et al., 2012),
(2) 3D registration of the STP data sets
to a reference mouse brain, and (3)
statistical analyses of the whole-brain
distribution of the c-fos-GFP+ cells
(Figure 1A).
The mouse brains were imaged by
STP tomography as data sets of 280 serial coronal sections, with x-y resolution
1.0 mm and z-spacing 50 mm, which
required an imaging time of 21 hr per brain (Figure 1B) (Ragan
et al., 2012). To achieve a reliable computational detection of
the c-fos-GFP+ cells throughout the whole brain, we used convolutional networks (CNs) that can learn to recognize image
features in complex data sets (V. Jain et al., 2007, IEEE, conference; Turaga et al., 2010) (Figure S1; Experimental Procedures). Since nearby c-fos-GFP+ cells were sometimes merged
in the CN output, a postprocessing step was devised that could
separate such touching cells (Figure S1). The CN performance was then quantied on a new set of marked-up elds
of view from a second c-fos-GFP brain using the F-score measure, which represents the harmonic mean of the precision and
recall (i.e., the false positive and false negative error rate),
where F score 1 is the best and 0 the worst. The CN performance reached F-score 0.88 (precision 0.86, recall 0.90), which
was comparable to human interuser variability represented by
F-score 0.90 (precision 0.90, recall 0.90) (Figure S1; Experimental Procedures). We conclude that the trained CN provides
an automated and highly accurate method for detection of
c-fos-GFP+ cells in whole mouse brains imaged by STP
tomography.
Cell Reports 10, 292305, January 13, 2015 2015 The Authors 293

Figure 2. ROI-Based Segmentation and

Sample Size Calculation
(AD) ROI-based segmentation of the whole-brain
c-fos-GFP+ cell count. (A) Whole-brain view of
360,183 c-fos-GFP+ cells (same brain as in Figure 1H). (B and C) Examples of ABA ROI segmentation (B) and the corresponding c-fos-GFP+
cell counts (C): hippocampus: dark blue; 33,508
cells; medial amygdalar nucleus: light blue; 3,035
cells; nucleus accumbens: green; 13,627 cells;
and infralimbic cortical area: red; 4,665 cells. (D)
Further segmentation of the infralimbic region by
cortical layers; top shows the layer ROIs, from
layer 1 (orange) to layer 6 (purple), and bottom
shows the c-fos-GFP+ cell counts (ILA1 = 223,
ILA2 = 243, ILA2/3 = 1,572, ILA5 = 1,731, ILA6 =
896 c-fos-GFP+ cells). The spacing between the
layers was enlarged for better visualization.
(E and F) Estimation of the sample size based on
power analysis of c-fos-GFP+ cell counts. (E) The
simulation of the relationship between the number
of sufciently powered ROIs and the sample size
shows a steep increase until N = 10, which then
begins to plateau. For the current study, we chose a sample size of N = 13 (dashed line). (F) The plot of the relationship between the statistical power of each ROI
and the effect size for N = 13 group. Of the total 763 ROIs analyzed, 601 (78.8%) showed sufcient statistical power at the effect size 0.6 and 699 (91.6%) at the
effect size 1.0.

Anatomical Registration of the Whole-Brain

c-fos-GFP Data
Results from the CN-based cell counting produce a number of
c-fos-GFP+ cells per the individual 280-section data sets, with
each cell having an xyz location. To be able to compare patterns
of c-fos activation between experimental groups in one common
brain volume, we created a Reference STP (RSTP) brain coregistered to the digital Allen Brain Atlas (ABA) for 8-week-old C57BL/
6 mice (Sunkin et al., 2013) (Figure S2A; Experimental Procedures; Movie S1). The image registrations were done by a 3D
afne transformation, followed by a 3D B-spline transformation
with Mattes Mutual information as the similarity measure (Mattes
et al., 2003). The 3D registration accuracy was calculated to be
65.0 39.9 mm (mean SD; (Figures S2B and S2C; Experimental
Procedures), which is also the accuracy for the registration of all
STP experimental data sets to the RSTP brain for data analysis.
The alignment of the RSTP and ABA Nissl brains was further
improved by 2D afne and B-spline transformations using STP
tomography-imaged CAG-Keima brain, which has a Nissl-like
uorescent labeling from the broadly expressing CAG (cytomegalovirus-IE/chicken b-actin) promoter (Figure S3A; Experimental
Procedures). Finally, the alignment of many ABA anatomical labels was validated, and in some cases manually corrected,
based on a comparison to brain structures delineated by tissue
autouorescence or uorescent protein expression in parvalbumin-, glutamic acid decarboxylase-, and somatostatin-specic
transgenic reporters (Taniguchi et al., 2011) (Figures S3BS3D).
Calculation of the Sample Size for c-fos-GFP-Based
Mapping of Mouse Brain Activation
The RSTP brain allows us to calculate the number of c-fos-GFP+
cells per anatomical ABA regions in the 280-section data sets. To
estimate the required sample size for statistical comparisons, we
used power analysis on data from a baseline group of mice
294 Cell Reports 10, 292305, January 13, 2015 2015 The Authors

(Experimental Procedures). The brains of 7 c-fos-GFP mice (no

experimental manipulation) were imaged by STP tomography,
warped to the RSTP brain, and the c-fos-GFP+ cells were
counted per each anatomical ROI. To determine the optimal
sample size, Monte Carlo methods were applied to this data to
simulate ROI counts for two groups at various effect sizes. As
shown in Figure 2, the number (N) of sufciently powered ROIs
(a < 0.05, power > 0.80) increased at an approximately constant
rate until N = 10, where it started to plateau. We chose N = 1213
as sample size per group, which assures high statistical power
for most ROIs.
The Selection of the Social Behavioral Protocols and
Characterization of c-fos-GFP Induction
Interactions between a male and a female mouse, and between a
male and a male mouse, include initial common social behaviors,
such as anogenital snifng and close following, and consequent
sex-specic behaviors, such mounting and ghting. In the current study, we wished to focus on the comparison of brain activation evoked during the initial social exploration-based phase
of the male-female and male-male interactions, during which
the male is expected to recognize the social stimulus and to
discriminate the sex of the interacting partner.
The social comparison was based on two experimental
groups. In the male-female interaction group, an ovariectomized
(OVX) conspecic female was introduced for 90 s in the home
cage of a naive c-fos-GFP+ male, while in the male-male interaction group, a conspecic male was used as the 90 s stimulus
(Movie S2). As described before in studies of social recognition
(Ferguson et al., 2000, 2001), the brief interaction period
included exploratory behavioral activities of anogenital snifng,
close following, and nose-to-nose snifng, but no sexual
behavior or aggression (Figure S4). The OVX female, which
was recognized by the male as a social stimulus comparable

to an intact female (Figure S4), was chosen to limit experimental

variability due to the estrous cycle (Ferguson et al., 2000; Winslow, 2003).
For control, we included four groups. Baseline group included
mice that were not handled or otherwise manipulated. The
handling group included mice that were transferred to the experimental area for 90 s, the object group included mice that
received a novel object for 90 s, the olfactory group included
mice that were exposed for 90 s to a novel object enriched
with banana-like odor (isoamyl acetate [ISO]; note that ISO is a
monomolecular odor and as such it is likely to induce simpler
activation patterns compared to complex volatile odors.).
In order to characterize the time course of c-fos-GFP induction, we used the 90 s ISO stimulation and tested c-fos-GFP
increase in the main olfactory bulb at 0.5, 1.5, 3, and 5 hr poststimulus. This protocol revealed a peak induction at 3 hr after
the stimulation, which returned to the baseline level at 5 hr (Figures S5AS5C). The time of 3 hr poststimulus was selected for
analysis of all behavioral experiments.
In order to compare the c-fos-GFP signal to native c-fos signal,
we analyzed female interaction-driven induction in eight selected
brain regions by anti-c-fos immunohistochemistry in wild-type
C57BL/6 mice and by STP tomography in c-fos-GFP mice
(note that the c-fos signal was analyzed at 1 hr poststimulus
because of the short half-life of the native c-fos protein). Overall,
the c-fos-GFP signal represented 59% 6% (mean SEM) of
anti-c-fos immunosignal, indicating that the direct c-fos-GFP
uorescence detects approximately 60% of all c-fos induced
cells (Figures S5DS5I). Importantly, the female interactiondriven increase was also highly comparable between the wildtype and c-fos-GFP mice (Figures S5DS5I).
ROI- and Voxel-Based Statistical Analyses
The distribution of the c-fos-GFP+ cells among the different
behavioral groups was compared using ROI- and voxel-based
statistical tests corrected for multiple comparisons by false discovery rate (FDR) (Experimental Procedures). The 694 ROIs
analyzed represent the segmentation of the RSTP Brain volume
by the ABA anatomical regions and the c-fos-GFP cell counts are
compared ROI-to-ROI between the experimental groups (Experimental Procedures). The RSTP brain voxelization (done by overlapping sphere voxels of 100 mm diameter) generates discrete
digitization unbiased of anatomical regions and the c-fos-GFP
cell counts are compared voxel-to-voxel (Experimental Procedures). The voxel-based statistics can reveal hot spot areas
of activation and subregional differences within the anatomical
ROIs (Figure S6).
In the rst ROI analysis, the comparison of the male-female,
male-male, olfactory, and handling groups to the baseline group
revealed broad patterns of brain activation, with 69%, 76%,
79%, and 35% of ROIs activated by the respective manipulations (Table S1). Since all ROIs activated in the handling group
were also activated in the other three groups, the handlinginduced brain activation represents nonspecic shared stimuli,
such as moving the cage to the experimental area. In order to
determine the stimulus-specic brain activations, we next
compared the male-female, male-male, and olfactory groups
to the handling group by both ROI and voxel-based analysis.

Female and Male Interaction-Evoked Brain Activation

It has been proposed that the detection of volatile pheromones
by the main olfactory epithelium (MOE) and MOB is necessary
for sex discrimination (Baum and Kelliher, 2009) (see Discussion). However, the mechanism of such detection at the level
of downstream brain structures is not known. A comparison of
the female and male interaction-evoked activation by ROI statistics revealed largely overlapping c-fos-GFP induction among
MOB-connected brain regions, including the anterior olfactory
nucleus (AON), piriform cortex (PIR), nucleus of the lateral olfactory tract (NLOT), anterior amygdala area (AAA), piriform-amygdala area (PAA), anterior and posterior lateral cortical amygdala
(COAa, COApl), and entorhinal cortex lateral (ENTl), in addition
to a female specic activation of taenia tecta (TT) and postpiriform transition area (TR) (Figure 3; Table S2; note that the heat
map data in Figures 3, 4, and 5 show statistical signicance,
while the magnitude of c-fos upregulation is provided in Figure S7). A further analysis by voxel-based statistics revealed a
mainly dorsal MOB activation by both stimuli and a clear dorsal-ventral separation between the two stimuli in the PIR and
ENT (Figures 3C3F; it should be noted that overlapping voxel
activation between the male and female data sets, seen as yellow areas in Figures 3C3F, represents activation of the same
area, but not necessarily the same neurons). These data suggest that spatial organization of the dorsal MOB outputs leads
to activation of distinct neuronal populations in the PIR and
ENT, which may contribute to sex discrimination in the male
brain.
The sensing of nonvolatile pheromones by the vomeronasal
organ (VNO) and AOB has been proposed to play a critical role
in mate recognition and behavioral motivation (Baum and Kelliher, 2009). Our analysis of brain regions downstream of the
AOB revealed a strong bias toward the female interactionevoked brain activation, including the AOB granular cell layer
(AOBgr), the posterior medial cortical amygdala (COApm), the
entire medial amygdala (MEA), bed nucleus of the accessory olfactory tract (BA), and bed nuclei of the stria terminalis (BST) (Figure 4; Table S2; Movie S3). In contrast, male-male interaction
induced activation in fewer AOB-linked areas, including the BA
and MEA anterior dorsal (ad), anterior ventral (av), and posterior
dorsal (pd) (Figure 4; Table S2). Voxel analysis revealed focal
activation in the AOB in the male-male interaction (Figure 4C)
and a largely overlapping activation in the MEAad, av, and pd
in the male-female and male-male data sets (Figures 4D and
4E; Table S2; Movie S3).
The male-female interaction also showed strongly evoked
activation of brain areas linked to behavioral motivation, including the olfactory tubercle (OT) and nucleus accumbens shell
(ACBsh) of the ventral striatum, prelimbic, infralimbic, and orbital
medial (PL, ILA, and ORBm) prefrontal cortical areas, agranular
insular cortex (AI), substantia innominata (SI; also known as
ventral pallidum), medial dorsal thalamus (MDm), hippocampal
ventral subiculum (SUBv), and serotonergic dorsal raphe (DR)
(Figure 5A; Table S2). In contrast, the male-male interaction
had a comparable induction only in the AI; much weaker activation in the prefrontal cortices, SI, OT, and SUBv; and no signicant activation in the MDm and DR (Figure 5B; Table S2).
Voxel-based analysis revealed that activation in the medial
Cell Reports 10, 292305, January 13, 2015 2015 The Authors 295

Figure 3. Social Behavior-Activated Areas: The MOB and Its Direct Downstream Circuitry
(A and B) ROI analysis:.The male-female (A) and male-male groups (B) are compared to the handling group and signicantly activated ROIs downstream of the
MOB are displayed. Most of the regions were activated by both stimuli. Heatmap in (A) represents FDR corrected statistical signicance. Numbers in (B) represent
bregma A/P coordinates. See Table S2 for ROI full names.
(CF) Voxel-based analysis revealed activation pattern selective for the female stimulus (red), the male stimulus (green), and shared by both stimuli (yellow). (C)
Both male and female stimuli induced dorsal activation in the MOB. (DF) Dorsoventral separation was detected between the male- and female-evoked activation
in the PIR (D and F) and ENT (E and F).
See also Movie S3 for the full data set.

prefrontal cortices in the male-male interaction was limited to supercial cortical layers (Figure 5C; Movie S3). We also observed
a focal activation in the DR at a specic A/P bregma location in
the male-female data set (Figure 5E; Movie S3).
296 Cell Reports 10, 292305, January 13, 2015 2015 The Authors

The activation of the septal and hypothalamic nuclei is

known to mediate both sexual and defensive/aggressive behaviors (Anderson, 2012; Swanson, 2000). We therefore asked
whether the brief interaction used in our experiments was

Figure 4. Social Behavior-Activated Areas: the AOB and Its Direct Downstream Circuitry
(A and B) ROI analysis. The male-female (A) and male-male groups (B) are compared to the handling group, and signicantly activated ROIs downstream of the
AOB are displayed. The female stimulus activated all AOB downstream regions, while the male stimulus induced only a partial activation of these areas. Heatmap
in (A) represents FDR-corrected statistical signicance. Numbers in (A) represent bregma A/P coordinates. See Table S2 for ROI full names.
(CE) Voxel-based analysis revealed a largely overlapping activation pattern (yellow) in the coactivated AOB (C), MEAad and MEAav (D), and MEApd (E), and
selective female-evoked activation in the MEApv and COApm (E).
See also Movie S3 for the full data set.

sufcient to activate these regions even though it lacked overt

mating and ghting. The ROI analysis revealed that the female
stimulus induced activation of the rostral lateral septum (LSr)
and neuroendocrine nuclei, including the medial preoptic nucleus (MPN), medial preoptic area (MPO), ventral premammillary nucleus (PMv), ventrolateral part of the ventromedial
nucleus (VMHvl), paraventricular hypothalamic nucleus (PVH),
dorsomedial hypothalamus (DMH), anteroventral periventricular nucleus (AVPV), posterior periventricular hypothalamic nucleus (PVp), and tuberal nucleus (TU) (Figure 6A; Table S2).
The male stimulus activated the VMHvl, DMH, PVH, PVp,
and TU from the structures of the male-female data set, in
addition to a male-specic activation of the dorsomedial part
of the ventromedial nucleus (VMHdm), the anterior, preoptic,
and intermediate periventricular nuclei (PVa, PVpo, PVi), retrochiasmatic area (RCH), subparaventricular zone (SBPV), supraoptic nucleus (SO), and arcuate nucleus (ARH) (Figure 6B;
Figure S7; Table S2). Voxel-based analysis revealed very

distinct and focal LSr activation at A/P coordinates between +0.345 and 0.145 (Figure 6C; Movie S3). The activation in the VMHvl, which was previously shown to play a role
in both sexual and aggressive behaviors (Lin et al., 2011),
was highly overlapping between the male-female and malemale data sets (Figure 6D; Movie S3). In addition, only a medial
part of the PMv was activated in the male-male data set, suggesting a functional subdivision within this structure (Figure 6E;
Movie S3).
Finally, among additional brain areas, the claustrum (CLA), basomedial amygdala (BMA), and intercalated amygdala (IA) were
activated by both the female and male interactions; the capsular
central amygdala (CEAc), basolateral amygdala (BLA), and
thalamic parataenial nucleus (PT) were activated only in
response to the female stimulus; and the temporal associational,
perirhinal, and ectorhinal (TEa, PERI, and ECT) cortical areas
were activated only in response to the male stimulus (Table
S2). The activation of the hippocampal CA2 region linked to
Cell Reports 10, 292305, January 13, 2015 2015 The Authors 297

Figure 5. Social Behavior-Activated Areas: Motivational Circuitry

(A and B) ROI analysis. The male-female (A) and male-male groups (B) are compared to the handling group, and signicantly activated ROIs previously implicated
in behavioral motivation are displayed. The female stimulus activated frontal cortical areas (PL, ORBm, ILA, AI), ventral striatum (OT, ACB, SI), midline thalamus
(MDm), ventral hippocampus (SUBv), and seretonergic DR, while the male stimulus activated AI; only supercially layer of PL, ORBm, and ILA; and weakly SI, OT,
and SUBv. Heatmap in (A) represents FDR corrected statistical signicance. Numbers in (A) represent bregma A/P coordinates. See Table S2 for ROI full names.
(CE) Voxel analysis showed that (C) the entire ventral part of PL and dorsal half of ILA was activated by the female stimulation, while only the upper layers of the
same regions were activated by male stimulation. (D) Ventral striatum (ACB, SI, OT) showed a patch-shaped, strong activation pattern by female stimulus, but not
by male stimulus. (E) Voxel analysis pinpointed the maximal activation in the DR by the female stimulus at A/P coordinate 4.78.

social memory (Hitti and Siegelbaum, 2014) was also detected in

both the male-female and male-male data sets (Table S2).
Social Behavior-Specic Brain Activation
In addition to the male versus female comparison described
above, we also asked which of the activated brain regions are
specic to social behavior, i.e., are shared between the male-female and male-male data sets and are not activated in response
to a nonsocial stimulus represented by a novel object enriched
with a volatile odor (banana-like ISO).
First, we compared the ISO data set to the handling control.
This analysis revealed the expected activation of the PIR and
other areas downstream of the MOB, which was similar to the
social behavior-evoked activation (Table S2). The activation
throughout the rest of the brain, however, was highly divergent
from the pattern evoked by the social stimuli, as it included
many cortical areas, the entire hippocampus, and the hypotha298 Cell Reports 10, 292305, January 13, 2015 2015 The Authors

lamic subfornical organ (SFO) regulating autonomic functions

(Smith and Ferguson, 2010); the suprachiasmatic nucleus
(SCH) regulating sleep, waking, and locomotor activity (Saper
et al., 2005); and the arcuate nucleus (ARH) linked to feeding
(Sternson, 2013) (Table S2).
Second, we compared the shared male-female and malemale brain activation to the ISO data set. This analysis revealed
the subset of areas specic to social behavior, which included
the amygdalar regions BA, COApl, MEAav, MEApd, BMAp,
BLAv, and PA, the hypothalamic VMHvl and PVH, and the SI (Figure 7; Table S3).
Correlation of c-fos Activation to Time Spent in Social
Behaviors
The time spent in a specic behavioral activity may be expected
to correlate to the number of c-fos-GFP+ cells in brain regions
driving this activity. We next tested whether this correlation

Figure 6. Social Behavior-Activated Areas: Septal and Hypothalamic Activation

(A and B) ROI analysis. The male-female (A) and male-male groups (B) are compared to the handling group, and signicantly activated ROIs of the septum and
hypothalamus are displayed. The female stimulus activated the rostral lateral septum (LSr), AVPV, medial preoptic area (MPO, MPN), PVH, TU, VMHvl, posterior
and ventral DMH, PMv, and PVp. The male stimulus also activated the PVH, VMHvl, DMH (anterior part), TU, and PVp, in addition to a selective activation of the
periventricular hypothalamic nuclei (PVpo, PVi), SBPV, RCH, SO, ARH, and VMHdm. Heatmap in (A) represents FDR-corrected statistical signicance. Numbers
in (A) represent bregma A/P coordinates. See Table S2 for ROI full names.
(C and D) Voxel analysis. (C) A distinct voxel activation was observed in the LSr only by female stimulation. (D) VMHvl showed largely overlapping activation by
both stimuli, while VMHdm and ARH showed activation only by the male stimulus.
(E) PMv is highly activated by the female stimulus, while the medial part of PMv was also activated by the male stimulus.
See also Movie S3 for the full data set.

may be used to functionally link the activated brain areas in the

male-female and male-male data sets to specic features of
the social behavior.
The correlation to the time spent in anogenital snifng identied mainly areas connected to volatile and nonvolatile olfactory
signaling, such as the COAa, COApl, COApm, MEA, and BST,
and hypothalamic neuroendocrine areas including the MPN,
PMv, and VMHvl (Table 1). Correlation to the time spent in close
following identied some of the same areas, such as the MEA
and BST, but also areas linked to behavioral motivation,
including the ACB, OT, SI, ILA, PL, ORBm, MDm, and DR (Table
1). Finally, the correlation to the time spent in nose-to-nose sniffing did not identify any positive association, suggesting that this
behavioral feature is not quantitatively linked to any brain regions
in our data sets. These data suggest that distinct aspects of the

social behavior engage distinct sets of brain areas and that

whole-brain cellular c-fos-GFP analysis is able to reveal this
structure-function relationship.
Calculation of the Density of c-fos-GFP+ Cells per ROIs
While the above analyses identied the activated brain areas,
the cellular resolution of our data allowed us to also estimate
the total numbers and the densities of c-fos-GFP+ cells per
anatomical ROIs. Since the z planes in the 280-section data
sets are spaced 50 mm apart, we transformed the serial 2D
data into 3D whole-brain estimates with a stereological
method (Williams and Rakic, 1988) applied to a high-resolution
5,600-section data set with z spacing of 2.5 mm (Experimental
Procedures). The obtained 2D-to-3D conversion factor of
2.5 was then used to multiply the 2D ROI counts in order to
Cell Reports 10, 292305, January 13, 2015 2015 The Authors 299

Figure 7. Social Behavior-Specic Brain Areas

Brain areas activated by both female and male stimulus, but not by ISO stimulation, are displayed as ROIs. Unique social behavior-activated areas included the
amygdalar BA, COApl, MEAav, MEApd, BLAv, BMAp, PA, hypothalamic PVH, VMHvl, and ventral pallidum (SI).

estimate the total numbers of c-fos-GFP+ cells, and the

total counts were divided by the ROI volumes in order to estimate the densities of c-fos-GFP+ cells per activated ROIs
(Figure S7).
The average cell density in the structures signicantly activated in the female and male data sets were, respectively,
4,993 400 and 4,519 283 per cubic mm (mean SEM),
whereas the average density in these structures in the handling
control was 3,127 201 per cubic mm (Figure S7). Therefore,
the social interactions evoked on average 1,500 to 2,000 cfos-GFP+ cells per cubic mm compared to the handling control,
suggesting a sparse activation of a few percent of neurons per
brain areas (see Discussion).
DISCUSSION
While the general organization of the brain structures regulating
sexual and aggressive behavior is beginning to be understood
(Anderson, 2012; Sokolowski and Corbin, 2012), much remains
unknown about how information is processed from the sensory
periphery (the olfactory system in rodents) to give rise to sexspecic behavioral responses. Here, utilizing a pipeline of
computational methods, including ROI-based whole-brain mapping of c-fos activation, voxel-based mapping of subregional differences in c-fos activation, and correlation analysis linking ROI
activation to behavior, we compared brief female interactionevoked activation in the brain of a male mouse to the activation
evoked by brief interaction with a male. Some more salient ndings from our analyses are discussed below following the
method discussion, while the complete ROI- and voxel-based
results are provided as a resource in Tables S1, S2, and S3
and Movie S3.
The Method Pipeline for c-fos-GFP-Based Mouse Brain
Screening
The entire method pipeline is automated, highly standardized
and operates at a reasonably high-throughput: the imaging
time per one brain is 21 hr, while the imaged processing and
computational analyses take 24 hr that occur in parallel with
the STP imaging (Ragan et al., 2012; Vousden et al., 2014).
300 Cell Reports 10, 292305, January 13, 2015 2015 The Authors

The rst key part of the computational pipeline is the detection

of c-fos-GFP+ cells in the STP data sets. We chose to use CNs,
because these algorithms rely on the learning procedure to account for signal to noise ratio variability and improved performance is achieved by simply increasing the training data set
(V. Jain et al., 2007, IEEE, conference; Turaga et al., 2010). The
trained CN performance (F-score = 0.88) was in fact close to human expert performance (F-score = 0.9), demonstrating the power of this approach for analysis of uorescent labeling in STP
tomography-imaged mouse brains. We have also tested two
other cell detection methodscell counting in the Volocity Image analysis software (Perkin Elmer) and cell counting based
on watershed algorithm (Kopec et al., 2011)but these were
considerable less reliable (F score < 0.5) compared to the CNbased detection.
The second critical step of the method is the registration of the
data sets to the RSTP Brain and the Allen Mouse Brain Atlas. Fixation-induced tissue autouorescence provides rich image content for the registration by the warping algorithm Elastix (Mattes
et al., 2003). As a result, we were able to achieve a high level of
precision (60 mm jitter) for the registration of the experimental
data sets to the RSTP brain (Figure S2). The alignment of the
ABA Nissl-stained sections to the RSTP brain was further helped
by the use of the transgenic CAG-Keima brain with a cellular uorescent protein labeling that matched in most brain regions the
cellular Nissl signal and by several interneuron-specic reporter
mice (Taniguchi et al., 2011) that helped to validate and improve
the matching of the labels to specic brain nuclei (Figure S3).
Consequently, the precision ABA labels became closely aligned
to the RSTP brain, as judged based on brain landmarks, such
as the corpus callosum, hippocampal pyramidal layers, and
many structural borders visible in the autouorescence signal
(Figure S3).
The last part of the method pipeline includes statistical analyses of the brain-wide c-fos-GFP+ cell counts. Since it was rst
established in rat models of seizure, the inducibility of c-fos has
been utilized to map neuronal activation in many behavioral and
pharmacological experiments, demonstrating that c-fos can be
used as an activity reporter in most if not all areas of the brain
(Dragunow and Robertson, 1987; Morgan et al., 1987). The

Table 1. c-fos-GFP Count to the Social Behavioral Correlation

Table 1.

ROIs

ROIs

Anogentical Snifng

Close Following

Isocortex

Anogentical Snifng

Close Following

Midbrain

ILA

DR

Pearson correlation between the time spent in anogenital snifng and

close following and c-fos-GFP cell counts in the regions activated in
the male-female and male-male data sets. Signicance is based on
FDR q value adjusted for multiple comparisons: (+) = 0.01 % FDR q <
0.05; (++) = 0.001 % FDR q < 0.01; and (+++) = FDR q < 0.001.

PL

++

ORBm

++

TT

++

Olfactory area
DP

++

AOBgr

++

COAa

COApl

+++

COApm

+++

PAA

+++

TR

Hippocampal formation
ENTmv

Continued

++

CA3

Cortical subplate
EP

BLAa

++

BLAp

BLAv

++

BMAa

++

BMAp

+++

PA

+++

Cerebral nuclei
ACBsh

OT

AAA

LSr

++
+

CEAc

IA

++

MEAad

+++

MEAav

+++

MEApd

+++

MEApv

+++

SI

BSTmg

++

BSTv

++

BSTp

++

Thalamus
MDm

PT

Hypothalamus
AVPV

++

MPN

++

PMv

+++

VMHvl

TU

++

ROI- and voxel-based statistical analyses established here

transform the traditional laborious immunostaining or in situ hybridization based c-fos mapping into an automated whole-brain
assay.
In addition to the current application, these methods can also
be used to detect and quantify other uorescent protein-expressing transgenic mouse brains by simply training new CN
on a different ground-truth data. This makes our pipeline easily
adaptable to many other applications in quantitative whole-brain
mapping, such as the generation of whole-brain cell counts in
cell type-specic GFP reporter mice (Taniguchi et al., 2011).
Female- and Male-Evoked Maps of Whole-Brain
Activation in the Male Brain
By focusing on the initial period of social exploratory behaviors
between a naive male and a novel conspecic female or male
mouse, we set out to determine the brain activation patterns
that underlie social recognition and sex discrimination in the
male brain. Our results revealed that while the brief interactions
led to an activation of the expected sex-specic response at
the hypothalamic level (indicating that the behaviors were sufcient for correct sex discrimination), the upstream patterns of
brain activation strongly diverged between the two stimuli.
At the level of the AOB and MOB signaling, the female stimulus
evoked activation of all downstream connected brain structures,
while the male stimulus showed activation of all MOB-linked
structures but only a subset of the AOB-linked structures. The
strong MOB-driven brain activation in both behaviors agrees
with the role of volatile signaling in sex discrimination proposed
by studies using chemical lesion of the MOE (Keller et al., 2006)
or genetic disruption of cellular signaling in the MOE (Mandiyan
et al., 2005). The nding that the male and female stimuli activate
different parts of the PIR and ENT areas suggests that topologically distinct MOB cortical outputs may discriminate the sexspecic stimuli. This dorsoventral separation is an example of
a novel spatial organization in the piriform cortex, which until
now has been considered to lack gross sensory input-based topology (Ghosh et al., 2011; Sosulski et al., 2011).
The role of the VNO and AOB-driven activation in social behaviors appears to be less clear than that of the MOE/MOB
signaling. Lesioning of the VNO failed to affect sex discrimination
in male mice (Pankevich et al., 2004), even though it did impair
vocalization after nasal contact with female urine (Bean, 1982),
while genetic disruption of VNO signaling caused male-male
mounting instead of aggressive behavior without affecting
male-female behavior (Stowers et al., 2002). Our data point to
a more prominent role of the AOB-connected brain structures
Cell Reports 10, 292305, January 13, 2015 2015 The Authors 301

in the male-female interaction, as MEA, BST, BA, and COApm

were all activated in the male-female data set, but only MEA
and BA were activated in the male-male data set. The selective
BST activation in the male-female data set included the posterior
division nuclei (principal, interfascicular, and transverse) proposed to function in reproductive behaviors (Dong and Swanson, 2004), and the magnocellular nucleus of the anterior division
proposed to control neuroendocrine functions and pelvic functions, including penile erection (Dong and Swanson, 2006).
The female, but not the male, stimulus also evoked activation
of brain areas of the striatopallidothalamocortical circuit known
to positively regulate behavioral motivation (Ikemoto, 2007;
Sesack and Grace, 2010), including the ventral striatum (OT,
ACB), ventral pallidum (SI), thalamus (MDm), and prefrontal cortex (ILA, PL, ORB). While we did not detect activation of the
dopaminergic neurons of the ventral tegmental area (VTA),
which are known to reinforce ACB functions within this circuit
during sexual behavior (Ikemoto, 2007; Sesack and Grace,
2010), we did detect activation of the serotonergic DR, which
was recently shown to be necessary for ACB functions in social
reward (Dolen et al., 2013). The switch between the DR and
VTA modulation of ventral striatum may contribute to a transition between exploratory and consummatory male-female
behavior.
The analysis of the hypothalamic brain areas revealed activation of structures regulating sexual and aggressive behaviors
(Anderson, 2012; Swanson, 2000): the MPN and PMv regulating
male reproductive behavior (Simerly, 2002; Yang and Shah,
2014) were selectively activated in the male-female data set,
the VMHvl regulating both male sexual behavior and aggression
(Anderson, 2012; Lin et al., 2011; Yang et al., 2013) was activated
in response to both female and male stimuli, and the VMHdm
regulating male defensive behaviors (Lin et al., 2011; Sokolowski
and Corbin, 2012) was activated only in the male-male data set.
Since the brief social interactions did not comprise mating or
aggression, these data show that the activation of the hypothalamic nuclei can precede the manifestation of these behaviors as
part of the male-female and male-male social exploration-based
behaviors.
The Quantication of the Whole-Brain Activation Maps
at Cellular Level
The cellular resolution of our data also allowed us to search
for correlations between behavioral activity and brain activation and to estimate the density of activated cells per brain
area.
The correlation between behavior and c-fos activation can be
expected to identify the most behaviorally relevant brain regions
in which the number of c-fos activated cells reects the behavioral performance in individual animals. In agreement with this
hypothesis, regions correlated to the time spent in anogenital
snifng included mainly amygdalar and hypothalamic areas of
the vomeronasal sensory-motor system transforming the chemosensory information into sexual or aggressive behavior
(Swanson, 2000), while the brain areas correlated to the time
spent in following included the structures linked to behavioral
motivation and described above as part of the striatopallidothalamocortical circuit.
302 Cell Reports 10, 292305, January 13, 2015 2015 The Authors

The correlation analysis can be used to add functional signicance to activated regions that were not previously known to be
involved in social behaviors. For example, the activation of the
amygdalar IA and CEAc nuclei was correlated to the anogenital
snifng time, while the PT thalamus activation was correlated
to following. Since both IA and CEAc can inhibit the medial central amygdala (CEAm), which is the output fear pathway (Pitkanen et al., 1997), these data suggest that the IA and CEAc are
activated by chemosensory cues and may act to modulate fear
behaviors during social exploration. The PT, a part of the dorsal
group of thalamic nuclei, projects to the ACB (Kelley and Stinus,
1984) and may play a role in motivational modulation of the malefemale social behavior.
Finally, the quantication of the numbers of c-fos-GFP+ cells
per brain area can provide information about the approximate
percentage of neurons behaviorally recruited in the identied
brain areas. For example, we observed on average 2-fold increase (3,600 c-fos-GFP+ cells per cubic mm) in the prefrontal
cortical areas in the male-female data sets, compared to the
handling control (Figure S7). Since neuronal density in the mouse
cortex is estimated at 80,000100,000 per cubic mm (Herculano-Houzel et al., 2006; Keller and Carlson, 1999; Meyer et al.,
2010), these data suggest less than 5% of neurons is recruited
in response to the female social stimulus. Further, as most brain
areas showed similar c-fos-GFP+ densities, the behavioral
recruitment of a few percent of neurons is likely a general feature
of c-fos activation. This may represent c-fos induction occurring
only in the most strongly activated cells, and such sparse c-fos
induction may be relevant for the proposed sparse coding of
sensory inputs (Olshausen and Field, 2004).
Caveats of the Current Study
There are several caveats associated with our study. First, while
the behavior is limited to 90 s, our assay cannot determine
whether the observed c-fos-GFP induction occurred entirely
during this brief time period or whether some downstream activation occurred during a longer time interval. Second, the behavioral paradigm includes both the introduction and removal of the
stimulus animal from the home cage of the c-fos-GFP male, and
some of the observed activation pattern thus may reect stress
induced by these manipulations. Third, the use of the OVX females in our study restricts the interpretation of the male-female
activation data to social exploration that lacks the effects of
estrous hormones. Thus, male-female interaction with, for
example, estradiol-induced OVX mice may be expected to
induce brain activation partially distinct from the one described
in the current study. Fourth, since the c-fos-GFP reporter labels
60% of all c-fos+ cells detected by immunostaining, it is
possible that some areas with native c-fos activation were
missed in our assay. Finally, fth, c-fos is a member of a family
of IEGs regulated by neuronal activity, and the detection of other
IEGs (such as Arc, homer-1A, or zif-268) can be expected to
reveal partially overlapping activation maps compared to the
c-fos-GFP map identied in our paper. Because neuronal activation in some brain areas may induce other IEGs but fail to induce
c-fos, the c-fos-GFP-based network of brain areas described
here should not be interpreted as a complete brain activation
map evoked by social behavior.

CONCLUSIONS
Our method of c-fos-GFP-based screening generates cellularresolution maps of behaviorally evoked whole-brain activation
in the mouse. The patterns of female and male interactionevoked brain activation revealed clear separation between the
two stimuli, including at the level of brain structures downstream
of both volatile and nonvolatile chemosensory signaling. These
activation patterns were also markedly different from the activation pattern evoked during nonsocial olfactory-enhanced exploratory behavior. These ndings demonstrate that our method
can be used for screening behavior-evoked whole-brain activation, and we envision that future experiments will yield brainmap-like descriptions for other innate behaviors, such as
aggression and defensive behaviors, or cognitive behaviors,
such as attention and decision making. Further, the same
method can be applied to genetic mouse models of neurodevelopmental disorders with the aim of identifying circuit decits
underlying changes in social, cognitive, and other higher-order
brain functions.

were killed and perfused 1 hr later and the brains were xed overnight in 4%
paraformaldehyde, then cut as 50 mm coronal sections. For immunohistochemistry, sections were exposed to rabbit anti-c-fos antibody (1:10,000,
Santa Cruz SC052) and labeled by DAB solution. FIJI (ImageJ) and Volocity
(Perkin-Elmer) were used for cell counting.
Statistics
We ran statistical comparisons between different behavioral groups based on
either ROIs or evenly spaced voxels. Voxels were overlapping 3D spheres with
100 mm diameter each and spaced 20 mm apart from each other. The cell count
of each voxel was calculated as the number of nuclei found within 100 mm from
the center of the voxel in all 3D. To account for multiple comparisons across all
voxel/ROI locations, we thresholded the p values and reported FDRs. For correlation between c-fos-GFP cell counts and social behavior, Pearson correlation R values were calculated between c-fos-GFP cell counts and time spent in
social behaviors. See the Supplemental Experimental Procedures for more
details.
SUPPLEMENTAL INFORMATION
Supplemental Information includes Supplemental Experimental Procedures,
seven gures, three tables, and three movies and can be found with this article
online at http://dx.doi.org/10.1016/j.celrep.2014.12.014.

EXPERIMENTAL PROCEDURES

ACKNOWLEDGMENTS

Animals
Animal procedures were approved by the Cold Spring Harbor Laboratory Animal Care and Use Committee. The c-fos-GFP mice, Tg(Fos-tTA,Fos-EGFP)
line, were obtained from The Jackson Laboratory. In our study, we used the
direct c-fos-GFP signal, whereas several other studies used the tTA protein
to drive other reporter molecules (Garner et al., 2012; Liu et al., 2012; Matsuo
et al., 2008; Reijmers et al., 2007).

We thank G. Fitzgerald, R. Palaniswamy, and J. Zambratto for expert technical

assistance, J. Kuhl for comments and graphics work, W. Denk and K. Baldwin
for comments on the manuscript, and members of the P.O. lab for helpful discussions. P.O. is supported by the National Institute of Mental Health grant
1R01MH096946-01, Simons Foundation for Autism Research grants 204719
and 253447, and CSHL; Y.K. is supported by NARSAD Young Investigator
Grant, M.H. is supported by the Allen Institute for Brain Science, and H.S.S.
is supported by Howard Hughes Medical Institute and Gatsby Charitable
Foundation.

Behavioral Tests and c-fos-GFP Induction Time Course

Heterozygous c-fos-GFP male mice (811 weeks old) were individually housed
for 1 week before the test. The behavioral stimuli were transfer of the animal to
the experimental arena (handling control) or plus introduction of an OVX
conspecic female (male-female group), conspecic male (male-male group),
50 ml falcon tube (object group), and 50 ml falcon tube with a side opening in
which was cotton ball with isoamyl acetate (1:100 in mineral oil, 40 ml per
experiment, freshly made each day). The stimulus was placed in the home
cage for 90 s and then removed. The behavior was video-recorded and manually scored. After the behavioral stimulus was removed, the mice remained
in the home cage for additional 3 hr and then killed by transcardial perfusion.
For the time course of c-fos-GFP induction, isoamylacetate was introduced
into the mouse home cage for a brief period of 90 s. The mice were killed at
selected time points of 0.5, 1.5, 3, and 5 hr poststimulation.
Brain Preparation, STP Tomography Imaging, and Data Processing
The brains were prepared as described in our previous study (Ragan et al.,
2012). Briey, the brains were embedded in oxidized 4% agarose, crosslinked,
and imaged as 280 serial sections. The raw image tiles were corrected for illumination and stitched in 2D in MATLAB and aligned in 3D in Fiji (Ragan et al.,
2012). The CNs for detection of c-fos-GFP+ cells was trained based on ground
truth data marked up by an expert biologist. The CN performance was scored
based on the F-score (the harmonic mean of the precision and recall). Stereological procedure was used to calculate how CNs 2D based counting can be
converted into 3D counting to calculate the densities of c-fos-GFP+ cells per
activated ROIs. 3D registration methods with Elastix were the same as
described previously (Ragan et al., 2012), but with modied parameters. See
the Supplemental Experimental Procedures for more details.
c-fos Immunohistochemistry and Comparison to c-fos-GFP+ Cell
Counting
Wild-type C57BL/6 mice (810 weeks old) underwent the same behaviors as
the c-fos-GFP mice of the male-to-female and handling groups. The mice

Received: August 24, 2014

Revised: October 21, 2014
Accepted: December 5, 2014
Published: December 31, 2014
REFERENCES
Anderson, D.J. (2012). Optogenetics, sex, and violence in the brain: implications for psychiatry. Biol. Psychiatry 71, 10811089.
Baum, M.J., and Kelliher, K.R. (2009). Complementary roles of the main and
accessory olfactory systems in mammalian mate recognition. Annu. Rev.
Physiol. 71, 141160.
Bean, N.J. (1982). Olfactory and vomeronasal mediation of ultrasonic vocalizations in male mice. Physiol. Behav. 28, 3137.
Bia1y, M., and Kaczmarek, L. (1996). c-Fos expression as a tool to search for
the neurobiological base of the sexual behaviour of males. Acta Neurobiol.
Exp. (Warsz.) 56, 567577.
Brennan, P.A., and Zufall, F. (2006). Pheromonal communication in vertebrates. Nature 444, 308315.
Buzsaki, G. (2004). Large-scale recording of neuronal ensembles. Nat. Neurosci. 7, 446451.
Clayton, D.F. (2000). The genomic action potential. Neurobiol. Learn. Mem. 74,
185216.
Coolen, L.M., Peters, H.J., and Veening, J.G. (1996). Fos immunoreactivity in
the rat brain following consummatory elements of sexual behavior: a sex comparison. Brain Res. 738, 6782.
Dolen, G., Darvishzadeh, A., Huang, K.W., and Malenka, R.C. (2013). Social
reward requires coordinated activity of nucleus accumbens oxytocin and serotonin. Nature 501, 179184.

Cell Reports 10, 292305, January 13, 2015 2015 The Authors 303

Dong, H.W., and Swanson, L.W. (2004). Projections from bed nuclei of
the stria terminalis, posterior division: implications for cerebral hemisphere
regulation of defensive and reproductive behaviors. J. Comp. Neurol. 471,
396433.
Dong, H.W., and Swanson, L.W. (2006). Projections from bed nuclei of the stria
terminalis, magnocellular nucleus: implications for cerebral hemisphere regulation of micturition, defecation, and penile erection. J. Comp. Neurol. 494,
108141.
Dragunow, M., and Robertson, H.A. (1987). Kindling stimulation induces c-fos
protein(s) in granule cells of the rat dentate gyrus. Nature 329, 441442.
Fenno, L., Yizhar, O., and Deisseroth, K. (2011). The development and application of optogenetics. Annu. Rev. Neurosci. 34, 389412.
Ferguson, J.N., Young, L.J., Hearn, E.F., Matzuk, M.M., Insel, T.R., and Winslow, J.T. (2000). Social amnesia in mice lacking the oxytocin gene. Nat.
Genet. 25, 284288.
Ferguson, J.N., Aldag, J.M., Insel, T.R., and Young, L.J. (2001). Oxytocin in the
medial amygdala is essential for social recognition in the mouse. J. Neurosci.
21, 82788285.
Garner, A.R., Rowland, D.C., Hwang, S.Y., Baumgaertel, K., Roth, B.L.,
Kentros, C., and Mayford, M. (2012). Generation of a synthetic memory trace.
Science 335, 15131516.
Ghosh, S., Larson, S.D., Hefzi, H., Marnoy, Z., Cutforth, T., Dokka, K., and
Baldwin, K.K. (2011). Sensory maps in the olfactory cortex dened by longrange viral tracing of single neurons. Nature 472, 217220.
Grewe, B.F., and Helmchen, F. (2009). Optical probing of neuronal ensemble
activity. Curr. Opin. Neurobiol. 19, 520529.

Matsuo, N., Reijmers, L., and Mayford, M. (2008). Spine-type-specic recruitment of newly synthesized AMPA receptors with learning. Science 319, 1104
1107.
Mattes, D., Haynor, D.R., Vesselle, H., Lewellen, T.K., and Eubank, W. (2003).
PET-CT image registration in the chest using free-form deformations. IEEE
Trans. Med. Imaging 22, 120128.
Meyer, H.S., Wimmer, V.C., Oberlaender, M., de Kock, C.P., Sakmann, B., and
Helmstaedter, M. (2010). Number and laminar distribution of neurons in a thalamocortical projection column of rat vibrissal cortex. Cereb. Cortex 20, 2277
2286.
Morgan, J.I., Cohen, D.R., Hempstead, J.L., and Curran, T. (1987). Mapping
patterns of c-fos expression in the central nervous system after seizure. Science 237, 192197.
Olshausen, B.A., and Field, D.J. (2004). Sparse coding of sensory inputs. Curr.
Opin. Neurobiol. 14, 481487.
Pankevich, D.E., Baum, M.J., and Cherry, J.A. (2004). Olfactory sex discrimination persists, whereas the preference for urinary odorants from estrous females disappears in male mice after vomeronasal organ removal. J. Neurosci.
24, 94519457.
Pfaus, J.G., and Heeb, M.M. (1997). Implications of immediate-early gene induction in the brain following sexual stimulation of female and male rodents.
Brain Res. Bull. 44, 397407.
Pitkanen, A., Savander, V., and LeDoux, J.E. (1997). Organization of intraamygdaloid circuitries in the rat: an emerging framework for understanding
functions of the amygdala. Trends Neurosci. 20, 517523.

Guzowski, J.F., Timlin, J.A., Roysam, B., McNaughton, B.L., Worley, P.F., and
Barnes, C.A. (2005). Mapping behaviorally relevant neural circuits with immediate-early gene expression. Curr. Opin. Neurobiol. 15, 599606.

Ragan, T., Kadiri, L.R., Venkataraju, K.U., Bahlmann, K., Sutin, J., Taranda, J.,
Arganda-Carreras, I., Kim, Y., Seung, H.S., and Osten, P. (2012). Serial twophoton tomography for automated ex vivo mouse brain imaging. Nat. Methods
9, 255258.

Herculano-Houzel, S., Mota, B., and Lent, R. (2006). Cellular scaling rules for
rodent brains. Proc. Natl. Acad. Sci. USA 103, 1213812143.

Reijmers, L.G., Perkins, B.L., Matsuo, N., and Mayford, M. (2007). Localization
of a stable neural correlate of associative memory. Science 317, 12301233.

Hitti, F.L., and Siegelbaum, S.A. (2014). The hippocampal CA2 region is essential for social memory. Nature 508, 8892.

Saper, C.B., Lu, J., Chou, T.C., and Gooley, J. (2005). The hypothalamic integrator for circadian rhythms. Trends Neurosci. 28, 152157.

Ikemoto, S. (2007). Dopamine reward circuitry: two projection systems from

the ventral midbrain to the nucleus accumbens-olfactory tubercle complex.
Brain Res. Brain Res. Rev. 56, 2778.

Sesack, S.R., and Grace, A.A. (2010). Cortico-Basal Ganglia reward network:
microcircuitry. Neuropsychopharmacology 35, 2747.

Keller, A., and Carlson, G.C. (1999). Neonatal whisker clipping alters intracortical, but not thalamocortical projections, in rat barrel cortex. J. Comp. Neurol.
412, 8394.
Keller, M., Douhard, Q., Baum, M.J., and Bakker, J. (2006). Sexual experience
does not compensate for the disruptive effects of zinc sulfatelesioning of the
main olfactory epithelium on sexual behavior in male mice. Chem. Senses 31,
753762.

Simerly, R.B. (2002). Wired for reproduction: organization and development of

sexually dimorphic circuits in the mammalian forebrain. Annu. Rev. Neurosci.
25, 507536.
Smith, P.M., and Ferguson, A.V. (2010). Circulating signals as critical regulators of autonomic statecentral roles for the subfornical organ. Am. J. Physiol.
Regul. Integr. Comp. Physiol. 299, R405R415.
Sokolowski, K., and Corbin, J.G. (2012). Wired for behaviors: from development to function of innate limbic system circuitry. Front. Mol. Neurosci. 5, 55.

Kelley, A.E., and Stinus, L. (1984). The distribution of the projection from the
parataenial nucleus of the thalamus to the nucleus accumbens in the rat: an
autoradiographic study. Exp. Brain Res. 54, 499512.

Sosulski, D.L., Bloom, M.L., Cutforth, T., Axel, R., and Datta, S.R. (2011).
Distinct representations of olfactory information in different cortical centres.
Nature 472, 213216.

Kopec, C.D., Bowers, A.C., Pai, S., and Brody, C.D. (2011). Semi-automated
atlas-based analysis of brain histological sections. J. Neurosci. Methods
196, 1219.

Sternson, S.M. (2013). Hypothalamic survival circuits: blueprints for purposive

behaviors. Neuron 77, 810824.

Lee, H.M., Giguere, P.M., and Roth, B.L. (2014). DREADDs: novel tools for
drug discovery and development. Drug Discov. Today 19, 469473.
Lin, D., Boyle, M.P., Dollar, P., Lee, H., Lein, E.S., Perona, P., and Anderson,
D.J. (2011). Functional identication of an aggression locus in the mouse hypothalamus. Nature 470, 221226.
Liu, X., Ramirez, S., Pang, P.T., Puryear, C.B., Govindarajan, A., Deisseroth,
K., and Tonegawa, S. (2012). Optogenetic stimulation of a hippocampal
engram activates fear memory recall. Nature 484, 381385.
Mandiyan, V.S., Coats, J.K., and Shah, N.M. (2005). Decits in sexual
and aggressive behaviors in Cnga2 mutant mice. Nat. Neurosci. 8,
16601662.

304 Cell Reports 10, 292305, January 13, 2015 2015 The Authors

Stowers, L., Holy, T.E., Meister, M., Dulac, C., and Koentges, G. (2002). Loss of
sex discrimination and male-male aggression in mice decient for TRP2. Science 295, 14931500.
Sunkin, S.M., Ng, L., Lau, C., Dolbeare, T., Gilbert, T.L., Thompson, C.L., Hawrylycz, M., and Dang, C. (2013). Allen Brain Atlas: an integrated spatio-temporal portal for exploring the central nervous system. Nucleic Acids Res. 41,
D996D1008.
Swanson, L.W. (2000). Cerebral hemisphere regulation of motivated behavior.
Brain Res. 886, 113164.
Taniguchi, H., He, M., Wu, P., Kim, S., Paik, R., Sugino, K., Kvitsiani, D., Fu, Y.,
Lu, J., Lin, Y., et al. (2011). A resource of Cre driver lines for genetic targeting of
GABAergic neurons in cerebral cortex. Neuron 71, 9951013.

Turaga, S.C., Murray, J.F., Jain, V., Roth, F., Helmstaedter, M., Briggman, K.,
Denk, W., and Seung, H.S. (2010). Convolutional networks can learn to
generate afnity graphs for image segmentation. Neural Comput. 22, 511538.

Williams, R.W., and Rakic, P. (1988). Three-dimensional counting: an accurate

and direct method to estimate numbers of cells in sectioned material. J. Comp.
Neurol. 278, 344352.

Veening, J.G., Coolen, L.M., de Jong, T.R., Joosten, H.W., de Boer, S.F.,
Koolhaas, J.M., and Olivier, B. (2005). Do similar neural systems subserve
aggressive and sexual behaviour in male rats? Insights from c-Fos and pharmacological studies. Eur. J. Pharmacol. 526, 226239.

Winslow, J.T. (2003). Mouse social recognition and preference. Curr. Protoc.
Neurosci. Chapter 8, 16.

Vousden, D.A., Epp, J., Okuno, H., Nieman, B.J., van Eede, M., Dazai, J., Ragan, T., Bito, H., Frankland, P.W., Lerch, J.P., and Henkelman, R.M. (2014).
Whole-brain mapping of behaviourally induced neural activation in mice. Brain
Struct. Funct.

Yang, C.F., Chiang, M.C., Gray, D.C., Prabhakaran, M., Alvarado, M., Juntti,
S.A., Unger, E.K., Wells, J.A., and Shah, N.M. (2013). Sexually dimorphic neurons in the ventromedial hypothalamus govern mating in both sexes and
aggression in males. Cell 153, 896909.
Yang, C.F., and Shah, N.M. (2014). Representing sex in the brain, one module
at a time. Neuron 82, 261278.

Cell Reports 10, 292305, January 13, 2015 2015 The Authors 305

Cell Reports

Resource
Insights into the Evolution of Longevity
from the Bowhead Whale Genome
Michael Keane,1,18 Jeremy Semeiks,2,18 Andrew E. Webb,3,18 Yang I. Li,4,18,19 Vctor Quesada,5,18 Thomas Craig,1
Lone Bruhn Madsen,6 Sipko van Dam,1 David Brawand,4 Patrcia I. Marques,5 Pawel Michalak,7 Lin Kang,7 Jong Bhak,8
Hyung-Soon Yim,9 Nick V. Grishin,2 Nynne Hjort Nielsen,10 Mads Peter Heide-Jrgensen,10 Elias M. Oziolor,11
Cole W. Matson,11 George M. Church,12 Gary W. Stuart,13 John C. Patton,14 J. Craig George,15 Robert Suydam,15
Knud Larsen,6 Carlos Lopez-Otn,5 Mary J. OConnell,3 John W. Bickham,16,17 Bo Thomsen,6
and Joao Pedro de Magalhaes1,*
1Integrative

Genomics of Ageing Group, Institute of Integrative Biology, University of Liverpool, Liverpool L69 7ZB, UK
Hughes Medical Institute and Departments of Biophysics and Biochemistry, University of Texas Southwestern Medical Center,
Dallas, TX 75390-9050, USA
3Bioinformatics and Molecular Evolution Group, School of Biotechnology, Dublin City University, Glasnevin, Dublin 9, Ireland
4MRC Functional Genomics Unit, University of Oxford, Oxford OX1 3QX, UK
5Departamento de Bioqumica y Biologa Molecular, Instituto Universitario de Oncologa (IUOPA), Universidad de Oviedo, 33006 Oviedo,
Spain
6Department of Molecular Biology and Genetics, Aarhus University, 8830 Tjele, Denmark
7Virginia Bioinformatics Institute, Virginia Tech, Blacksburg, VA 24061, USA
8Personal Genomics Institute, Genome Research Foundation, Suwon 443-270, Republic of Korea
9KIOST, Korea Institute of Ocean Science and Technology, Ansan 426744, Republic of Korea
10Greenland Institute of Natural Resources, 3900 Nuuk, Greenland
11Department of Environmental Science, Center for Reservoir and Aquatic Systems Research (CRASR) and Institute for Biomedical Studies,
Baylor University, Waco, TX 76798, USA
12Department of Genetics, Harvard Medical School, Boston, MA 02115, USA
13The Center for Genomic Advocacy (TCGA) and Department of Biology, Indiana State University, Terre Haute, IN 47809, USA
14Department of Forestry and Natural Resources, Purdue University, West Lafayette, IN 47907, USA
15North Slope Borough, Department of Wildlife Management, Barrow, AK 99723, USA
16Battelle Memorial Institute, Houston, TX 77079, USA
17Department of Wildlife and Fisheries Sciences, Texas A&M University, College Station, TX 77843, USA
18Co-rst author
19Present address: Department of Genetics, Stanford University, Stanford, CA 94305, USA
*Correspondence: [email protected]
http://dx.doi.org/10.1016/j.celrep.2014.12.008
This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/3.0/).
2Howard

SUMMARY

The bowhead whale (Balaena mysticetus) is estimated to live over 200 years and is possibly
the longest-living mammal. These animals should
possess protective molecular adaptations relevant
to age-related diseases, particularly cancer. Here,
we report the sequencing and comparative analysis
of the bowhead whale genome and two transcriptomes from different populations. Our analysis
identies genes under positive selection and bowhead-specic mutations in genes linked to cancer
and aging. In addition, we identify gene gain and
loss involving genes associated with DNA repair,
cell-cycle regulation, cancer, and aging. Our results
expand our understanding of the evolution of
mammalian longevity and suggest possible players
involved in adaptive genetic changes conferring cancer resistance. We also found potentially relevant
changes in genes related to additional processes,
112 Cell Reports 10, 112122, January 6, 2015 2015 The Authors

including thermoregulation, sensory perception,

dietary adaptations, and immune response. Our
data are made available online (http://www.
bowhead-whale.org) to facilitate research in this
long-lived species.
INTRODUCTION
The lifespan of some animals, including quahogs, tortoises, and
certain whale species, is far greater than that of humans (Austad,
2010; Finch, 1990). It is remarkable that a warm-blooded species
such as the bowhead whale (Balaena mysticetus) has not only
been estimated to live over 200 years (estimated age of one
specimen 211 SE 35 years), suggesting it is the longest-lived
mammal, but also exhibits very low disease incidence until an
advanced age compared to humans (George et al., 1999; Philo
et al., 1993). As in humans, the evolution of longevity in whales
was accompanied by low fecundity and longer developmental
time (Tacutu et al., 2013), as predicted by evolutionary theory.
The cellular, molecular, and genetic mechanisms underlying
longevity and resistance to age-related diseases in bowhead

Table 1. Statistics of the Bowhead Whale Genome Sequencing

Sequence Data Generated
Sequence Coverage
(for 2.91 Gb)

Libraries

Total Data (Gb)

200 bp paired-end

149.1

500 bp paired-end

141.7

48.73

3 kb mate-paired

57.3

19.73

5 kb mate-paired

72.5

24.93

10 kb mate-paired

28.5

9.83

449.1

154.33

Total

51.23

Genome Assembly Statistics

Assembly
Contigs
Scaffolds

N50 (kb)
34.8
877

Number

Total Size (Gb)

113,673

2.1

7,227

2.3

See also Figures S1 and S2.

whales are unknown, but it is clear that, in order to live so long,

these animals must possess preventative mechanisms against
cancer, immunosenescence, and neurodegenerative, cardiovascular, and metabolic diseases. In the context of cancer,
whales, and bowhead whales, in particular, must possess effective antitumor mechanisms. Indeed, given their large size (in
extreme cases adult bowhead whales can weigh up to 100
tons and are therefore among the largest whales) and exceptional longevity, bowhead whale cells must have a signicantly
lower probability of neoplastic transformation relative to humans
(Caulin and Maley, 2011; de Magalhaes, 2013). Therefore, studying species such as bowhead whales that have greater natural
longevity and resistance to age-related diseases than humans
may lead to insights on the fundamental mechanisms of aging.
Here, we report the sequencing and analysis of the genome of
the bowhead whale, a species of the right whale family Balaenidae that lives in Arctic and sub-Arctic waters. This work provides
clues regarding mechanisms underlying mammalian longevity
and will be a valuable resource for researchers studying the evolution of longevity, disease resistance, and basic bowhead whale
biology.
RESULTS
Sequencing and Annotation of the Bowhead Whale
Genome
We sequenced the nuclear genome of a female bowhead whale
(Balaena mysticetus) using the Illumina HiSeq platform at 1503
coverage. We followed established standards in the eld in
terms of sequencing paired-end libraries at high coverage
plus mate-paired libraries of varying (3, 5, and 10 kb) insert
sizes (Table 1). Contigs and scaffolds were assembled with
ALLPATHS-LG (Gnerre et al., 2011). In line with other genomes
sequenced with second-generation sequencing platforms, the
contig N50 was 34.8 kb and scaffold N50 was 877 kb (Table 1);
the longest scaffold in our assembly was 5,861 kb. In total, our
assembly is 2.3 Gb long. Genome size was estimated experimentally to be 2.91 Gb in another female and 2.87 Gb averaged
with one male (see Supplemental Results and Figure S1), but this

discrepancy likely reects highly repetitive regions, as observed

for the genomes of other species with similar reported sizes such
as the minke whale (Yim et al., 2014).
The full and partial completeness of the bowhead whale draft
genome assembly was evaluated as 93.15% and 97.18%,
respectively, by the CEGMA pipeline (Parra et al., 2007), which
is comparable to the minke whale genome assembly (Yim
et al., 2014). We also generated RNA sequencing (RNA-seq)
data from seven adult bowhead whale tissues (cerebellum, kidney, muscle, heart, retina, liver, and testis) from specimens
from Greenland and Alaska, resulting in two transcriptome assemblies (see Experimental Procedures) and annotated the
genome using MAKER2, which combines ab initio methods, homology-based methods, and transcriptome data to derive gene
models (Holt and Yandell, 2011). Our annotation contains 22,672
predicted protein-coding genes with an average length of 417
(median 307) amino acid residues. In addition, based on transcriptome data from two Alaskan individuals (Table S1), we estimated 0.50.6 SNPs per kilobase of RNA (Table S2). To begin
annotation of the bowhead genome, we identied orthologs
based on similarity with cow, human, and mouse genes/proteins
(see Experimental Procedures), which allowed us to assign predicted gene symbols to 15,831 bowhead genes.
Moreover, to annotate microRNAs in the bowhead genome,
we sequenced small RNA libraries prepared from kidney and
skeletal muscle. The miRDeep algorithm (Friedlander et al.,
2008, 2012) was used to integrate the sequencing data into a
model of microRNA biogenesis by Dicer processing of predicted
precursor hairpin structures in the genome, thus identifying 546
candidate microRNA genes. Of the 546 candidate miRNAs identied in the bowhead, 395 had seed sequences previously identied in miRNAs from human, cow, or mouse, whereas 151 did
not. All of our data are available online from our Bowhead Whale
Genome Resource portal (http://www.bowhead-whale.org).
Analysis of the Draft Bowhead Whale Genome
Repeat sequences make up 41% of the bowhead genome
assembly, most of which (78%) belong to the group of transposable elements (TEs). Although long interspersed nuclear elements (LINEs), such as L1, and short interspersed nuclear
elements (SINEs) are widespread TEs in most mammalian
lineages, the bowhead genome, similar to other cetacean genomesminke, orca, and common bottlenose dolphinis virtually devoid of SINEs (Supplemental Folder 1). LINE-1 (L1) is the
most abundant TE, particularly in orca (90%) and minke whale
(89%) (Figure S2). In comparison, TE diversity (measured with
Shannons index) in the bowhead genome (0.947) is higher
than in orca (0.469) and minke whale (0.515) but lower than in
dolphin (1.389) and cow (Bovine Genome Sequencing and Analysis Consortium et al., 2009) (1.534).
As a rst assessment of coding genes that could be responsible for bowhead whale adaptations, we used bowhead coding
sequences to calculate pairwise dN/dS ratios for 9,682, 12,685,
and 11,158 orthologous coding sequences from minke whale
(Balaenoptera acutorostrata), cow (Bos taurus), and dolphin (Tursiops truncatus), respectively. It is interesting to note that there
are high levels of sequence conservation in the protein coding
regions between bowhead and these species: 96% (minke),
Cell Reports 10, 112122, January 6, 2015 2015 The Authors 113

Figure 1. Phylogeny of Mammals Used in

Codon-Based Maximum Likelihood Comparison of Selective Pressure Variation
The number of candidate genes under positive
selection on each lineage is indicated.

92% (dolphin), and 91% (cow). This is not surprising, however,

given the long generation time of cetaceans and of the bowhead
whale, in particular, with animals only reaching sexual maturity
at >20 years (Tacutu et al., 2013).
Because the minke whale is the closest relative to the bowhead (divergence time 2530 million years ago [Gatesy et al.,
2013]) with a sequenced genome and is smaller (<10 tons) and
probably much shorter lived (maximum lifespan 50 years)
(Tacutu et al., 2013), comparisons between the bowhead and
minke whale genomes may provide insights on the evolution of
bowhead traits and of longevity, in particular. A number of agingand cancer-associated genes were observed among the 420
predicted bowhead-minke orthologs with dN/dS exceeding 1,
including suppressor of cytokine signaling 2 (SOCS2), aprataxin
(APTX), noggin (NOG), and leptin (LEP). In addition, the top 5%
genes with high dN/dS values for bowhead-minke relative to
the values for minke-cow and minke-dolphin orthologs included
forkhead box O3 (FOXO3), excision repair cross-complementing
rodent repair deciency, complementation group 3 (ERCC3),
and broblast growth factor receptor 1 (FGFR1). The data on
dN/dS ratios are also available on our portal to allow researchers
to do their own analysis and quickly retrieve gene(s) of interest.
In a complementary and more detailed analysis of selective
pressure variation, we used codon-based models of evolution
(Yang, 2007) to identify candidate genes with evidence of lineage-specic positive selection (see Experimental Procedures).
Using bowhead, minke, and orca protein-coding data along
with a variety of available high-quality completed genomes
from Laurasiatheria, Euarchontoglires, marsupial, and monotreme species, we identied a total of 866 single-gene ortholog
families (SGOs) (i.e., these gene families have no more than
114 Cell Reports 10, 112122, January 6, 2015 2015 The Authors

one copy in each species). We tested

each of the extant whale lineages, the
ancestral baleen whale, and the most
recent common ancestor (MRCA) of
bowhead, minke, and orca, a total of
ve lineages (Figure 1), for evidence
of lineage-specic positive selection.
Of the two extant whales analyzed, the
number of SGOs exhibiting signatures of
lineage-specic positive selection were
as follows: bowhead (15 gene families)
and minke (ten gene families). The small
number of candidates under positive selection likely reects the high level of protein conservation between bowhead and
other cetaceans as well as the stringent
ltering of candidates due to data-quality
concerns; all results and alignments are
provided in Supplemental Folder 1. A
few genes associated with disease were
identied, including BMP and activin membrane-bound inhibitor
(BAMBI), which has been associated with various pathologies,
including cancer, and also poorly studied genes of potential
interest like GRB2-binding adaptor protein, transmembrane
(GAPT).
In addition to the codon-based models of evolution, we
wished to identify bowhead whale specic amino acid replacement substitutions. To this end, we aligned orthologous
sequences between the bowhead whale and nine other mammalsa total of 4,358 alignments (see Experimental Procedures). Lineage-specic residues identied in this way have
previously been shown to be indicative of signicant changes
in protein function (Tian et al., 2013). Our analysis revealed
several proteins associated with aging and cancer among the
top 5% of unique bowhead residues by concentration (i.e.,
normalized by protein length), including ERCC1 (excision repair
cross-complementing rodent repair deciency, complementation group 1), HDAC1 (histone deacetylase 1), and HDAC2 (Figure 2A). ERCC1 is a member of the nucleotide excision repair
pathway (Gillet and Scharer, 2006), and disruption results in
greatly reduced lifespan in mice and accelerated aging (Weeda
et al., 1997). Histone deacetylases play an important role in the
regulation of chromatin structure and transcription (Lee et al.,
1993) and have been associated with longevity in Drosophila
(Rogina et al., 2002). As such, these represent candidates
involved in adaptive genetic changes conferring disease resistance in the bowhead whale. The full results are available in Supplemental Folder 1.
In addition to genes related to longevity, several interesting
candidate genes emerged from our analysis of lineage-specic
residues of potential relevance to other bowhead traits. Of

Figure 2. Multiple Protein Sequence Alignments of HDAC2 and UCP1

(A) Partial alignment of bowhead HDAC2 with mammalian orthologs. Unique bowhead residues are highlighted at human positions 68, 95, and 133.
(B) Partial alignment of whale UCP1 with mammalian orthologs. Conserved regions involved in UCP1 are marked in red.

note, a number of proteins related to sensory perception of

sound were also identied with bowhead-specic mutations,
including otoraplin (OTOR) and cholinergic receptor, nicotinic,
alpha 10 (CHRNA10), which could be relevant in the context of
the bowheads ability to produce high- and low-frequency tones
simultaneously (Tervo et al., 2011). In addition, many proteins
must play roles in the large differences in size and development
between the bowhead and related species and our results reveal
possible candidates for further functional studies; for example, in
the top ten proteins, SNX3 (sorting nexin 3) has been associated
in one patient with eye formation defects and microcephaly (Vervoort et al., 2002), and WDR5 (WD repeat-containing protein 5)

has been associated with osteoblast differentiation and bone

development (Gori et al., 2006).
In the naked mole rat, a poikilotherm with a low metabolic rate
and body temperature when compared to other mammals,
unique changes in uncoupling protein 1 (UCP1), which is used
to generate heat, have been previously found (Kim et al.,
2011). Because the specic metabolic power output of cells
in vivo for large whales must be much less than for smaller mammals (West et al., 2002), it is interesting to note that UCP1 of
whales has a premature stop codon in C-terminal region, which
is functionally important and conserved in other mammals (Figure 2B). It is tempting to speculate that these changes are related
Cell Reports 10, 112122, January 6, 2015 2015 The Authors 115

Figure 3. Gene Family Expansion and PCNA

(A) Gene family expansion. Numbers in red correspond to the predicted number of gene expansion events during mammalian evolution. Mean divergence time
estimates were used from TimeTree (Hedges et al., 2006) for scaling.
(B) Multiple sequence alignment of PCNA residues 28107, showing bowhead whale-specic duplication (gene IDs: bmy 16007 and bmy 21945). Lineage-specic
amino acids in the duplicated PCNA of bowhead whales are highlighted in red.
(C) Crystal structure of the PCNA (green) and FEN-1 (yellow) complex. Lineage-specic residues on the PCNA structure are colored in red. A zoom in on the
structures reveals a putative interaction between two b sheets, one within PCNA and another within FEN-1. This interaction may be altered through a second
interaction between the PCNA b sheet and a lineage-specic change from glutamine to histidine within PCNA. Distance measurements between pairs of atoms are
marked in black. PDB accession number: 1UL1.
See also Table S3 and Figure S3.

to differences in thermoregulation between whales and smaller

mammals.
Potential Gene Duplications and Gene Losses
Gene duplication is a major mechanism through which phenotypic innovations can evolve (Holland et al., 1994; Kaessmann,
2010). Examples of mammalian phenotypic innovations associated to gene duplication include duplication of RNASE1, a
pancreatic ribonuclease gene, in leaf-eating monkeys that
contributed to adaptative changes in diet and digestive physiology (Zhang et al., 2002), a duplication of GLUD1 in hominoids
that subsequently acquired brain-specic functions (Burki and
Kaessmann, 2004), and domestication of two syncytin gene
copies that contributed to the emergence of placental development in mammals (Dupressoir et al., 2009). We surveyed the
bowhead whale genome for expanded gene families that may
reect lineage-specic phenotypic adaptations and traits.
In the bowhead whale lineage, 575 gene families were predicted to have expanded (Figure 3). However, because gene
expansion predictions are susceptible to false-positives owing
to pseudogenes and annotation artifacts among other biases,
116 Cell Reports 10, 112122, January 6, 2015 2015 The Authors

we applied a stringent lter based on percentage of identity

(Experimental Procedures) that reduced the number of candidate expansions to 41 (see Supplemental Folder 1 for the
complete list). A functional enrichment analysis of these gene
families, using default parameters in DAVID (Huang et al.,
2009), only revealed a statistically signicant enrichment (after
correction for multiple hypothesis testing; Bonferroni <0.001)
for genes associated with translation/ribosome. Given the association between translation and aging, for instance, in the
context of loss of proteostasis (Lopez-Otn et al., 2013), it is
possible that these results reect relevant adaptations in the
bowhead whale.
Upon manual inspection of the gene expansion results, we
found several duplicates of note. For instance, proliferating cell
nuclear antigen (PCNA) is duplicated in bowhead whales with
one copy harboring four lineage-specic residue changes (Figure 3B). Based on our RNA-seq data mapped to the genome
(see Experimental Procedures and full results in Supplemental
Folder 1), both PCNA copies are expressed in bowhead whale
muscle, kidney, retina, and testis. By mapping the lineagespecic residues onto the structure of PCNA in complex with

FEN-1, we uncovered one amino acid substitution (Q38H), which

may affect the interaction between PCNA and FEN-1 (Figure 3C).
A subsequent branch-site test for selective pressure variation
(see Experimental Procedures and Table S3) revealed that one
substitution, D58S, may have undergone positive selection in
the bowhead-whale lineage (with a posterior probability score
of 0.983). The duplication of PCNA during bowhead-whale evolution is of particular interest due to its involvement in DNA damage repair (Hoege et al., 2002) and association with aging in that
its levels in aged rat liver seem to relate to the decrease in the
rate of cell proliferation (Tanno et al., 1996).
Another notable duplicated gene is late endosomal/lysosomal
adaptor, MAPK and MTOR activator 1 (LAMTOR1), in which six
bowhead-specic amino acid changes were identied (Figure S3). LAMTOR1 is involved in amino acid sensing and activation of mTORC1, a gene strongly associated with aging and
cancer (Cornu et al., 2013). The original LAMTOR1 copy was expressed in all bowhead whale adult tissues for which we have
data, with the duplicate having much lower (but detectable)
expression in heart and retina. Also of note, putative duplications
of 26S proteasome non-ATPase regulatory subunit 4 (PSMD4)
and ubiquitin carboxyl-terminal esterase L3 (UCHL3) were identied with evidence of expression, which is intriguing considering
the known involvement of the proteasome-ubiquitin system in
aging (Lopez-Otn et al., 2013) and given previous evidence
that this system is under selection specic to lineages where
longevity increased (Li and de Magalhaes, 2013); UCHL3 has
also been involved in neurodegeneration (Kurihara et al., 2001).
Other gene duplications of potential interest for their role in
mitosis, cancer, and stress response include cAMP-regulated
phosphoprotein 19 (ARPP19), which has three copies even
though we only detected expression of two copies, stomatinlike 2 (STOML2), heat shock factor binding protein 1 (HSBP1)
with four copies of which two appear to be expressed, spermine
synthase (SMS) and suppression of tumorigenicity 13 (ST13).
Similar to previous genome characterizations, we chose the
complete set of known protease genes for a detailed supervised
analysis of gene loss (Quesada et al., 2009). This procedure highlighted multiple gene loss events potentially related to the evolution of several cetacean traits, including adaptations affecting
the immune system, blood homeostasis, digestive system, and
dentition (Figure S4). Thus, the cysteine protease CASP12, a
modulator of the activity of inammatory caspases, has at least
one conserved premature stop codon in bowhead and minke
whales. Interestingly, whereas this protease is conserved and
functional in almost all of the terrestrial mammals, most human
populations display different deleterious variants (Fischer et al.,
2002), presumably with the same functional consequences as
the premature stop codons in whales. Likewise, two paralogues
of carboxypeptidase A (CPA2 and CPA3) have been pseudogenized in bowhead and minke whales. Notably, CPA variants have
been associated with increased risk for prostate cancer in humans (Ross et al., 2009), which could be of interest in the context
of reduced cancer susceptibility in whales compared with humans (de Magalhaes, 2013).
Additionally, we found that multiple coagulation factors have
been lost in bowhead and minke whales. The nding of bowhead
whale-specic changes is also noteworthy because it could be

related to the special characteristics of this mammal. For

example, OTUD6A, a cysteine protease with a putative role in
the innate immune system (Kayagaki et al., 2007), is specically
lacking in the assembled genome and expressed sequences of
the bowhead whale. In addition, whereas the enamel metalloprotease MMP20 has been lost in bowhead and minke whales (Yim
et al., 2014), our analysis suggests that these genomic events
happened independently (see alignments in Supplemental
Folder 1). Finally, as aforementioned, the cysteine protease
UCHL3 seems to have been duplicated through a retrotranscription-mediated event in a common ancestor to bowhead and
minke whales, although only the genome of the bowhead whale
shows a complete, putatively functional open reading frame for
this extra copy of the gene. UCHL3 may play a role in adipogenesis (van Beekum et al., 2012), which indicates that this duplication might be related to the adaptation of the bowhead whale to
the challenging arctic environment. These results suggest
specic scenarios for the role of proteolysis in the evolution of
Mysticetes. Specically, given the relationship between immunity and aging (Lopez-Otn et al., 2013), some of these ndings
might open new approaches for the study of this outstanding
cetacean.
DISCUSSION
The genetic and molecular mechanisms by which longevity
evolves remain largely unexplained. Given the declining costs
of DNA sequencing, de novo genome sequencing is rapidly
becoming affordable. The sequencing of genomes of long-lived
species allows comparative genomics to be employed to study
the evolution of longevity and has already provided candidate
genes for further functional studies (de Magalhaes and Keane,
2013). Nonetheless, deciphering the genetic basis of species differences in longevity has major intrinsic challenges (de Magalhaes and Keane, 2013), and much work remains to uncover
the underlying mechanisms by which some species live much
longer than others. In this context, studying a species so long
lived and with such an extraordinary resistance to age-related
diseases as the bowhead whale will help elucidate mechanisms
and genes conferring longevity and disease resistance in mammals. Remarkably, large whales with over 1,000 times more cells
than humans do not exhibit an increased cancer risk (Caulin and
Maley, 2011), suggesting the existence of natural mechanisms
that can suppress cancer more effectively in these animals. Having the genome sequence of the bowhead whale will allow
researchers to study basic molecular processes and identify
maintenance mechanisms that help preserve life, avoid entropy,
and repair molecular damage. When compared to transcriptome
data (Seim et al., 2014), the genomes greater completeness and
quality permits additional (e.g., gene loss and duplication) and
more thorough analyses. Besides, whereas the genomes of
many commercially important agricultural species have been reported, the bowhead genome sequence is the rst for a species
key to a subsistence diet of indigenous communities. One of the
outputs of this project will be to facilitate and drive research in
this long-lived species. Data and results from this project are
thus made freely available to the scientic community on an
online portal (http://www.bowhead-whale.org/). We provide
Cell Reports 10, 112122, January 6, 2015 2015 The Authors 117

this key resource for studying the bowhead whale and its various
traits, including its exceptional longevity and resistance to
diseases.
EXPERIMENTAL PROCEDURES
DNA and RNA Sampling in Greenland
Bowhead (Balaena mysticetus) DNA used for genome sequencing was isolated from muscle tissue sampled from a 51-year-old female (ID no. 325)
caught in the Disko Bay, West Greenland in 2009 (Heide-Jrgensen et al.,
2012). Tissue samples were stored at 20 C immediately after collection.
Age estimation was performed using the aspartic acid racemization technique
(Garde et al., 2007). CITES no. 12GL1003387 was used for transfer of biological material. Bowhead RNA used for RNA-seq and small RNA analysis
was isolated from two different individuals: kidney samples were from a
44-year-old female (ID no. 500) and muscle samples were isolated from a
44-year-old male (ID no. 322). For more details of the individual whales, see
Heide-Jrgensen et al. (2012).
Genome Sequencing
DNA was extracted following standard protocols, quantied using Qubit and
run on an agarose gel to ensure no degradation had occurred. We then generated 1503 coverage of the genome using the Illumina HiSeq 2000 platform
with 100 bp reads, sequencing paired-end libraries, and mate-paired libraries
with insert sizes of 3, 5, and 10 kb (Table 1). Sequencing was performed at
the Liverpool Centre for Genomic Research (CGR; http://www.liv.ac.uk/
genomic-research/).
Genome Assembly
Libraries were preprocessed in-house by the CGR to remove adaptor sequences. The raw fastq les were trimmed for the presence of the Illumina
adaptor sequence using Cutadapt and then subjected to window-based quality trimming using Sickle with a minimum window quality score of 20. A minimum read-length lter of 10 bp was also applied. Libraries were then assembled with ALLPATHS-LG (Gnerre et al., 2011), which performed all assembly
steps including read error correction, initial read alignment, and scaffolding.
ALLPATHS-LG build 43762 was used with the default input parameters,
including K = 96. Several build parameters were automatically determined by
the software at run time per its standard algorithm. Of 2.88 3 109 paired fragment reads and 1.87 3 109 paired jumping reads, 0.015% were removed as
poly(A) and 1.5% were removed due to low-frequency kmers; 54% of jumping
read pairs were error-corrected, and overall 33% of jumping pairs were redundant. In total, we used 216 Gbp for the 2.3 Gb assembly, meaning that coverage
retained for the assembly was 953. Full assembly and read usage data are
shown in Supplemental Folder 2. Assembly completeness was assayed with
CEGMA by searching for 248 core eukaryotic genes (Parra et al., 2007).
Genome Size Determination
To determine the genome size for bowhead whale, spleen tissues were
acquired from one male (10B17) and one female (10B18). Both whales were
harvested in 2010 as part of the native subsistence hunt in Barrow, Alaska.
Sample processing and staining followed the methods of Vindelv and Christensen (1994). Instrument description and additional methodological details
are provided in Oziolor et al. (2014). Briey, ow cytometric genome size determination is based on propidium iodide uorescent staining of nuclear DNA.
Mean uorescence is calculated for cells in the G0 and G1 phases of the cell
cycle. This method requires direct comparison to known standards to convert
measured uorescence to pg of DNA. The primary standard used in this study
was the domestic chicken (Gallus gallus domesticus). Chicken red blood cells
are widely used as a genome size standard, with an accepted genome size of
C = 1.25 pg. Chicken whole blood was purchased from Innovative Research.
Mouse (Mus musculus) and rat (Rattus norvegicus) were included as internal
checks, with estimates for both falling within 3% of previously published
genome size estimates (Vinogradov, 1998). Spleen tissues from three male
129/SvEvTac laboratory mice and a single male Harlan SD Sprague-Dawley
laboratory rat were used.

118 Cell Reports 10, 112122, January 6, 2015 2015 The Authors

Transcriptome Sequencing and Assembly: Greenland Samples

Total RNA was extracted from the kidney and muscle employing the
mirVanaTM RNA extraction kit (Ambion). RNA integrity of the individual RNA
samples was assessed on a 1% agarose gel using an Agilent 2100 Bioanalyzer
(Agilent Technologies). Library preparation was performed using the
ScriptSeqTM mRNA-seq library preparation kit from Epicenter according to
the manufacturers protocol (Epicenter) and sequenced (100 bp paired end)
as multiplexed samples using the Illumina HiSeq 2000 analyzer. Fastq generation and demultiplexing were performed using the CASAVA 1.8.2 package
(Illumina). The fastq les were ltered for adapters, quality, and length using
Trimmomatic (v.0.27), with a window size of 4, a base quality cutoff of 20,
and a minimum length of 60 (Lohse et al., 2012). De novo transcriptome assembly was performed using the short read assembler software Trinity (release
2013-02-25), which is based on the de Bruijn graph method for assembly, with
default settings (Grabherr et al., 2011).
Transcriptome Sequencing and Assembly: Alaskan Samples
Tissue biopsies were obtained from two male bowhead whales harvested by
Inupiat hunters at Barrow, Alaska during the Fall hunt of 2010; heart, cerebellum, liver, and testes were biopsied from male bowhead number 10B16,
and retina from male bowhead 10B20. Samples were immediately placed in
liquid nitrogen and transported in a dry shipper to Purdue University. RNA
was extracted using TRIZOL reagent (Invitrogen) following the manufacturers
protocol. RNA was puried using an Invitrogen PureLink Micro-to-Midi columns from the Total RNA Purication System using the standard protocol.
RNA quantity and quality was estimated with a spectrophotometer (Nanodrop)
and by gel electrophoresis using an Agilent model 2100 Bioanalyzer. cDNA
libraries were constructed by random priming of chemically sheared poly A
captured RNA. Randomly primed DNA products were blunt ended. Products
from 450650 bp were then isolated using a PippenPrep. After the addition
of an adenine to the fragments, a Y primer amplication was used to produce
properly tailed products. Paired-end sequences of 100 bp per end were generated using the Illumina HiScan platform. Sequences with primer concatamers,
weak signal, and/or poly A/T tails were culled. The Trinity software package for
de novo assembly (Grabherr et al., 2011) was used for transcript reconstruction (Table S1).
Small RNA Sequencing and Annotation
To annotate microRNA genes in the bowhead genome, we conducted deep
sequencing of two small RNA libraries prepared from muscle and kidney tissues (Greenland samples). Total RNA was isolated using mirVana miRNA
Isolation Kit (Ambion). Small RNA in the 15-40 nucleotides range was gel puried and small RNA libraries were prepared for next-generation sequencing
using the ScriptMiner Small RNA-Seq Library Preparation Kit (Epicenter). The
two libraries were sequenced on an Illumina Hi-Seq 2000 instrument to
generate single end sequences of 50 nucleotides. Primary data analysis
was done using the Illumina CASAVA Pipeline software v.1.8.2, and the
sequence reads were further processed by trimming for adapters and
ltering for low quality using Trimmomatic (Lohse et al., 2012). Identication
of conserved and novel candidate microRNA genes in the bowhead genome
was accomplished by applying the miRDeep2 algorithm (Friedlander et al.,
2008, 2012).
Evaluation of Repeat Elements
To evaluate the percentage of repeat elements, RepeatMasker (v.4.0.3; http://
www.repeatmasker.org/) was used to identify repeat elements, with parameters set as -s -species mammal. RMBlast was used as a sequence search
engine to list out all types of repeats. Percentage of repeat elements was
calculated as the total number of repeat region divided by the total length of
the genome, excluding the N-region. Genomes of minke whale (Balaenoptera
acutorostrata), orca (Orcinus orca), common bottlenose dolphin (Tursiops truncates), and cow (Bos taurus) were downloaded from NCBI and run in parallel
for comparison with the bowhead genome.
Genome Annotation
Putative genes were located in the assembly by structural annotation with
MAKER2 (Holt and Yandell, 2011), which combined both bowhead

transcriptomes with comparative and de novo prediction methods including

BLASTX, Exonerate, SNAP, Genemark, and Augustus. In addition to the
RNA-seq data, the entire SwissProt database and the draft proteome of dolphin were used as input to the comparative methods. Repetitive elements
were found with RepeatMasker (http://www.repeatmasker.org/). The complete set of MAKER input parameters, including training sets used for the
de novo prediction methods, are listed in Supplemental Folder 2. In total,
22,672 protein-coding genes were predicted with an average length of 417
(median 307) amino acid residues.
The RNA-seq data from seven adult bowhead tissues described above
were then mapped to the genome: FastQC (http://www.bioinformatics.
babraham.ac.uk/projects/fastqc/) was used for quality control to make sure
that data of all seven samples was of acceptable quality. STAR (Dobin
et al., 2013) was used to generate genome les from the bowhead assembly
and to map the reads to the bowhead genome with 70.3% of reads mapping,
which is in line with other results including those in the minke whale (Yim
et al., 2014). To count the reads overlapping genes, we used ReadCounter
(van Dam et al., 2015). The results obtained from all seven samples were
combined into a single le describing the number of nonambiguously mapping reads for each gene (full results in Supplemental Folder 1). Of the
22,672 predicted protein-coding genes, 89.5% had at least ten reads mapping and 97.5% of predicted genes had at least one read mapping to them,
which is again comparable to other genomes like the minke whale genome
(Yim et al., 2014).
To allow the identication of orthologous relationships with bowhead proteins, all cow protein sequences were downloaded from Ensembl (Flicek
et al., 2013). Cow was initially used because it is the closest relative to the bowhead with a high-quality annotated genome available. First, BLASTP (105)
was used to nd the best hit in the cow proteome for every predicted bowhead
protein, and then the reciprocal best hit for each cow protein was dened as an
ortholog. In addition, human and mouse orthologs from the OPTIC pipeline
(see below) were used to assign predicted gene symbols to genes and proteins. A total of 15,831 bowhead genes have a putative gene symbol based
on these predictions. Homologs in minke whale and dolphin were also derived
and are available on our bowhead genome portal.
Genome Portal
To facilitate further studies of these animals, we constructed an online genome
portal: The Bowhead Whale Genome Resource (http://www.bowhead-whale.
org/). Its database structure, interface, and functionality were adapted from
our existing Naked Mole Rat Genome Resource (Keane et al., 2014). Our
data and results are available from the portal, and supplemental methods
and data les are also available on GitHub (https://github.com/maglab/
bowhead-whale-supplementary).
Pairwise dN/dS Analysis
The CodeML program from the PAML package was used to calculate
pairwise dN/dS ratios (Yang, 2007). This is done using the ratio of nonsynonymous substitutions per nonsynonymous site (dN) to synonymous substitutions per synonymous site (dS), dN/dS, or u (Yang, 2007). Specically, these
pairwise dN/dS ratios were calculated for bowhead coding sequences and
orthologous sequences from minke, cow, and dolphin, excluding coding sequences that were less than 50% of the length of the orthologous sequence.
The results were then ranked by decreasing dN/dS and are available on our
bowhead genome portal. In addition, the ratio of the bowhead-minke dN/dS
value to the higher of the dN/dS values for minke-cow and minke-dolphin
was calculated to identify genes that evolved more rapidly on the bowhead
lineage.
Assessment of Selective Pressure Variation across Single-Gene
Orthologous Families Using Codon-Based Models of Evolution
To accurately assess variation in selective pressure on the bowhead, minke,
and orca lineages in comparison to extant terrestrial mammals, we created
a protein-coding database spanning the placental mammals. Along with the
orca (http://www.ncbi.nlm.nih.gov/bioproject/189949), minke (Yim et al.,
2014), and bowhead data described above, we extracted protein coding
sequences from Ensembl Biomart v.73 (Flicek et al., 2013) for the following

18 genomes: chimpanzee, cow, dog, elephant, gibbon (5.63 coverage),

gorilla, guinea pig, horse, human, macaque, marmoset, microbat, mouse,
opossum, orangutan, platypus, rabbit, and rat. These genomes were all high
coverage (mostly >63 coverage) with the exception of gibbon (Supplemental
Folder 2). Sequence similarity searches were performed using mpi-BLAST
(v 1.6.0) (Altschul et al., 1990) (http://www.mpiblast.org/) on all proteins using
a threshold of 107. Gene families were identied using in-house software that
clusters genes based on reciprocal BLAST hits (Altschul et al., 1990). We identied a total of 6,630 gene families from which we extracted the single-gene
orthologous families (SGOs). Families were considered SGOs if we identied
a single-gene representative in each species (one-to-one orthologs), and to
account for lower coverage genomes and missing data we also considered
cases where a specic gene was not present in a species, i.e., one-to-zero orthology. SGOs were only considered for subsequent analysis if they contained
more than seven species in total and if they contained no internal stop codons
(indicative of sequencing errors). In total, we retained 866 SGOs for further
analysis. Multiple sequence alignments (MSAs) were generated using default
parameters in PRANK (v.100802) (Loytynoja and Goldman, 2008). To minimize
potential false-positives due to poor sequence quality, the MSAs of the 866
SGOs underwent strict data-quality ltering. The rst lter prohibited the
presence of gaps in the MSA if created by unique insertions (>12 bp) in either
bowhead or minke sequences. The second lter required unaligned bowhead
or minke sequences to be at least half the length of their respective MSA.
These two lters rened the number of testable SGOs to 319. The gene phylogeny of each SGO was inferred from the species phylogeny (Morgan et al.,
2013). CodeML from the PAML software package (v.4.4e) (Yang, 2007) was
employed for our selective pressure variation analyses. We analyzed each of
the 319 rened SGOs using the nested codon-based models of evolution under a maximum likelihood framework. We employed the likelihood ratio test
(LRT) using nested models of sequence evolution to evaluate a variety of
models of codon sequence evolution (Yang, 2007). In general, these codon
models allow for variable dN/dS ratios (referred to as u throughout) among
sites in the alignment, along different lineages on our phylogenetic tree, or a
combination of both variations across lineages and sites. To assess the significance of t of each model to the data, we used the recommended LRTs
in CodeML (Yang, 2007) for comparing nested models (see Supplemental
Folder 2). The LRT test statistic approximates the chi-square (c2) distribution
critical value with degrees of freedom equal to the number of additional free
parameters in the alternative model. The goal of the codon-based modeling
is to determine the selective pressures at work in a lineage and site-specic
manner.
The models applied follow the standard nomenclature (i.e., model M1, M2,
A, and A null) (Yang, 2007). Model M1 assumes that there are two classes of
sitesthose with an u value of zero and those with an u value of 1. Model
M2 allows for three classes of sitesone with an u value of zero, one with
an u value of one and one with an u value that is not xed to any value. Given
the relationship between M1 and M2, they can be tested for the signicance of
the difference of the t of these two models using an LRT with df = 2. Finally, we
used model A that allows the u value to vary across sites and across different
lineages in combination. With model A, we can estimate the proportion of sites
and the dN/dS ratio in the foreground lineage of interest in comparison to the
background lineages and the estimated dN/dS ratio is free to vary above 1 (i.e.,
positive selection). Model A can be compared with its site-specic counterpart
(model M1) using the LRT with df = 2. In addition, the lineage and site-specic
model model A null was applied as a second LRT with model A. In model A null,
the additional site category is xed at neutral rather than being estimated from
the data, and this LRT provides an additional test for model A (Zhang et al.,
2005). In this way, we performed independent tests on each of the extant cetacean lineages (orca, minke, and bowhead), as well as testing each ancestral
cetacean branch (the MRCA of the two baleen whales and the MRCA of all
three cetaceans), to determine if there were signatures of positive selection
that are unique to each lineage (Yang and dos Reis, 2011). Using empirical
Bayesian estimations, we identied the specic residues that are positively
selected in each lineage tested. Positive selection was inferred if all of the
following criteria were met: (1) if the LRT was signicant, (2) if the parameters
estimated under that model were concurrent with positive selection, and (3) if
the alignment in that region was of high quality (as judged by alignment

Cell Reports 10, 112122, January 6, 2015 2015 The Authors 119

completeness and quality in that region). The posterior probability (PP) of a

positively selected site is estimated using two calculations: Naive Empirical
Bayes (NEB) or Bayes Empirical Bayes (BEB) (Yang, 2007). If both NEB and
BEB are predicted, we reported the BEB results as they have been shown to
be more robust under certain conditions (Yang et al., 2005). For all models
used in the analysis where u is estimated from the data, a variety of starting
u values was used for the calculation of likelihood estimates. This ensures
that the global minimum is reached.
Identication of Proteins with Bowhead-Unique Residues
An in-house Perl pipeline was used to align each bowhead protein with orthologs from nine other mammals: human (Homo sapiens), dog (Canis familiaris),
mouse (Mus musculus), rat (Rattus norvegicus), minke whale (Balaenoptera
acutorostrata), cow (Bos taurus), dolphin (Tursiops truncatus), horse (Equus
caballus), and elephant (Loxodonta africana) and then identify the unique bowhead amino acid residues. Gaps were excluded from the analysis, and a
maximum of one unknown residue was allowed in species other than the bowhead. The results were ranked by the number of unique residues normalized by
the protein length (full results in Supplemental Folder 1).
Gene Expansion Analysis, Filtering, and Expression
Human, mouse, dog, cow, dolphin, and platypus genomes and gene annotations were obtained from Ensembl (Flicek et al., 2013), the genome and gene
annotation of minke whale were obtained from Yim et al. (2014). In total,
21,069, 22,275, 19,292, 19,988, 15,769, 17,936, 20,496, and 22,733 human,
mouse, dog, cow, dolphin, platypus, minke whale, and bowhead whale genes,
respectively, were used to construct orthology mappings using OPTIC (Heger
and Ponting, 2007). Briey, OPTIC builds phylogenetic trees for gene families
by rst assigning orthology relationships based on pairwise orthologs
computed using PhyOP (Goodstadt and Ponting, 2006). Then, a tree-based
method, PhyOP, is used to cluster genes into orthologous groups, and, last,
gene members are aligned and phylogenetic trees built with TreeBeST (Vilella
et al., 2009). Further details are available in the OPTIC paper (Heger and Ponting, 2007). Predicted orthology groups can be accessed at http://genserv.
anat.ox.ac.uk/clades/vertebrates_bowhead.
To identify gene families that underwent expansion, gene trees were
reconciled with the consensus species tree, and duplicated nodes were
identied. The tree used, derived from TimeTree (Hedges et al., 2006), was:
(mm_oanatinus5, ((mm_cfamiliaris3, (mm_btaurus, (mm_ttruncatus, (mm_
balaenoptera, mm_bmysticetus)))), (mm_hsapiens10, mm_mmusculus5))).
The following algorithm was used to reconcile gene and species trees.
A stringent lter was applied to the data so that gene duplicates in bowhead whales were required to differ by at most 10% in protein sequence
from a cognate copy but were also required to differ by at least 1% to avoid
assembly artifacts and to remove recently duplicated copies with no function. Further manual inspection of the alignments was performed. Gene
expression inferred from our RNA-seq data was used to check the expression of duplicates.
An in-house peptide-sensitive approach was used to align the PCNA cDNA
into codons, and CodeML/PAML was used to test M0, a one-rate model that
assumes the same rate of evolution in all branches against M2^a, a branch site
test with one rate for the background and one rate for the bowhead whale
branch (Yang, 2007).
ACCESSION NUMBERS
Our data and results can be downloaded from the Bowhead Whale Genome
Resource (http://www.bowhead-whale.org/downloads/). In addition, data
are available at the NCBI BioProject PRJNA194091 with raw sequencing reads
in the Sequence Read Archive (SRP050351).

AUTHOR CONTRIBUTIONS
G.M.C., J.C.G., R.S., J.W.B., B.T., and J.P.M. conceived and designed the
study; L.B.M., E.M.O., and C.W.M. performed the experiments; M.K., J.S.,
A.E.W., Y.I.L., V.Q., L.B.M., S.v.D., D.B., P.I.M., P.M., L.K., J.B., H.-S.Y.,
G.W.S., J.C.P., C.L.-O., M.J.O., J.W.C., and J.P.M. analyzed the data; T.C.,
N.V.G., N.H.N., M.P.H.-J., R.S., and K.L. contributed reagents/materials/analysis tools; and M.K. and J.P.M. wrote the paper.
ACKNOWLEDGMENTS
This project was supported by grants from the Life Extension Foundation and
the Methuselah Foundation to J.P.M. We thank the Inuit whaling captains of
the Alaska Eskimo Whaling Commission and the Barrow Whaling Captains
Association for allowing us to sample their whales and their willingness to support the transcriptome study. This study was also partly funded by the Augustinus Foundation to K.L. T.C. was supported by a Wellcome Trust grant
(WT094386MA) to J.P.M. and M.K. was supported by a studentship from the
University of Liverpools Faculty of Health and Life Sciences. J.S. was supported by grants from the NIH (GM094575 to N.V.G.) and the Welch
Foundation (I-1505 to N.V.G.). Y.I.L. was supported by a Nufeld Department
of Medicine Prize studentship from the University of Oxford. C.L.-O. is an
Investigator of the Botin Foundation also supported by grants from Ministerio
de Economa y Competitividad-Spain and Instituto de Salud Carlos III (RTICC)Spain. M.J.O. and A.E.W. are funded by Science Foundation Ireland Research
Frontiers Programme Grant (EOB2763) to M.J.O. M.J.O. would also like to
acknowledge the Fulbright Commission for the Fulbright Scholar Award
2012-2013. M.J.O. and A.E.W. thank the SFI/HEA Irish Centre for High-End
Computing (ICHEC) for processor time and technical support. This work was
also supported by the Korea Institute of Ocean Science and Technology
(KIOST) in-house program (PE99212). Further thanks to Prof. Chris Ponting
and Dr. Andreas Heger for hosting gene orthology predictions from OPTIC,
the University of Liverpool High Performance Computing facilities for processor time, Eric de Sousa for help with RNA-seq data analysis, and to Louise
Crompton for assistance in compiling and formatting the bibliography. Last,
we are grateful to the staff at the Liverpool Centre for Genomic Research for
advice during this project.
Received: September 7, 2014
Revised: November 21, 2014
Accepted: December 3, 2014
Published: December 24, 2014
REFERENCES
Altschul, S.F., Gish, W., Miller, W., Myers, E.W., and Lipman, D.J. (1990). Basic
local alignment search tool. J. Mol. Biol. 215, 403410.
Austad, S.N. (2010). Methusalehs Zoo: how nature provides us with clues for
extending human health span. J. Comp. Pathol. 142 (Suppl 1), S10S21.
Burki, F., and Kaessmann, H. (2004). Birth and adaptive evolution of a
hominoid gene that supports high neurotransmitter ux. Nat. Genet. 36,
10611063.
Caulin, A.F., and Maley, C.C. (2011). Petos Paradox: evolutions prescription
for cancer prevention. Trends Ecol. Evol. 26, 175182.
Bovine Genome Sequencing and Analysis Consortium, Elsik, C.G., Tellam,
R.L., Worley, K.C., Gibbs, R.A., Muzny, D.M., Weinstock, G.M., Adelson,
D.L., Eichler, E.E., Elnitski, L., et al. (2009). The genome sequence of taurine
cattle: a window to ruminant biology and evolution. Science 324, 522528.
Cornu, M., Albert, V., and Hall, M.N. (2013). mTOR in aging, metabolism, and
cancer. Curr. Opin. Genet. Dev. 23, 5362.

SUPPLEMENTAL INFORMATION

de Magalhaes, J.P. (2013). How ageing processes inuence cancer. Nat. Rev.
Cancer 13, 357365.

Supplemental Information includes Supplemental Results, four gures, three

tables, and two supplemental data les and can be found with this article
online at http://dx.doi.org/10.1016/j.celrep.2014.12.008.

de Magalhaes, J.P., and Keane, M. (2013). Endless paces of degeneration

applying comparative genomics to study evolutions moulding of longevity.
EMBO Rep. 14, 661662.

120 Cell Reports 10, 112122, January 6, 2015 2015 The Authors

Dobin, A., Davis, C.A., Schlesinger, F., Drenkow, J., Zaleski, C., Jha, S., Batut,
P., Chaisson, M., and Gingeras, T.R. (2013). STAR: ultrafast universal RNA-seq
aligner. Bioinformatics 29, 1521.

Holt, C., and Yandell, M. (2011). MAKER2: an annotation pipeline and genomedatabase management tool for second-generation genome projects. BMC
Bioinformatics 12, 491.

Dupressoir, A., Vernochet, C., Bawa, O., Harper, F., Pierron, G., Opolon, P.,
and Heidmann, T. (2009). Syncytin-A knockout mice demonstrate the critical
role in placentation of a fusogenic, endogenous retrovirus-derived, envelope
gene. Proc. Natl. Acad. Sci. USA 106, 1212712132.

Huang, W., Sherman, B.T., and Lempicki, R.A. (2009). Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat.
Protoc. 4, 4457.

Finch, C. (1990). Longevity, Senescence, and the Genome (Chicago: University of Chicago Press).
Fischer, H., Koenig, U., Eckhart, L., and Tschachler, E. (2002). Human caspase
12 has acquired deleterious mutations. Biochem. Biophys. Res. Commun.
293, 722726.
Flicek, P., Ahmed, I., Amode, M.R., Barrell, D., Beal, K., Brent, S., CarvalhoSilva, D., Clapham, P., Coates, G., Fairley, S., et al. (2013). Ensembl 2013.
Nucleic Acids Res. 41, D48D55.

Kaessmann, H. (2010). Origins, evolution, and phenotypic impact of new

genes. Genome Res. 20, 13131326.
Kayagaki, N., Phung, Q., Chan, S., Chaudhari, R., Quan, C., ORourke, K.M.,
Eby, M., Pietras, E., Cheng, G., Bazan, J.F., et al. (2007). DUBA: a deubiquitinase that regulates type I interferon production. Science 318, 16281632.
Keane, M., Craig, T., Alfoldi, J., Berlin, A.M., Johnson, J., Seluanov, A., Gorbunova, V., Di Palma, F., Lindblad-Toh, K., Church, G.M., and de Magalhaes, J.P.
(2014). The Naked Mole Rat Genome Resource: facilitating analyses of cancer
and longevity-related adaptations. Bioinformatics 30, 35583560.

Friedlander, M.R., Chen, W., Adamidi, C., Maaskola, J., Einspanier, R., Knespel, S., and Rajewsky, N. (2008). Discovering microRNAs from deep
sequencing data using miRDeep. Nat. Biotechnol. 26, 407415.

Kim, E.B., Fang, X., Fushan, A.A., Huang, Z., Lobanov, A.V., Han, L., Marino,
S.M., Sun, X., Turanov, A.A., Yang, P., et al. (2011). Genome sequencing reveals insights into physiology and longevity of the naked mole rat. Nature
479, 223227.

Friedlander, M.R., Mackowiak, S.D., Li, N., Chen, W., and Rajewsky, N. (2012).
miRDeep2 accurately identies known and hundreds of novel microRNA
genes in seven animal clades. Nucleic Acids Res. 40, 3752.

Kurihara, L.J., Kikuchi, T., Wada, K., and Tilghman, S.M. (2001). Loss of UchL1 and Uch-L3 leads to neurodegeneration, posterior paralysis and dysphagia.
Hum. Mol. Genet. 10, 19631970.

Garde, E., Heide-Jorgensen, M.P., Hansen, S.H., Nachman, G., and Forchhammer, M.C. (2007). Age-specic growth and remarkable longevity in narwhals (Monodon monoceros) from West Greenland as estimated by aspartic
acid racemization. J. Mammal. 88, 4958.

Lee, D.Y., Hayes, J.J., Pruss, D., and Wolffe, A.P. (1993). A positive role for histone acetylation in transcription factor access to nucleosomal DNA. Cell 72,
7384.

Gatesy, J., Geisler, J.H., Chang, J., Buell, C., Berta, A., Meredith, R.W.,
Springer, M.S., and McGowen, M.R. (2013). A phylogenetic blueprint for a
modern whale. Mol. Phylogenet. Evol. 66, 479506.
George, J.C., Bada, J., Zeh, J., Scott, L., Brown, S.E., OHara, T., and Suydam,
R. (1999). Age and growth estimates of bowhead whales (Balaena mysticetus)
via aspartic acid racemization. Can. J. Zool. 77, 571580.
Gillet, L.C.J., and Scharer, O.D. (2006). Molecular mechanisms of mammalian
global genome nucleotide excision repair. Chem. Rev. 106, 253276.
Gnerre, S., Maccallum, I., Przybylski, D., Ribeiro, F.J., Burton, J.N., Walker,
B.J., Sharpe, T., Hall, G., Shea, T.P., Sykes, S., et al. (2011). High-quality draft
assemblies of mammalian genomes from massively parallel sequence data.
Proc. Natl. Acad. Sci. USA 108, 15131518.
Goodstadt, L., and Ponting, C.P. (2006). Phylogenetic reconstruction of orthology, paralogy, and conserved synteny for dog and human. PLoS Comput. Biol.
2, e133.
Gori, F., Friedman, L.G., and Demay, M.B. (2006). Wdr5, a WD-40 protein, regulates osteoblast differentiation during embryonic bone development. Dev.
Biol. 295, 498506.
Grabherr, M.G., Haas, B.J., Yassour, M., Levin, J.Z., Thompson, D.A., Amit, I.,
Adiconis, X., Fan, L., Raychowdhury, R., Zeng, Q., et al. (2011). Full-length
transcriptome assembly from RNA-seq data without a reference genome.
Nat. Biotechnol. 29, 644652.

Li, Y., and de Magalhaes, J.P. (2013). Accelerated protein evolution analysis
reveals genes and pathways associated with the evolution of mammalian
longevity. Age (Dordr.) 35, 301314.
Lohse, M., Bolger, A.M., Nagel, A., Fernie, A.R., Lunn, J.E., Stitt, M., and Usadel, B. (2012). RobiNA: a user-friendly, integrated software solution for RNAseq-based transcriptomics. Nucleic Acids Res. 40, W622W627.
Lopez-Otn, C., Blasco, M.A., Partridge, L., Serrano, M., and Kroemer, G.
(2013). The hallmarks of aging. Cell 153, 11941217.
Loytynoja, A., and Goldman, N. (2008). A model of evolution and structure for
multiple sequence alignment. Philos. Trans. R. Soc. Lond. B Biol. Sci. 363,
39133919.
Morgan, C.C., Foster, P.G., Webb, A.E., Pisani, D., McInerney, J.O., and
OConnell, M.J. (2013). Heterogeneous models place the root of the placental
mammal phylogeny. Mol. Biol. Evol. 30, 21452156.
Oziolor, E.M., Bigorgne, E., Aguilar, L., Usenko, S., and Matson, C.W. (2014).
Evolved resistance to PCB- and PAH-induced cardiac teratogenesis, and
reduced CYP1A activity in Gulf killish (Fundulus grandis) populations from
the Houston Ship Channel, Texas. Aquat. Toxicol. 150, 210219.
Parra, G., Bradnam, K., and Korf, I. (2007). CEGMA: a pipeline to accurately
annotate core genes in eukaryotic genomes. Bioinformatics 23, 10611067.
Philo, L.M., Shotts, E.B., and George, J.C. (1993). Morbidity and mortality. In
The Bowhead Whale, J.J. Burns, J.J. Montague, and C.J. Cowles, eds. (Lawrence, Kansas: Allen Press), pp. 275312.

Hedges, S.B., Dudley, J., and Kumar, S. (2006). TimeTree: a public knowledgebase of divergence times among organisms. Bioinformatics 22, 29712972.

Quesada, V., Ordonez, G.R., Sanchez, L.M., Puente, X.S., and Lopez-Otn, C.
(2009). The Degradome database: mammalian proteases and diseases of proteolysis. Nucleic Acids Res. 37, D239D243.

Heger, A., and Ponting, C.P. (2007). Evolutionary rate analyses of orthologs
and paralogs from 12 Drosophila genomes. Genome Res. 17, 18371849.

Rogina, B., Helfand, S.L., and Frankel, S. (2002). Longevity regulation by

Drosophila Rpd3 deacetylase and caloric restriction. Science 298, 1745.

Heide-Jrgensen, M.P., Garde, E., Nielsen, N.H., Andersen, O.N., and Hansen, S.H. (2012). A note on biological data from the hunt of bowhead whales
in West Greenland 2009-2011. J. Cetacean Res. Manag. 12, 329333.

Ross, P.L., Cheng, I., Liu, X., Cicek, M.S., Carroll, P.R., Casey, G., and Witte,
J.S. (2009). Carboxypeptidase 4 gene variants and early-onset intermediateto-high risk prostate cancer. BMC Cancer 9, 69.

Hoege, C., Pfander, B., Moldovan, G.L., Pyrowolakis, G., and Jentsch, S.
(2002). RAD6-dependent DNA repair is linked to modication of PCNA by ubiquitin and SUMO. Nature 419, 135141.

Seim, I., Ma, S., Zhou, X., Gerashchenko, M.V., Lee, S.G., Suydam, R.,
George, J.C., Bickham, J.W., and Gladyshev, V.N. (2014). The transcriptome
of the bowhead whale Balaena mysticetus reveals adaptations of the
longest-lived mammal. Aging (Albany, N.Y. Online) 6, 879899.

Holland, P.W., Garcia-Ferna`ndez, J., Williams, N.A., and Sidow, A. (1994).

Gene duplications and the origins of vertebrate development. Dev. Suppl.,
125133.

Tacutu, R., Craig, T., Budovsky, A., Wuttke, D., Lehmann, G., Taranukha, D.,
Costa, J., Fraifeld, V.E., and de Magalhaes, J.P. (2013). Human Ageing

Cell Reports 10, 112122, January 6, 2015 2015 The Authors 121

Genomic Resources: integrated databases and tools for the biology and genetics of ageing. Nucleic Acids Res. 41, D1027D1033.

Vinogradov, A.E. (1998). Genome size and GC-percent in vertebrates as determined by ow cytometry: the triangular relationship. Cytometry 31, 100109.

Tanno, M., Ogihara, M., and Taguchi, T. (1996). Age-related changes in proliferating cell nuclear antigen levels. Mech. Ageing Dev. 92, 5366.

Weeda, G., Donker, I., de Wit, J., Morreau, H., Janssens, R., Vissers, C.J.,
Nigg, A., van Steeg, H., Bootsma, D., and Hoeijmakers, J.H.J. (1997). Disruption of mouse ERCC1 results in a novel repair syndrome with growth failure,
nuclear abnormalities and senescence. Curr. Biol. 7, 427439.

Tervo, O.M., Christoffersen, M.F., Parks, S.E., Kristensen, R.M., and Madsen,
P.T. (2011). Evidence for simultaneous sound production in the bowhead
whale (Balaena mysticetus). J. Acoust. Soc. Am. 130, 22572262.
Tian, X., Azpurua, J., Hine, C., Vaidya, A., Myakishev-Rempel, M., Ablaeva, J.,
Mao, Z., Nevo, E., Gorbunova, V., and Seluanov, A. (2013). High-molecularmass hyaluronan mediates the cancer resistance of the naked mole rat. Nature
499, 346349.

West, G.B., Woodruff, W.H., and Brown, J.H. (2002). Allometric scaling of
metabolic rate from molecules and mitochondria to cells and mammals.
Proc. Natl. Acad. Sci. USA 99 (Suppl 1), 24732478.
Yang, Z. (2007). PAML 4: phylogenetic analysis by maximum likelihood. Mol.
Biol. Evol. 24, 15861591.

van Beekum, O., Gao, Y., Berger, R., Koppen, A., and Kalkhoven, E. (2012). A
novel RNAi lethality rescue screen to identify regulators of adipogenesis. PLoS
ONE 7, e37680.

Yang, Z., and dos Reis, M. (2011). Statistical properties of the branch-site test
of positive selection. Mol. Biol. Evol. 28, 12171228.

van Dam, S., Craig, T., and de Magalhaes, J.P. (2015). GeneFriends: a human
RNA-seq-based gene and transcript co-expression database. Nucleic Acids
Res. http://dx.doi.org/10.1093/nar/gku1042

Yang, Z., Wong, W.S.W., and Nielsen, R. (2005). Bayes empirical bayes inference of amino acid sites under positive selection. Mol. Biol. Evol. 22, 1107
1118.

Vervoort, V.S., Viljoen, D., Smart, R., Suthers, G., DuPont, B.R., Abbott, A., and
Schwartz, C.E. (2002). Sorting nexin 3 (SNX3) is disrupted in a patient with a
translocation t(6;13)(q21;q12) and microcephaly, microphthalmia, ectrodactyly, prognathism (MMEP) phenotype. J. Med. Genet. 39, 893899.

Yim, H.S., Cho, Y.S., Guang, X., Kang, S.G., Jeong, J.Y., Cha, S.S., Oh, H.M.,
Lee, J.H., Yang, E.C., Kwon, K.K., et al. (2014). Minke whale genome and
aquatic adaptation in cetaceans. Nat. Genet. 46, 8892.

Vilella, A.J., Severin, J., Ureta-Vidal, A., Heng, L., Durbin, R., and Birney, E.
(2009). EnsemblCompara GeneTrees: Complete, duplication-aware phylogenetic trees in vertebrates. Genome Res. 19, 327335.

Zhang, J., Zhang, Y.P., and Rosenberg, H.F. (2002). Adaptive evolution of a
duplicated pancreatic ribonuclease gene in a leaf-eating monkey. Nat. Genet.
30, 411415.

Vindelv, L.L., and Christensen, I.J. (1994). Detergent and proteolytic enzymebased techniques for nuclear isolation and DNA content analysis. Methods
Cell Biol. 41, 219229.

Zhang, J., Nielsen, R., and Yang, Z. (2005). Evaluation of an improved branchsite likelihood method for detecting positive selection at the molecular level.
Mol. Biol. Evol. 22, 24722479.

122 Cell Reports 10, 112122, January 6, 2015 2015 The Authors

933>
G=C@
47<53@
=<
B63
>C:A3
EWbV1SZZ>`Saa@SdWSea
:]]YW\UT]`OcbV]`WbObWdS`SdWSea]\bVS
T]`ST`]\b]TaQWS\QS-Bc`\b]1SZZ>`Saa
=c`SRWb]`aabOgOP`SOab]TbVSZObSab
RSdSZ]^[S\baa]bVSgQO\Q][[WaaW]\
Sf^S`bQcbbW\USRUS`SdWSea

4W\Rg]c`eOg
eWbV1SZZ>`Saa@SdWSea
eeeQSZZQ][`SdWSea

B]^`]^SZg]c``SaSO`QVT]`eO`RTOabS`bc`\b]
1SZZ>`Saa@SdWSea=c`W\aWUVbTcZOcbV]`WbObWdS
`SdWSea^cPZWaVSROQ`]aabVSZWTSaQWS\QSaW\
]c`^`W[O`g`SaSO`QVO\RB`S\RaX]c`\OZaU]
PSg]\Rag\bVSaWaO\R]TTS`O^]W\b]TdWSe

The 2016 Keystone Symposia Meeting Series

Systems Immunology: From Molecular Networks to Human Biology (A1)
Jan 1014, 2016 | Big Sky, Montana | USA

Ubiquitin Signaling (X3) joint with 1)% 0$3.LQDVH6LJQDOLQJLQ

,QDPPDWLRQ;

Cytokine JAK-STAT Signaling in Immunity & Disease (A2)

Mar 1317, 2016 | Whistler, British Columbia | Canada

Jan 1014, 2016 | Steamboat Springs, Colorado | USA

Molecular & Cellular Basis of Growth & Regeneration (A3)

Jan 1014, 2016 | Breckenridge, Colorado | USA

Nuclear Receptors: Full Throttle (J1)

joint with Metabolism, Transcription & Disease (J2)
Jan 1014, 2016 | Snowbird, Utah | USA

Biology of Down Syndrome: Impacts Across the Biomedical Spectrum (A4)

Jan 2427, 2016 | Santa Fe, New Mexico | USA

Islet Biology: From Cell Birth to Death (X5)

joint with Stem Cells & Regeneration in the Digestive Organs (X6)
Mar 1317, 2016 | Keystone, Colorado | USA

Chromatin & Epigenetics (C2)

Mar 2024, 2016 | Whistler, British Columbia | Canada
+,93HUVLVWHQFH3DWKRJHQHVLV (UDGLFDWLRQ;joint with HIV Vaccines (X8)
Mar 2024, 2016 | Olympic Valley, California | USA

Cancer Pathophysiology: Integrating the Host & Tumor Environments (C3)

Mar 28Apr 1, 2016 | Breckenridge, Colorado | USA

Traumatic Brain Injury: Clinical, Pathological & Translational Mechanisms (J3) 0RGHUQ3KHQRW\SLF'UXJ'LVFRYHU\'HQLQJWKH3DWK)RUZDUG'
joint with Axons: From Cell Biology to Pathology (J4)
Apr 26, 2016 | Big Sky, Montana | USA
Jan 2427, 2016 | Santa Fe, New Mexico | USA

Drug Discovery for Parasitic Diseases (A5)

Jan 2428, 2016 | Tahoe City, California | USA

Mitochondrial Dynamics (D2)

Apr 37, 2016 | Steamboat Springs, Colorado | USA

Small RNA Silencing: Little Guides, Big Biology (A6)

Heart Failure: Genetics, Genomics & Epigenetics (Z1)

joint with Cardiac Development, Regeneration & Repair (Z2)

Jan 2428, 2016 | Keystone, Colorado | USA

Apr 37, 2016 | Snowbird, Utah | USA

Purinergic Signaling (J5) joint with Cancer Immunotherapy:

Immunity & Immunosuppression Meet Targeted Therapies (J6)

Myeloid Cells (D3)

Apr 1014, 2016 | Killarney, County Kerry | Ireland

Jan 2428, 2016 | Vancouver, British Columbia | Canada

New Therapeutics for Diabetes & Obesity (G1)

1HXURORJLFDO'LVRUGHUVRI,QWUDFHOOXODU7UDIFNLQJ$

Apr 1720, 2016 | La Jolla, California | USA

Jan 31Feb 4, 2016 | Keystone, Colorado | USA

Gut Microbiota, Metabolic Disorders & Beyond (D4)

Cell Biology & Immunology of Persistent Infection (A8)

Apr 1721, 2016 | Newport, Rhode Island | USA

Jan 31Feb 4, 2016 | Banff, Alberta | Canada

The Cancer Genome (Q1) joint with Genomics & Personalized Medicine (Q2)

Epigenetic & Metabolic Regulation of Aging & Aging-Related Diseases (E1)

May 15, 2016 | Santa Fe, New Mexico | USA

Feb 711, 2016 | Banff, Alberta | Canada

Positive-Strand RNA Viruses (N1)

Fibrosis: From Basic Mechanisms to Targeted Therapies (Q3)

joint with Stromal Cells in Immunity (Q4)

Nucleic Acid Sensing Pathways:

Innate Immunity, Immunobiology & Therapeutics (E2)

Feb 711, 2016 | Keystone, Colorado | USA

Plant Epigenetics: From Genotype to Phenotype (B1)

Feb 1519, 2016 | Taos, New Mexico | USA

May 15, 2016 | Austin, Texas | USA

May 812, 2016 | Dresden | Germany

State of the Brain (R1)

May 2226, 2016 | Alpbach | Austria

Obesity & Adipose Tissue Biology (B2)

New Approaches to Vaccines for Human & Veterinary Tropical Diseases (M1)

Feb 1519, 2016 | Banff, Alberta | Canada

May 2226, 2016 | Cape Town | South Africa

Noncoding RNAs in Health & Disease (Q5)

joint with Enhancer Malfunction in Cancer (Q6)

B Cells at the Intersection of Innate & Adaptive Immunity (E3)

May 29Jun 2, 2016 | Stockholm | Sweden

Feb 2124, 2016 | Santa Fe, New Mexico | USA

Understanding the Function of Human Genome Variation (K1)

G Protein-Coupled Receptors: Structure, Signaling & Drug Discovery (B3)

May 31Jun 4, 2016 | Uppsala | Sweden

Feb 2125, 2016 | Keystone, Colorado | USA

Autophagy: Molecular & Physiological Mechanisms (V1)

1HZ)URQWLHUVLQ8QGHUVWDQGLQJ7XPRU0HWDEROLVP4
joint with ,PPXQRPHWDEROLVPLQ,PPXQH)XQFWLRQ ,QDPPDWRU\'LVHDVH4

Jun 59, 2016 | Whistler, British Columbia | Canada

Feb 2125, 2016 | Banff, Alberta | Canada

T Follicular Helper Cells & Germinal Centers (B4)

Feb 26Mar 1, 2016 | Monterey, California | USA

Immunity in Skin Development, Homeostasis & Disease (B5)

Feb 28Mar 2, 2016 | Tahoe City, California | USA

Tuberculosis Co-Morbidities & Immunopathogenesis (B6)

Feb 28Mar 3, 2016 | Keystone, Colorado | USA

Stem Cells & Cancer (C1)

Mar 610, 2016 | Breckenridge, Colorado | USA

Cancer Vaccines: Targeting Cancer Genes for Immunotherapy (X1)

joint with Antibodies as Drugs (X2)
Mar 610, 2016 | Whistler, British Columbia | Canada

Common Mechanisms of Neurodegeneration (Z3)

joint with Microglia in the Brain (Z4)
Jun 1216, 2016 | Keystone, Colorado | USA

Exosomes/Microvesicles: Novel Mechanisms of Cell-Cell Communication (E4)

Jun 1922, 2016 | Keystone, Colorado | USA

Translational Vaccinology for Global Health (S1)

Oct 2630, 2016 | London | United Kingdom

Phytobiomes: From Microbes to Plant Ecosystems (S2)

Nov 812, 2016 | Guaruj, So Paulo | Brazil

Hemorrhagic Fever Viruses (S3)

Dec 48, 2016 | Santa Fe, New Mexico | USA

Cellular Stress Responses and Infectious Agents (S3)

Dec 48, 2016 | Santa Fe, New Mexico | USA

www.keystonesymposia.org/meetings | 1.800.253.0685 | 1.970.262.1230

2016

SCIENTIFIC
CONFERENCES
Presenting the most signicant research on cancer etiology, prevention, diagnosis, and treatment

Fourth AACR-IASLC International Joint Conference:

Lung Cancer Translational Science
From the Bench to the Clinic
Conference Chairpersons: Karen L. Kelly and
Alice T. Shaw
Co-Chairpersons: Stephen B. Baylin, Jeffrey A.
Engelman, Roy S. Herbst, and Pierre P. Massion

*1.5*%!#+

The Function of Tumor Microenvironment in
Cancer Progression
Conference Co-Chairpersons: Raghu Kalluri, Robert A.
Weinberg, Douglas Hanahan, and Morag Park

*1.5*%!#+

Patient-Derived Cancer Models: Present and Future
Applications from Basic Science to the Clinic
Conference Co-Chairpersons: Manuel Hidalgo,
Hans Clevers, S. Gail Eckhardt, and Joan Seoane
!.1.5!3.(!*/

10th AACR-JCA Joint Conference on
Breakthroughs in Cancer Research:
From Biology to Therapeutics
Co-Chairpersons: Frank McCormick and Tetsuo Noda
!.1.5 1%
AACR Precision Medicine Series:
Cancer Cell Cycle-Tumor Progression
and Therapeutic Response
Conference Chairperson: Julien Sage
Co-Chairpersons: Karen E. Knudsen and
J. Alan Diehl
!.1.5 .$.(* +
AACR Annual Meeting 2016
Program Committee Chairperson: Scott A. Armstrong

,.%(!3.(!*/

Learn more and register at

www.AACR.org/Calendar

Pancreatic Cancer
Conference Co-Chairpersons: Manuel Hidalgo,
Christine Iacobuzio-Donahue, and
Robert H. Vonderheide
5.(* +
AACR Precision Medicine Series:
Targeting the Vulnerabilities of Cancer
Conference Co-Chairpersons: Stephen W. Fesik,
Jeffrey E. Settleman, and Paul Workman
5 %)%
Engineering and Physical Sciences in Oncology
Conference Co-Chairpersons: Rakesh Jain,
Robert Langer, and Joan Brugge

1*!+/0+*

Translational Control of Cancer:
A New Frontier in Cancer Biology and Therapy
Conference Co-Chairpersons: Jennifer A. Doudna,
Frank McCormick, Davide Ruggero, and
Nahum Sonenberg
0+!.*.*%/+

DNA Repair: Tumor Development and
Therapeutic Response
Conference Co-Chairpersons: Robert G. Bristow,
Theodore S. Lawrence, and Maria Jasin
+2!)!. +*0.!(1!!* 
EORTC-NCI-AACR Molecular Targets and
Cancer Therapeutics Symposium
+2!)!.!!)!. 1*%$!.)*5