pub_NanoSIMS_cell-press2_Cell press selection 03/11/2020 14:24 Page 4

Upgrade your knowledge on

microanalytical techniques!

The Essential Guide on

Dynamic Secondary Ion Mass
Spectrometry (SIMS)
Still the most sensitive microanalytical technique available, Dynamic
SIMS provides localized elemental, isotopic and molecular characterization
of the sample surface.
Our free Dynamic SIMS guide provides a simple introduction to the basic principles of this essential
microanalytical technique, outlining its advantages over alternative surface analysis techniques. Four
case studies allow readers to explore current practices and applications of the NanoSIMS and other
dynamic SIMS instruments including research conducted at Harvard Medical School (see extract below).

Scan the code or visit

cameca.com/focus/sims-tuto
to download a free guide
on Dynamic SIMS!

Extract from Dynamic SIMS guide, case study Nr 4:

Protein turnover quantification by feeding animals with a 15N-labeled
precursor amino acid to measure appearance of new protein.
In contrast to previous studies, direct measurement with NanoSIMS
multi-isotope imaging mass spectrometry revealed unexpectedly low
turnover of protein in hair-cell stereocili. Measured volume: 8x8x2μm.

From: Multi-isotope imaging mass spectrometry reveals slow

protein turnover in hair-cell stereocilia. Duan-Sun Zhang et al.
NATURE, vol 481, 26 January 2012

cameca.info@ametek.com
www.cameca.com
pub_NanoSIMS_cell-press2_Cell press selection 03/11/2020 14:24 Page 1

NanoSIMS 50L
Quantitative measurement
of metabolic activity and
cellular exchanges with
sub-cellular resolution
The CAMECA NanoSIMS 50L is a unique imaging Secondary Ion Mass Spectrometer enabling
quantitative measurements as well as 2D and 3D imaging of molecular distribution, kinetics and fluxes
in tissues or at intra-cellular level through a strategy of stable isotope or rare element labelling.
The NanoSIMS 50L delivers simultaneously:
• 2D & 3D imaging of molecular incorporation with 50nm resolution;
• Mapping of stable isotopes and/or trace elements tagged molecules;
• Up to 7 isotopes or masses recorded in parallel at high mass resolution.

Metabolic origin of glial lipid droplets revealed with NanoSIMS multi-isotope imaging mass spectrometry.

fold change
12
C / 13C 12
C / 13C 12
C / 13C above natural
ratio

Drosophilia larvae were raised on diets containing 13C-labeled glucose, acetate or linoleate, major carbon
sources for lipogenesis. Glial lipid droplets were then induced with 22 hr hypoxia.
NanoSIMS analyses reveal strong 13C incorporation of labeled fatty acids into the core of hypoxic lipid droplets.
De novo fatty acid synthesis contributes neutral lipid cargo to glial lipid droplets.
From: Antioxidant Role for Lipid Droplets in a Stem Cell Niche of Drosophila. Andrew P. Bailey et al.
CELL 163, 340–353, 8 October 2015

cameca.info@ametek.com
www.cameca.com
Foreword
The ability of cells to undergo metabolic reprogramming in response to their changing environment is crucial to their
function, and disruption of these complex responses underlies a growing number of metabolic diseases. However,
until recently, research has relied on bulk measurements to understand a cell’s underlying metabolic status and its
response to and interaction with its environment.

With the emergence of ever-more-sophisticated single-cell techniques, we are now beginning to get a glimpse of
the metabolic status of individual cells within a heterogeneous population and to understand the dynamics of these
cell populations in space and time. These technologies have painted a broader picture of how cells interact within
tissues and with their environment during health and disease.

In this Cell Press Selection, Metabolism at the Single-Cell Level, we showcase diverse cutting-edge techniques,
including live-cell imaging, chemical imaging, and single-cell transcriptomic technologies that have been harnessed
for understanding metabolism at the single-cell and even sub-cellular level. These technologies have been used
in diverse ways and have challenged us to ask better questions and get more complete answers to how biology
works.

This compilation represents a snapshot of the exciting research in this area. We hope you will visit cell.com and cell.
com/cell-metabolism to keep up with the ever-changing landscape.

Lastly, we are grateful for the support of Cameca, who made the publication of this collection possible.

Allyson Evans
Editor-in-Chief, Cell Metabolism

For more information about Cell Press Selections:

Gordon Sheffield
Program Director, Cell Press Selections
g.sheffield@cell.com
617-386-2189 ll
pub_NanoSIMS_cell-press2_Cell press selection 03/11/2020 14:24 Page 2

NanoSIMS 50L
A revolutionary imaging
and analytical instrument
serving medical research
and drug discovery
The CAMECA NanoSIMS 50L is a unique imaging Secondary Ion Mass Spectrometer capable of
tracking molecules and quantifying biological processes at subcellular level. It has been successfully
applied to numerous life science research projects: molecular mechanisms of lipid transport, action
mechanisms of pharmaceutical compounds, investigation of the toxicology of nanomaterials...

Multiplexed antibody imaging:

the NanoSIMS 50L provides new
insights into disease pathogenesis
that will be valuable for
basic research and
drug discovery.

False color rendering of

multiplexed antibody imaging.
Ten different rare earth
elements were conjugated
with ten different antibodies.
From: Multiplexed ion beam
imaging of human breast tumor.
M. Angelo et al. NATURE
MEDECINE, 2014 Apr, 20 (4).

Scan the code

or visit cameca.com/go/nanosims-bio
to download your free copy of our booklet
“NanoSIMS 50L, a revolution in cellular biology”

cameca.info@ametek.com
www.cameca.com
Metabolism at the Single-Cell Level
Techniques for Linking Inter- and Intra-cellular Responses to Environment
Perspectives and Reviews
Immunometabolism in the Single-Cell Era
Single-Cell RNA Sequencing in Cancer: Lessons Learned
and Emerging Challenges
Subcellular Chemical Imaging: New Avenues
in Cell Biology
GPIHBP1 and Lipoprotein Lipase, Partners in Plasma
Triglyceride Metabolism
Short Article
Age Mosaicism across Multiple Scales in Adult Tissues
Articles
Patch-Seq Links Single-Cell Transcriptomes to Human
Islet Dysfunction in Diabetes
Maxim N. Artyomov and Jan Van den Bossche
Mario L. Suvá and Itay Tirosh
Johan Decelle, Giulia Veronesi, Benoit Gallet,
Hryhoriy Stryhanyuk, Pietro Benettoni, Matthias Schmidt,
Rémi Tucoulou, Melissa Passarelli, Sylvain Bohic, Peta Clode,
and Niculina Musat
Stephen G. Young, Loren G. Fong, Anne P. Beigneux,
Christopher M. Allan, Cuiwen He, Haibo Jiang,
Katsuyuki Nakajima, Muthuraman Meiyappan, Gabriel Birrane,
and Michael Ploug
Rafael Arrojo e Drigo, Varda Lev-Ram, Swati Tyagi,
Ranjan Ramachandra, Thomas Deerinck, Eric Bushong,
Sebastien Phan, Victoria Orphan, Claude Lechene,
Mark H. Ellisman, and Martin W. Hetzer
Joan Camunas-Soler, Xiao-Qing Dai, Yan Hang,
Austin Bautista, James Lyon, Kunimasa Suzuki, Seung K. Kim,
Stephen R. Quake, and Patrick E. MacDonald
ll
Niche-Specific Reprogramming of Epigenetic
Landscapes Drives Myeloid Cell Diversity in Nonalcoholic
Steatohepatitis
NanoSIMS Analysis of Intravascular Lipolysis and Lipid
Movement across Capillaries and into Cardiomyocytes
TREM2 Regulates Microglial Cholesterol Metabolism
upon Chronic Phagocytic Challenge
Oxidative Metabolism Drives Immortalization of Neural
Stem Cells during Tumorigenesis
Jason S. Seidman, Ty D. Troutman, Mashito Sakai,
Anita Gola, Nathanael J. Spann, Hunter Bennett,
Cassi M. Bruni, Zhengyu Ouyang, Rick Z. Li, Xiaoli Sun,
BaoChau T. Vu, Martina P. Pasillas, Kaori M. Ego,
David Gosselin, Verena M. Link, Ling-Wa Chong,
Ronald M. Evans, Bonne M. Thompson, Jeffrey G. McDonald,
Mojgan Hosseini, Joseph L. Witztum, Ronald N. Germain,
and Christopher K. Glass
Cuiwen He, Thomas A. Weston, Rachel S. Jung, Patrick Heizer,
Mikael Larsson, Xuchen Hu, Christopher M. Allan,
Peter Tontonoz, Karen Reue, Anne P. Beigneux, Michael Ploug,
Andrea Holme, Matthew Kilburn, Paul Guagliardo, David A. Ford,
Loren G. Fong, Stephen G. Young, and Haibo Jiang
Alicia A. Nugent, Karin Lin, Bettina van Lengerich,
Steve Lianoglou, Laralynne Przybyla, Sonnet S. Davis,
Ceyda Llapashtica, Junhua Wang, Do Jin Kim, Dan Xia,
Anthony Lucas, Sulochanadevi Baskaran, Patrick C.G. Haddick,
Melina Lenser, Timothy K. Earr, Ju Shi, Jason C. Dugas,
Benjamin J. Andreone, Todd Logan, Hilda O. Solanoy,
Hang Chen, Ankita Srivastava, Suresh B. Poda,
Pascal E. Sanchez, Ryan J. Watts, Thomas Sandmann,
Giuseppe Astarita, Joseph W. Lewcock, Kathryn M. Monroe,
and Gilbert Di Paolo
François Bonnay, Ana Veloso, Victoria Steinmann,
Thomas Köcher, Merve Deniz Abdusselamoglu, Sunanjay Bajaj,
Elisa Rivelles, Lisa Landskron, Harald Esterbauer,
Robert P. Zinzen, and Juergen A. Knoblich
Resource
Single-Cell RNA Sequencing Maps Endothelial Metabolic
Plasticity in Pathological Angiogenesis
Katerina Rohlenova, Jermaine Goveia, Melissa García-Caballero,
Abhishek Subramanian, Joanna Kalucka, Lucas Treps,
Kim D. Falkenberg, Laura P.M.H. de Rooij, Yingfeng Zheng,
Lin Lin, Liliana Sokol, Laure-Anne Teuwen, Vincent Geldhof,
Federico Taverna, Andreas Pircher, Lena-Christin Conradi,
Shawez Khan, Steve Stegen, Dena Panovska,
Frederik De Smet, Frank J.T. Staal, Rene J. Mclaughlin,
Stefan Vinckier, Tine Van Bergen, Nadine Ectors,
Patrik De Haes, Jian Wang, Lars Bolund, Luc Schoonjans,
Tobias K. Karakach, Huanming Yang, Geert Carmeliet,
Yizhi Liu, Bernard Thienpont, Mieke Dewerchin, Guy Eelen,
Xuri Li, Yonglun Luo, and Peter Carmeliet

On the cover: Increasingly sophisticated imaging techniques now allow researchers to better
understand metabolism at the single-cell level. Pictured on the cover is an image, created with
NanoSIMS, showing transport of deuterium-labeled lipids (red) across a capillary endothelial
cell (yellow) into surrounding cardiomyocytes (blue). Image credit: Haibo Jiang, Stephen Young,
and Cuiwen He.
INSIGHTS DRIVE
DRUG DISCOVERY
Are you keeping pace with peer-reviewed research
that will affect the future of therapeutics? Are
you up to speed on the latest trends, delivery
mechanisms, and policies?

Register today to receive alerts from Research Arc, a resource for R&D
professionals that draws on the vast collection of top-tier life science
review journals published by Cell Press. It’s the perfect companion
for anyone in the biopharmaceutical community looking for targeted
scientific insights that mirror the trajectory of drug development.

cell.com/research-arc
 ll

Perspective
Immunometabolism in the Single-Cell Era
Maxim N. Artyomov1,* and Jan Van den Bossche2,*
1Department of Pathology and Immunology, Washington University School of Medicine, 660 South Euclid Avenue, St. Louis, MO 63110, USA
2Amsterdam UMC, Vrije Universiteit Amsterdam, Department of Molecular Cell Biology and Immunology, Amsterdam Cardiovascular
Sciences, Cancer Center Amsterdam, De Boelelaan 1108, 1081HZ Amsterdam, the Netherlands
*Correspondence: martyomov@wustl.edu (M.N.A.), j.vandenbossche@amsterdamumc.nl (J.V.d.B.)
https://doi.org/10.1016/j.cmet.2020.09.013

SUMMARY

Emerging research has identified metabolic pathways that are crucial for the proper regulation of immune
cells and how, when deranged, they can cause immune dysfunction and disease progression. However,
due to technical limitations such insights have relied heavily on bulk measurements in immune cells, often
activated in vitro. But with the emergence of single-cell applications, researchers can now estimate the meta-
bolic state of individual immune cells in clinical samples. Here, we review these single-cell techniques and
their ability to validate common principles in immunometabolism, while also revealing context-dependent
metabolic heterogeneity within the immune cell compartment. We also discuss current gaps and limitations,
as well as identify future opportunities to move the field forward toward the development of therapeutic tar-
gets and improved diagnostic capabilities.

INTRODUCTION sion, researchers need to be able to resolve the trajectory by

Metabolism is at the core of all biological processes and can be
broadly subdivided into anabolic processes that use energy and
building blocks for biosynthesis, and catabolic processes that
provide the raw material for anabolism as well as produce energy
to fuel biological functions. In addition to these functions, meta-
bolic enzymes and intermediates have important regulatory roles
and are major controllers of cellular behavior. Recent studies in
the field of immunometabolism have demonstrated a close
connection between metabolic programs and the specific im-
mune functions they support during both health and disease
(Ayres, 2020; Buck et al., 2017; Makowski et al., 2020; O’Neill
et al., 2016). Dysregulation of these core metabolic programs
is linked to many modern diseases, including cancer and chronic
inflammatory metabolic diseases such as diabetes, obesity,
atherosclerosis, and rheumatoid arthritis (Ketelhuth et al.,
2019; Koelwyn et al., 2018; Lim et al., 2020; Tabas and Bornfeldt,
2020; Turbitt et al., 2020). Yet a detailed understanding of the im-
munometabolic remodeling that occurs in pathogenic settings is
mostly lacking, largely due to the complex nature of the diseases
and involved cellular phenotypes.
Alterations in nutrient levels, oxygen availability, the presence
of signaling factors, and crosstalk with neighboring cells induce
metabolic changes that allow the cell to function in its specific
tissue microenvironment (Caputa et al., 2019; Van den Bossche
and Saraber, 2018). As every cell resides in a unique environ-
ment, no single cell within our body is completely identical meta-
bolically, phenotypically, and functionally. Addressing the
spatiotemporal aspect of immunometabolism and unravelling
this complexity at a single-cell resolution will move the immuno-
metabolism field forward, allowing translation to clinical applica-
tions and understanding the fundamentals of in vivo immu-
nology. Indeed, to fully understand the mechanisms by which
cellular metabolism regulates immunity and disease progres-
which immune cells metabolically transit into their effector
phenotype. Arguably, single-cell profiling technology possesses
the power to address all of the above questions, even at the cur-
rent state of development.
In this Perspective, we explore how current and emerging
state-of-the-art technologies contribute to our understanding
of metabolic regulation of immunity and discuss the transition
from bulk to single-cell immunometabolic profiling (Ahl et al.,
2020; Hartmann et al., 2020; Levine et al., 2020; Miller et al.,
2017; Xiao et al., 2019).
COMMON BULK METABOLIC ANALYSES
Extracellular Flux Analysis
Extracellular flux analyzers, such as the Seahorse apparatus,
perform real-time recordings of the extracellular acidification
rate (ECAR) and oxygen consumption rate (OCR) as key read-
outs of metabolism, and as such provide indirect measurements
of glycolysis and mitochondrial respiration, respectively. While
the use of specific substrates and inhibitors allows one to inves-
tigate the relative preference of a cell for glucose, glutamine, or
fatty acids, extracellular flux analysis does not yield detailed in-
formation about the activity of metabolic pathways outside of
glycolysis and the TCA cycle within the mitochondria. Neverthe-
less, the fact that this technology allows the easy, fast, and
reasonably affordable profiling of cells in a 96-well format
strongly contributed to the expansion of the immunometabolism
field (Pelgrom et al., 2016; Van den Bossche et al., 2015; van der
Windt et al., 2016). Some of the key findings obtained using the
Seahorse instrument include the glycolytic switch and mitochon-
drial dysfunction in inflammatory macrophages, increased
glycolysis in effector T cells and activated dendritic cells (DCs),
and the heightened mitochondrial respiration in T memory cells
and IL-4-activated macrophages (Everts et al., 2014; Huang

Cell Metabolism 32, November 3, 2020 ª 2020 Elsevier Inc. 1

 ll
et al., 2014; Van den Bossche et al., 2016; van der Windt et al.,
2012). Noteworthy, ECAR is a surrogate marker for glycolysis,
and it is important to stress that acidification may not always
result from enhanced glycolysis. Indeed, CO2 production within
respiring mitochondria and/or extracellular release of TCA cycle
intermediates (e.g., succinate) also acidifies the culture medium.
This limitation can be overcome by performing targeted metab-
olomics or fluxomics for the glycolysis pathway.
Steady-State Metabolomics
For a given time point, metabolomics measures steady-state
levels of a broad range of metabolites via liquid or gas chroma-
tography-mass spectrometry (LC-MS or GC-MS). While untar-
geted metabolomics measures hundreds of metabolites in bio-
logical samples, targeted metabolomics assesses preselected
metabolites and yields data with higher sensitivity and allows
for accurate absolute quantification of metabolite concentra-
tions. In addition to measuring small metabolites, high-resolution
MS-based profiling has been recently applied in ‘‘shotgun’’ lipi-
domics to profile the lipidome of differentially activated macro-
phages (Dennis et al., 2010; Hsieh et al., 2020). Most MS-based
approaches still require relatively high levels of input material (on
the order of 100,000s of cells), yet recent developments provide
an avenue to spatially measure metabolites at a single-cell reso-
lution. For more in-depth information on the recent develop-
ments in the field of spatial and single-cell metabolomics, we
refer to excellent reviews on key emerging methods, including
matrix-assisted laser desorption ionization (MALDI)-MS imaging
and secondary ion mass spectrometry (SIMS) (Duncan et al.,
2019; Gilmore et al., 2019). Combining such analysis with the sin-
gle-cell approaches discussed in this Perspective will certainly
help to unravel the immunometabolism at single-cell resolution
within complex tissue microenvironments in vivo.
Fluxomics
Altered flow through a given pathway does not necessarily corre-
late directly with metabolite levels within this pathway as
measured by metabolomic profiling at a set time point. The use
of stable isotope 13C-labeled or 15N-labeled substrates and
measurement of isotopologues through targeted MS at serial
time points allows one to determine metabolic fluxes and
pathway dynamics. While metabolomics provides a ‘‘snapshot’’
of the metabolites present at one particular moment, 13C/15N la-
bel-tracing (or so-called fluxomics) could be regarded as a
‘‘movie’’ that yields insight into the relative flux through distinct
metabolic paths. The use of such analyses is still in its infancy
at single-cell resolution but can be readily used by the commu-
nity to quantify the rates of specific metabolic reactions in bulk
approaches such as the mass isotopomer multi-ordinate spec-
tral analysis (MIMOSA fluxomics) as developed in the Kibbey
Lab (Alves et al., 2015).
KEY IMMUNOMETABOLISM CONCEPTS DERIVED
FROM BULK ANALYSES
The descriptions of the distinct metabolic pathways below pro-
vide minimal immunometabolism background for the distinct
metabolic proteins that are most frequently targeted in the sin-
gle-cell metabolic profiling approaches discussed later in this
2 Cell Metabolism 32, November 3, 2020
Perspective
article. For more details on immunometabolism, we refer to
excellent and more comprehensive reviews on this subject in
diverse immune cells and diseases (Ayres, 2020; Buck et al.,
2017; Makowski et al., 2020; O’Neill et al., 2016).
Glycolysis Supports Diverse Immune Effector Functions
Immune activation is an energy-demanding process that is typi-
cally accompanied by an increase in glycolytic flux. Glycolysis
starts with the uptake of glucose by glucose transporters (e.g.,
GLUT1, which is predominant in immune cells) and subsequent
processing in the cytosol to yield pyruvate, generating ATP and
reducing NAD+ to NADH during the process (Figure 1A). Elegant
research highlights the direct control of immune effector func-
tions by the glycolytic enzymes hexokinase 1 and 2 (HK1 and
HK2), glyceraldehyde-3-phosphate dehydrogenase (GAPDH),
enolase, and the pyruvate kinase isoenzyme M2 (PKM2) (Chang
et al., 2013; Galván-Peña et al., 2019; Moon et al., 2015; Pals-
son-McDermott et al., 2015). To maintain glycolytic flux in the
absence of mitochondrial oxidative phosphorylation (OXPHOS),
cells can reduce pyruvate to lactate via lactate dehydrogenase
(LDH) to recycle NADH into NAD+. This fermentation reaction
can be measured as ECAR in Seahorse analysis when protons,
along with glycolysis-derived lactate, are exported from the
cell by monocarboxylate transporter MCT1. While glycolysis is
most prominent in inflammatory immune effector cells (Krawczyk
et al., 2010; Michalek et al., 2011), it is critical for diverse modes
of immune activation, including alternative macrophage activa-
tion and the induction of T regulatory (Treg) cells, and glycolytic
rates differ between B and T lymphocytes (Caro-Maldonado
et al., 2014; De Rosa et al., 2015; Huang et al., 2016; Van den
Bossche et al., 2016) (Figure 1B).
The Pentose Phosphate Pathway Is Particularly
Important in Inflammatory Myeloid Cells
The pentose phosphate pathway (PPP) branches off from the
glycolytic pathway when hexokinase-derived glucose-6-phos-
phate is oxidized by G6P dehydrogenase (G6PD), the rate-
limiting step of the PPP (Figure 1A). This oxidative branch gen-
erates reducing equivalents of NADPH that have multiple roles
in immune cells. Neutrophils and inflammatory macrophages
use the PPP and especially NADPH oxidase to generate reac-
tive oxygen species (ROS) to combat infectious agents
(Figure 1B). Moreover, NADPH is used to generate fatty acids
(through fatty acid synthesis, FAS) and anti-oxidants, such as
glutathione, that prevent excessive tissue damage. Nucleotide
and amino acid precursors needed to support cell proliferation
are produced in the non-oxidative branch of the PPP. Together,
both branches of the PPP support anabolic programs and im-
mune effector functions (Baardman et al., 2018; Haschemi
et al., 2012).
The TCA Cycle and ETC as a Central Hub
The TCA cycle and mitochondrial electron transport chain (ETC)-
based energy production are key elements of life as most cells
rely on energy production via OXPHOS, which is typically evalu-
ated measuring OCR. Compared to glycolysis, OXPHOS is much
more efficient in generating energy, and thus this pathway is typi-
cally associated with the longevity of homeostatic immune cells
such as alternatively activated macrophages or B and T memory
 ll
Perspective

Figure 1. Key Metabolic Pathways in Immune Cell Subtypes

(A) Overview of the most important metabolic pathways regulating immune cells, including the metabolic targets that were used in recent single-cell profiling
approaches.
(B) Summary of the metabolic characteristics and needs of various immune cell subsets as detailed in the text and as reviewed earlier elsewhere (O’Neill et al.,
2016; Van den Bossche et al., 2017).

cells (Lam et al., 2016; van der Windt et al., 2013). Yet during pro-
inflammatory activation of macrophages, the TCA cycle is signif-
icantly remodeled via downregulation of IDH and SDH activity
(Jha et al., 2015). The latter is achieved through direct inhibition
by the immunoregulatory metabolite itaconate that is specifically
produced by ACOD1/IRG1 in activated myeloid cells (Lampro-
poulou et al., 2016; O’Neill and Artyomov, 2019; Swain
et al., 2020).
Fatty Acid Metabolism Supports Immune Cell
Phenotypes and Functions
Fatty acid oxidation (FAO) is a catabolic pathway converting
fatty acids into products like acetyl-CoA, NADH, and FADH2,
which are used in the mitochondria to generate energy. Carni-
tine palmitoyl transferase 1 and 2 (CPT1 and CPT2) shuttle
long-chain fatty acids into the mitochondrial matrix for subse-
quent oxidation to acetyl-CoA by hydroxyacyl-CoA

Cell Metabolism 32, November 3, 2020 3

 ll
dehydrogenase (HADHA). Experiments using often too high
concentrations of the CPT1-inhibitor etomoxir suggested that
FAO is particularly important in IL-4-induced macrophages,
memory T cells, and Tregs (Michalek et al., 2011; van der Windt
et al., 2012; Vats et al., 2006). Yet recent literature employing
cell-specific genetic knockouts for CPT1 and CPT2 reports
that FAO is mostly dispensable for these processes (Divakaruni
et al., 2018; Nomura et al., 2016; Raud et al., 2018; Van den
Bossche and van der Windt, 2018).
As opposed to FAO, FAS is an anabolic pathway converting
cytosolic acetyl-CoA into lipids. Acetyl-CoA carboxylase (ACC)
first carboxylates acetyl-CoA to malonyl-CoA, which is subse-
quently elongated by fatty acid synthase (FASN). FAS supports
cellular proliferation of effector T cells and is key to configuring,
the plasma membrane for inflammatory signaling in macro-
phages, and for endoplasmic reticulum synthesis to allow cyto-
kine secretion by activated DCs (Everts et al., 2014; Wang
et al., 2011).
Amino Acid Metabolism Is Differentially Regulated in
Distinct Immune Cells
Immune activation of T cells is associated with an increased
demand for amino acid metabolism, as exemplified by the
importance of L-type amino acid transporter 1 (LAT1, CD98,
and SLC7A5) and the serine pathway for successful signaling
upon TCR engagement (Hayashi et al., 2013; Ma et al., 2017,
2019). Furthermore, amino acids play a role in fine-tuning the
specific direction of immune responses. For instance, both
Th1 and Th17 cells increase glutamine usage via the trans-
porter protein ASCT2 in response to antigen receptor stimula-
tion, whereas anti-inflammatory Tregs are not affected by
altered glutamine supply (Nakaya et al., 2014). Glutaminase
(GLS) converts glutamine into glutamate to fuel the TCA cycle,
and this pathway promotes Th17 differentiation while diminish-
ing Th1 and cytotoxic T lymphocyte differentiation (Johnson
et al., 2018).
In macrophages, amino acids also play important functional
and regulatory roles. Glutamine metabolism has been associ-
ated with anti-inflammatory polarization programs, and serine
metabolism was implicated in the regulation of IL-1b produc-
tion (Jha et al., 2015; Rodriguez et al., 2019). Furthermore,
amino acids can also be used for the production of effector
molecules in macrophages. LPS(+IFNg)-induced macro-
phages mainly convert arginine into nitric oxide (NO) via
inducible NO synthase (iNOS), while IL-4-stimulated macro-
phages predominantly metabolize arginine into ornithine and
polyamines (Munder et al., 1998; Van den Bossche et al.,
2012). For more information on how amino acids regulate im-
munity, we refer to an excellent recent review on this topic
(Kelly and Pearce, 2020).
Importantly, most of the concepts described so far were
derived from bulk measurements in immune cells that were
cultured and activated in vitro. As discussed below, tech-
nological limitations largely prevented the validation of cur-
rent dogmas in vivo and the determination of the spatio-
temporal facets of immunometabolism. Next, we describe
the single-cell approaches addressing these key open
questions to push the immunometabolism field to the
next level.
4 Cell Metabolism 32, November 3, 2020
Perspective
LIMITATIONS OF CURRENT APPROACHES AND
ASSOCIATED GAPS IN OUR KNOWLEDGE
The Need for In Vitro to In Vivo Validation
Given the importance of environmental factors, immune cell
metabolism in laboratory cell culture is obviously different than
in vivo. Indeed, Ma et al. recently used a 13C-glucose infusion
method in mice to investigate the metabolism of CD8+ T cells
in response to Listeria infection and observed fundamentally
different metabolic profiles for activated T cells in vivo (Ma
et al., 2019). While in vitro-activated T cells show a shift away
from OXPHOS toward glycolysis, CD8+ T cells activated in vivo
display higher rates of oxidative metabolism and a reliance on
glucose-dependent biosynthesis of the amino acid serine. The
approach developed in this work requires delicate experimental
implementation and is not readily scalable or applicable to hu-
man samples. This translation to the human setting is important
since animal models have their limitations and do not always
replicate human immunology and biology. Generally, the further
development of in vivo (or ex vivo) assessments of cellular meta-
bolism with these and other techniques has a high priority in the
field. Meanwhile, the validation of key concepts in immunome-
tabolism research in vivo, especially in humans, requires alter-
nate options for metabolic profiling.
The Need to Understand the Spatiotemporal Aspects of
Immunometabolism
Immune cells heavily rely on a timely supply of nutrients, which
can differ depending on the location of the cell within the micro-
environment of a tumor or inflammation site (Caputa et al., 2019;
Van den Bossche and Saraber, 2018). Sorting and bulk measure-
ments inevitably average profiles over cell populations, which
obscures the effects of timing and environmental variables. As
such, understanding the cell’s metabolic state at single-cell res-
olution can explain the molecular mechanisms underlying their
phenotypic heterogeneity within complex tissue microenviron-
ments in vivo. This could help to understand the reasons behind
their functioning, or their inability to function properly in patho-
logical settings. Most recent transcriptional and imaging tech-
nologies can resolve the spatial arrangement of individual cells
or even metabolites, and their further development will be critical
to establish the next level of understanding of immunometabolic
remodeling in human samples and in in vivo settings (Mazumdar
et al., 2020; Van den Bossche et al., 2017). Timing is another
aspect that is often overlooked in the immunometabolism field.
We know most immune cells rapidly ramp up glycolysis upon
activation, and this ultimately leads to the increased metabolic
fluxes and abundance of specific metabolites in their effector
state (typically assessed after 24 h). Yet how this happens over
time and how it supports the acquisition of the effector pheno-
type and disease progression is not well understood. Resolving
the kinetics by which immune cells metabolically transit into their
effector phenotype is essential to define which events are the
cause and what are the subsequent consequences. Together,
understanding this spatiotemporal aspect of immunometabo-
lism will not only revolutionize our knowledge of fundamental
biology, but also aid the rational design of new therapeutic ap-
proaches.

Perspective
Cell Numbers and Sorting Time Limitation
The lack of in vitro to in vivo transitions and mouse-to-human
validations can be at least partially explained by the techno-
logical limitations of the commonly used bulk analysis tech-
niques in immunometabolism research. A major caveat of
both metabolomics and extracellular flux analysis is the
need for relatively high cell numbers that are often not present
in human clinical samples. For example, the Seahorse XF
analyzer measures extracellular fluxes in tens to hundreds of
thousands of cells per well and typically requires 4–5 replicate
wells per condition. For metabolomics, researchers typically
measure half a million pooled cells per replicate, and even
with this high input, the metabolic coverage is limited as we
typically measure a few hundred metabolites out of the esti-
mated thousands present in the cell. Another drawback of
extracellular flux analysis and metabolomics is the fact that
both assess metabolic features that are not very stable. As
such, the obtained results can be strongly affected by lengthy
and harsh tissue digestion and fluorescence-activated cell
sorting protocols (Llufrio et al., 2018). Therefore, new meta-
bolic profiling methods accompanied by novel, optimized
cell isolation approaches are needed to measure stable meta-
bolic features within small quantities of immune cells ex vivo.
Leveraging Different ‘omics Approaches for Indirect
Metabolic Assessment
The instability of metabolite levels can be at least partially
overcome by integrating metabolomics with other ‘omics
data that are more stable during processing such as tran-
scriptomics and proteomics. Metabolic fluxes can be re-
garded as the net result of metabolite levels and the pres-
ence and activity of the enzymes that metabolize them.
Gene expression analysis and proteomics can already give
some insight into the regulation of metabolic pathways but
become particularly powerful when integrated with metabo-
lomics data. Using a network integration approach, Jha
et al. identified an important metabolic break in the TCA cy-
cle that supports inflammatory macrophage responses (Jha
et al., 2015). Baardman et al. also combined metabolomics
and transcriptomics data to demonstrate that a defective
PPP is associated with decreased inflammatory responses
in lipid-laden macrophages (Baardman et al., 2018). Simi-
larly, integration of bulk proteomics and metabolomics re-
vealed the metabolic requirements for T cell activation (Gei-
ger et al., 2016). These distinct ‘omics approaches,
especially when integrated, can reveal (de)regulated meta-
bolic hubs that can be functionally tested using genetic
tools or pharmacological compounds.
Single-cell metabolomics is currently too premature for large-
scale applications, and thus alternate approaches are needed
for immunometabolism research to enter the single-cell era. Pre-
vious successes using bulk transcriptomics and proteomics to
delineate cellular metabolism show that using the related sin-
gle-cell ‘omics approaches is the way to go at the moment. To
this end, the latest single-cell RNA sequencing (scRNA-seq)
and cytometry-based approaches that map the metabolic land-
scape at single-cell resolution are discussed in the next parts of
this article.
ll
METABOLIC ANALYSIS BASED ON SINGLE-CELL
TRANSCRIPTOMICS
With the advent of scRNA-seq, one can resolve transcriptional
profiles of individual cells within complex multicellular ecosys-
tems, which is critically important in an immunological context.
Using the common 10x Genomics pipeline, scRNA-seq exper-
iments nowadays profile up to ten thousand cells per sample,
routinely detecting two to four thousand genes per cell. Even
direct examination of scRNA-seq data in an immunological
context can be very illuminating with regard to the key meta-
bolic processes that occur in vivo. For example, we can use
data on the immune cells within murine sarcomas that were
treated either with immune checkpoint blockade (ICB) anti-
PD-1/anti-CTLA-4 antibodies or with control antibodies (Gubin
et al., 2018). In this system, ICB treatment leads to tumor
rejection, which is paralleled by overall activation of the im-
mune compartment and the appearance of activated cells
within the myeloid compartment (Gubin et al., 2018). As illus-
trated in Figures 2A and 2B, we created a visualization of
the information from this scRNA-seq dataset to demonstrate
the behavior of some key metabolic genes in the immune
cell subsets, notably:
(i) Genes associated with increased glycolytic activity like
Glut1 are enriched in the activated macrophages that
appear after ICB treatment.
(ii) Nos2, a metabolic gene characteristic of inflammatory
macrophage activation, is enriched in macrophages that
are specific to the ICB treatment condition.
(iii and iv) Phgdh expression is associated with Mki67 ex-
pressing highly proliferative T cells, which is
consistent with bulk ex vivo analyses describing
the role of serine metabolism in T cell proliferation
(Ma et al., 2017, 2019).
(v and vi) Levels of NRF2 transcript (Nfe2l2), a key factor in
oxidative stress responses, are generally enriched
in macrophages relative to T cells and do not
appear to depend on the treatment condition.
However, its direct targets such as Hmox1 do in-
crease in subpopulations of inflammatory macro-
phages, consistent with potential oxidative/elec-
trophilic stress from NO and itaconate. Together,
this is in line with the fact that NRF2 is mainly regu-
lated on the protein level (Taguchi et al., 2011) and
highlights the need to profile protein rather than
gene expression for this key metabolic regulator.
A similar approach of direct examination of specific key meta-
bolic genes in scRNA-seq data proved to be powerful in
describing the metabolic aspects of apoptosis during Drosophila
eye development (Ariss et al., 2018). Together, these examples
illustrate the potential utility of scRNA-seq data for studying
metabolism. Yet methods to comprehensively map the meta-
bolic landscape at single-cell resolution remain relatively scarce
and/or are in the early stages of development. Most approaches
for metabolic analysis of scRNA-seq data arise from bulk RNA-
seq analysis techniques, and considerable challenges have to
be overcome when making this transition.
Cell Metabolism 32, November 3, 2020 5
 ll
Indeed, bulk transcriptional profiling provides an appealing
source of data as it yields exhaustive information about tran-
script levels within the sequenced cells. As such, the absence
of a signal typically implies absence of the corresponding tran-
script. This is in contrast to metabolomics profiling, where
coverage is limited and absence of a signal often arises due to
technological limitations rather than due to absence of the
metabolite itself. Yet this advantage of transcriptional profiling
is not as pronounced in scRNA-seq data. Due to generally lower
depth of sequencing and drop-out events, expression levels of
individual genes can appear zero in some cells even though
these genes are actually expressed. Typical depth of sequencing
is in the order of 25,000–50,000 reads per cell, which yields
confident detection of 3,000–4,000 genes per cell. Shallow
sequencing (i.e., on the order of 1,000 genes per cell) might result
in detection of mostly common and housekeeping genes and
only very few subpopulation-specific genes, which would lead
to significant difficulties when comparing the data with other
published signatures. Still, even the detection of 3,000–4,000
genes per cell is significantly lower than the estimated 12,000–
15,000 genes expressed in purified cell populations based on
the bulk RNA-seq data. In these situations, techniques such as
data imputations have to be used to post-process the data
and to obtain smooth transcriptional profiles. Within the same
scRNA-seq dataset different cell clusters can have very different
6 Cell Metabolism 32, November 3, 2020
Perspective
Figure 2. Expression of Some Key
Metabolic Genes in Intratumoral Immune
Cells Based on Published Single-Cell
Transcriptomics Data
The behavior of some key metabolic genes is
shown in immune cells of mouse sarcomas that
were treated with either anti-PD1/anti-CTLA4 im-
mune checkpoint blockade antibodies or control
antibody (Gubin et al., 2018) (A). To do so, we
created a direct visualization of the indicated
genes from the associated published dataset
(GEO: GSE119352) (B).
RNA content per cell, which results in un-
even coverage between cell types. For
example, a smaller number of total tran-
scripts is typically detected in neutrophils
as compared to T cells or macrophages.
Thus, care is required when comparing
pathway enrichments between different
cells since lower coverage of some clus-
ters can lead to false-negative results.
Most of the approaches for metabolic
analysis of scRNA-seq data can be
broadly classified into two main cate-
gories that are briefly described below:
(1) pathway-based analysis and (2) flux
balance analysis (FBA)-based methods.
Pathway-Based Approaches
Genes composing distinct metabolic
pathways are fairly well annotated (Kane-
hisa and Goto, 2000) and can be directly
probed as gene sets using standard bio-
informatics pathway enrichment techniques. This simple yet
powerful approach can reveal novel biological insights. For
example, the role of serine metabolism in T cells has been inves-
tigated based on the initial observation of the transcriptional
enrichment of this pathway in published bulk RNA-seq data
(Ma et al., 2017). As Figure 2 shows, this pathway is enriched
in the proliferating T cell cluster within the scRNA-seq data.
Recently, Xiao et al. demonstrated how pathway enrichment
pipelines can be adjusted for scRNA-seq data analysis and
used it to characterize the metabolic diversity within the tumor
microenvironment. Using a similar analysis strategy, Miragaia
et al. (2019) highlighted the metabolic heterogeneity among
Tregs across different tissues using pathway enrichments.
Flux Balance Analysis-Based Approaches
FBA is an elegant way to compute the steady-state distribution of
metabolic fluxes in the system that would satisfy constraints on
input and output fluxes given a specific network architecture
(Orth et al., 2010). Simply put, FBA views metabolic wiring in a
cell as a system of pipes with a limited number of input and output
faucets and requires that there is steady-state equilibrium estab-
lished at every node of a system. In other words, incoming meta-
bolic flow toward each reaction should equal metabolic flow going
away from that reaction. As such, specific expression levels of in-
dividual enzymes are not directly participating in modeling and the

Perspective
results are mostly dependent on network topology and require-
ments on input and output reactions of the model. Inputs are typi-
cally the most common substrate uptake reactions (e.g., glucose,
glutamine), and outputs (usually referred as objective function) are
associated with the specific studied cell behavior. For instance, an
objective can be to maximize the production of biomass for
growing cells or the production of certain metabolites such as
NO for activated macrophages (Bordbar et al., 2012).
An important advantage of the FBA-based approaches is that
they rely on network analysis that is not associated with prede-
fined pathway knowledge and therefore can reveal novel biolog-
ical concepts. As such, it provides a powerful tool to understand
changes in metabolic flux upon changes in network topologies,
e.g., due to gene knockout or tissue-specific pattern of expres-
sion. Indeed, Zhang et al. have demonstrated scRNA-seq-based
analysis of NAD+ biosynthesis in different tissues using an FBA-
based approach (Zhang et al., 2020). Conceptually, this work fol-
lows from the approach of studying tissue-specific metabolism
based on bulk RNA-seq data from different tissues (Bordbar
et al., 2011). In addition to looking at the dramatically different
network topologies, the FBA framework allows investigation of
metabolic remodeling within the same network, but upon the
change of the required cellular output induced by a stimulation.
This is feasible when at least some key metabolic features are
known. For instance, using this approach, Bordbar et al.
modeled activation of macrophages using increase in NO pro-
duction as a key objective of inflammatory macrophages (Bord-
bar et al., 2012). Key objectives can also be defined based on the
differential expression between conditions. This approach for
bulk data has previously been adopted to reveal itaconate as
an SDH inhibitor (Lampropoulou et al., 2016) and has recently
been translated to scRNA-seq data to investigate metabolic dif-
ferences associated with pathogenic/non-pathogenic Th17
cells, as well as their difference with Tregs (Wagner et al., 2020).
One significant caveat of the current approaches is that they
typically focus on differences between two distinct cell popula-
tions. In these situations, a more straightforward approach
would be to simply perform differential expression analysis be-
tween the two populations and perform either pathway enrich-
ment analysis or metabolic subnetwork analysis. We argue that
it is critical to move past the one-to-one comparison of known
cell types and to learn to capture metabolic diversity within the
whole dataset. Thus, the next generation metabolic scRNA-
seq analysis approaches is expected to reveal major metabolic
axes of variation independent of the assignments of the cellular
identities. While it is feasible at the moment to perform such clus-
tering simply in the space of all metabolic genes or using anno-
tated metabolic pathways as principal components, the most
powerful clustering approaches would incorporate network-
based analysis with no a priori knowledge of individual path-
ways. This would allow for unbiased definition of individual cells
as distinct metabolic entities and will provide an unbiased view of
the whole cellular ecosystem from a metabolic perspective.
PROTEIN-BASED SINGLE-CELL METABOLIC ANALYSIS
Single-Cell Metabolic Profiling by Mass Cytometry
While untargeted analysis of proteins within single cells is still in
its infancy (Slavov, 2020) and has not been applied in immuno-
ll
metabolism research yet, recent literature showcases the use
of mass cytometry (cytometry by time of flight, CyTOF) to esti-
mate the metabolic configurations of enzymes within single cells.
This technology allows for simultaneous quantification of 40
parameters at single-cell resolution (Hartmann and Bendall,
2020). In the context of immunometabolism, one should
consider the inclusion of antibodies targeting the main metabolic
features described above (Figure 1). Not surprisingly, the meta-
bolic panels that were independently designed by three distinct
research groups show considerable overlap (Table 1) (Ahl et al.,
2020; Hartmann et al., 2020; Levine et al., 2020).
To design an immunometabolic CyTOF panel, Hartmann et al.
first screened over 100 commercial antibodies against a broad
range of metabolic factors and came to a selection of 41 meta-
bolic antibodies that passed all biological controls and yielded
robust reproducible results in a proof-of-concept study of hu-
man white blood cells (Hartmann et al., 2020). After identifying
distinct immune cell subsets based on their lineage marker
expression pattern, the metabolic states of all immune cell sub-
sets could be estimated and compared without the need for prior
sorting. The metabolic state profiles acquired by mass cytometry
were in agreement with their described roles in specific immune
functions as exemplified by high expression levels of targets
associated with glucose and fatty acid metabolism in plasmacy-
toid DCs, and high levels of the rate-limiting PPP enzyme G6PD
in neutrophils (O’Neill et al., 2016). Furthermore, Hartman et al.
have demonstrated a high degree of consistency between meta-
bolic changes in T cells measured by Seahorse and the changes
observed at a single-cell resolution on the protein level. This
highlights that CyTOF-based single-cell metabolic profiling is a
powerful tool to gain new insights into the concepts of metabolic
remodeling established by bulk measurements.
Application of Metabolic CyTOF in the Ex Vivo Clinical
Context
Comparing tumor and healthy adjacent tissue from colorectal
carcinoma patients, along with healthy donor peripheral blood
mononuclear cells and lymph nodes, Hartmann et al. (2020) re-
vealed specific metabolic phenotypes that were enriched in tu-
mor-associated CD8+ T cells. These cells showed increased
expression of the large neutral amino acid transporter 1 (LAT1,
CD98) and decreased expression of a broad range of enzymes
across distinct metabolic pathways. These so-called metalow
cells displayed distinct signs of T cell exhaustion, including
increased expression of PD1 and CD39, reduced mitochondrial
capacity, and low levels of TCF1. Of note, a fraction of the met-
alow subset did not express PD1 or CD39, indicating that
combining metabolic and immune cell markers provides addi-
tional dimensions to phenotypically and functionally define tu-
mor-associated immune cells in relation to human disease.
Use of CyTOF-Based Metabolic Profiling in Animal
Model Research
In addition to human samples, metabolic CyTOF panels can be
useful to dissect metabolic remodeling in vivo in animal models.
Levine et al. used CyTOF-based metabolic profiling of T cell re-
sponses during Listeria monocytogenes infection as a well-
described model of CD8 T cell differentiation (Levine et al.,
2020). The metabolic panel configured here was largely based
Cell Metabolism 32, November 3, 2020 7
 ll
Perspective

Table 1. Metabolic Targets Used in Different Single-Cell Profiling Approaches

Metabolic Pathway and Inclusion in Panels
Target Function Hartmann Levine Ahl Miller Corea
Glycolysis
GLUT1 glucose uptake x x x x
HK first glycolytic enzyme x x (x)
PFK2 glycolytic enzyme and x
regulator
GAPDH glycolytic enzyme/ x x X x
immune regulator
PKM2 glycolysis/transcriptional x
regulator
Fermentation
LDH pyruvate to lactate x x (x)
conversion
MCT1 lactate export x (x)
PPP
G6PD G6P dehydrogenase, x x x x
rate-limiting PPP step
TCA/ETC
CS citrate synthase x x (x)
IDH isocitrate x x x x
dehydrogenase
OGDH alpha-ketoglutarate x (x)
dehydrogenase
SDH succinate x x
dehydrogenase/ETC
complex II
ATP5A ATP synthase/ETC x x x x
complex V
CytC cytochrome c, essential x x (x)
for ETC function
Mitochondrial Regulation
PGC1a mitochondrial biogenesis x
VDAC1 outer mitochondrial x x (x)
membrane porin
OPA1 mitochondrial fusion x
Fatty Acid Metabolism
CD36/FAT fatty acid transporter x
CPT1A fatty acid shuttling into x x x
mitochondria
HADHA long-chain fatty acid x
oxidation
ACADM medium-chain acyl-CoA x x
dehydrogenase
ACC/ACAC acetyl-CoA carboxylase/ x x x
fatty acid synthesis
Amino Acid Metabolism
CD98/LAT1 essential amino acid x x x
transporter
GLUD glutamate x
dehydrogenase
ASCT2 alanine/serine/cysteine- x
preferring amino acid
transporter 2
(Continued on next page)

8 Cell Metabolism 32, November 3, 2020

 ll
Perspective

Table 1. Continued
Metabolic Pathway and Inclusion in Panels
Target Function Hartmann Levine Ahl Miller Corea
GLS glutamine to glutamate x
conversion
GOT2 glutamic-oxaloacetic x
transaminase/malate-
aspartate shuttle
ASS1 arginosuccinate x
synthase
Metabolic Regulation/Signaling
NRF1 anti-oxidant transcription x x (x)
factor
NRF2 anti-oxidant transcription x
factor
KEAP1 NRF2 degradation x
PRDX2 peroxiredoxin 2 anti- x
oxidant enzyme
HIF1A hypoxia and x x x
inflammation-induced
transcription factor
mTOR central hub of nutrient x
signaling/anabolic
metabolism
S6-P ribosomal protein x x x
controlled by mTOR
XBP1 ER stress transcription x
factor regulating diverse
immune cells
Functional Output
CyclinB1 proliferation/clonal x
expansion
Ki-67 proliferation/clonal x x
expansion
BrU RNA biosynthesis x
IbU DNA biosynthesis x
Puromycin protein biosynthesis x
a
A core metabolic cytometry panel of 10 targets is proposed, along with additional targets (x) for further insight

on earlier bulk metabolic measurements and previous literature,
and as such, the obtained mass cytometry data nicely recapitu-
lated established concepts in the field. Indeed, effector T cells
showed an expected increase in glycolytic activity (Michalek
et al., 2011) as evident by GLUT1 and GAPDH induction in Cy-
TOF analysis and as validated by an increased ECAR by Sea-
horse analysis. Conversely, memory T cells showed induction
of CPT1a, a key player in FAO that was previously associated
with CD8 memory T cells (van der Windt et al., 2012). This meta-
bolic rewiring occurred in a stepwise manner that could be fol-
lowed over time at single-cell resolution. This metabolic change
included a uniform glycolytic switch early after activation, fol-
lowed by heterogeneous GAPDH expression after 2 days of acti-
vation. These two GAPDH populations showed functional differ-
ences, which would not have been revealed in bulk approaches.
Furthermore, this approach revealed a transitional pool of
effector T cells that could not be detected when solely using
common antibodies against lineage markers and T cell subsets.
This newfound subset appeared 4 days after infection, was high-
ly proliferative, showed strongly induced expression of both
glycolysis and mitochondrial oxidation markers, and appeared
to be fueled by both fatty acids and amino acids based on the
expression of CTP1a and CD98, respectively. Having detected
this unknown subset of transitional cells by CyTOF, the authors
next sorted these cells as CD62Llo CD44hi CD25hi to confirm their
increased glycolytic and oxidative metabolism through extracel-
lular flux analysis. This is an excellent example of how single-cell
technologies provide new insights and how established bulk
analysis can be used to validate such new discoveries.
Met-Flow: Single-Cell Metabolic Profiling by Flow
Cytometry
By using fluorescently labeled antibodies instead of heavy-
metal-conjugated ones, single-cell metabolic profiling can also
be performed by ‘‘regular’’ flow cytometry (with some advan-
tages and disadvantages; see later). Ahl et al. configured a

Cell Metabolism 32, November 3, 2020 9

 ll
flow cytometry-based method called Met-Flow to interrogate the
metabolic state of immune cells using 10 fluorochrome-conju-
gated antibodies targeting rate-limiting enzymes and key pro-
teins in distinct metabolic pathways, combined with phenotypic
markers to generate a 27-color flow cytometry panel (Ahl et al.,
2020). To acquire this extensive panel, the authors used an X-
30 FACSymphony cytometer from BD. Using the expression pro-
files of 10 metabolic markers allowed Ahl et al. to cluster and
separate blood leukocyte subsets into distinct metabolic islands.
Akin to the CyTOF analysis by Hartmann et al., CD4+ and CD8+
T cell subsets were not separated on metabolic proteins alone,
but most other immune cell subsets were quite efficiently defined
solely based on their metabolic profiles. This indicates that it is
probably more beneficial to extend our conventional cytometry
panels with antibodies against key metabolic enzymes that are
coupled to specific immune effector functions, instead of stain-
ing additional immune cell surface markers for which the actual
function is often unknown. Even when using relatively small cy-
tometry panels, including a few metabolic antibodies, it will prob-
ably be highly informative to further dissect the phenotype, func-
tion, and metabolism of immune cells using equipment that is
present in virtually every institute. This also supports the idea
of Poznanski and Ashkar, who previously addressed the ques-
tion, ‘‘If an NK Cell Cannot be Defined by How It Looks, Could
It be Defined by How It Is Fueled?’’ in an elegant review on which
factors regulate NK cell function (Poznanski and Ashkar, 2019).
Of note, Met-Flow with only 10 metabolic proteins identified
blood leukocyte subsets with a resolution comparable to that of
+500 genes by scRNA-seq analysis, whereas solely analyzing
the expression of the same 10 genes did not resolve the immune
populations. Additionally, Met-Flow demonstrated the well-docu-
mented upregulation of glycolysis, OXPHOS, and FAS upon T cell
activation, and this metabolic rewiring was confirmed by bulk
extracellular flux analysis. Follow-up single-cell Met-Flow analysis
indicated that most cells relied on glucose for their metabolic
switch, but also revealed a subset of CD4 central memory cells
with an alternative metabolic fuel. As such, single-cell, but not
bulk, metabolic profiling is able to identify specific metabolic char-
acteristics and requirements of individual T cell subsets.
PROS AND CONS OF DISTINCT SINGLE-CELL
METABOLIC PROFILING TECHNIQUES AND
DEVELOPMENTS
Single-Cell Transcriptomics versus Cytometry
Single-cell transcriptomics and multicolor cytometry are com-
plementary approaches in terms of information acquired and
experimental set-up required for profiling. As such they can be
used simultaneously for maximal resolution, but at the same
time either approach can be more accessible or informative.
Here, we describe some basic considerations for selecting the
most optimal approach (Table 2).
scRNA-seq typically yields 3,000–5,000 genes detected in
each cell. This allows for unbiased clustering of the data and po-
tential identification of novel subpopulations that have not been
described before. From the metabolic perspective, it allows
simultaneous detection of multiple genes across many pathways
and therefore yields a global picture of metabolic changes. On
the other side, cytometry approaches are currently limited to
10 Cell Metabolism 32, November 3, 2020
Perspective
the detection of 40 proteins, including cell lineage markers,
as well as selected metabolic enzymes such as the ones in
Figure 1A and Table 1. Therefore, utilization of cytometry ap-
proaches requires reasonably specific hypotheses to design
the antibody panel.
At the same time, the typical number of cells that is profiled by
the two approaches is very different, as scRNA-seq is currently
limited to about 10,000 cells per sample, while cytometry can
routinely process millions of cells and can be readily scaled up
if needed. This distinction becomes critical when minor cellular
populations like innate lymphoid cells or DCs need to be consid-
ered. Furthermore, the cost of single-cell transcriptional profiling
is nearly an order of magnitude larger than that of a cytometry
run, and therefore it is not feasible to profile either large numbers
of replicates or to increase statistical power by profiling many
cells from the same samples. Additionally, scRNA-seq is more
sensitive to cellular morphology compared to cytometry ap-
proaches. A number of cell types, such as myocytes or neurons,
cannot be reliably profiled due to incompatibility of cellular
shapes and instrumental design. In other cases, such as neutro-
phils, transcriptional profiling is complicated by fragility of the
cells and overall lower RNA content of the cells, which introduces
significant biases into the data.
On the other side, one significant advantage of transcriptional
profiling is rooted in the utilization of single-nucleus RNA-seq. In
this approach, one can isolate individual nuclei and profile corre-
sponding transcripts (Mereu et al., 2020). This approach has
several important advantages, even though the number of de-
tected transcripts is typically much smaller due to the fact that
most RNA is localized in the cytosol and not in the nucleus
(Ding et al., 2020). First, this approach avoids morphological
complications and all cell types within the sample can be profiled
on the same footing. But most importantly, this approach allows
profiling of frozen samples, including ones that have been bio-
banked years before. This opens up the ability to study cohorts
of tissues from clinical samples that typically have to be
collected across extended periods of time.
In spite of these advantages, transcriptional data provide only
a ‘‘twice-removed’’ proxy to the metabolic measurements, and
in that respect, cytometry is a more direct approach that yields
protein-level measurements. Antibody-based approaches
directly measure levels of enzymes, including their post-transla-
tional modifications (and the modification of their key regulators).
An additional benefit of using cytometry-based approaches is
that it allows inclusion of non-protein-based markers of cellular
phenotypes. For instance, by including tagged IdU, BrU, and pu-
romycin substrates, metabolic profiles and cellular phenotypes
can be functionally linked to anabolic process like DNA, RNA,
and protein synthesis (Kimmey et al., 2019).
Altogether, in the setting of controlled experimental design we
would advocate for a combinatorial approach where scRNA-seq
is run on a limited scale to explore the broad landscape of the
changes, followed by large-scale cytometry profiling using a hy-
pothesis-driven antibody panel for validation at the protein level.
Mass versus Flow Cytometry for Single-Cell
Immunometabolism Research
Flow cytometry is still the most commonly used approach for im-
mune cell profiling, but the number of parameters that can be
 ll
Perspective

Table 2. Strengths and Weaknesses of the Techniques

Extracellular flux analysis
Bulk metabolomics and fluxomics
Single-cell transcriptomics
Flow cytometry
Mass cytometry (CyTOF)
MIBI-TOF
In situ activity assays
MS imaging/single-cell metabolomics
Strengths
(indirect) functional readout
availability of apparatus
affordable and fast
info on a broad range of pathways
most direct assessment of metabolism
applicable in vivo/ex vivo (advanced)
Unbiased
combined metabolic and phenotypic info
generation of new hypotheses
combination with metabolic dyes
machinery available in most institutes
affordable and fast
largest antibody panel size
barcoding of multiple samples
stability of signal
(indirect) spatial assessment of metabolism
large antibody panels
can be performed on biobanked material
activity measurements in situ
measures selected enzymes in key
pathways
direct in situ measurements of metabolism
broad range of metabolites/lipids measured
Weaknesses
high cell number and replicates needed
limited to glycolysis and mitochondrial
respiration
affected by cell isolation and sorting
instability of metabolome
high levels of input material needed
costly
limited to 10,000 cells/sample
limited sequencing depth/coverage
costly
targets need to be selected carefully
(biased)
fluorescent spill-over and background
indirect metabolic measurements
only available in high-end (core) facilities/
costly
lower sensitivity
indirect metabolic measurements
advanced high-end technique (availability/
cost/time)
indirect metabolic measurements
limited to dehydrogenases
measured at saturated substrate levels
specialized technique
advanced high-end technique (availability)
to be combined with phenotypic/metabolic
readouts

simultaneously tested is relatively limited due to fluorescence
spillover. This occurs when the fluorescence emission of one
fluorochrome is sensed by the detector of another fluorochrome.
Therefore, flow cytometry data need to be carefully compen-
sated, and this can be particularly challenging for large panels.
Nevertheless, the number of colors and antibodies used in flow
cytometry is growing fast and panels of 20 and more markers
are now common. So far, it does not reach the +40 parameters
that can be assessed in CyTOF, but at this point flow cytometry
is clearly catching up. Opposed to fluorescence detectors, Cy-
TOF uses a mass cytometer to detect the TOF of each metal
that is set by its mass (Spitzer and Nolan, 2016). As such, both
the detection overlap and background (autofluorescence in
flow cytometry) are very low in CyTOF analyses because heavy
metals do not naturally occur in cells. Designing ideal CyTOF
panels is still key to minimize ‘‘spillover’’ due to impurities of
metal tags and to take differences in signal intensity of distinct
metals into account. Also, no mass channel in CyTOF is as sen-
sitive as bright fluorochromes such as PE in flow cytometry.
While stained cells need to be acquired fast in flow cytometry
to prevent photobleaching, metal-tagged samples can be
(cryo)preserved for weeks and acquired simultaneously when
all samples are collected over time during a clinical trial. Alterna-
tively, cells can be isolated, fixed, and stored over time and
handled together once the clinical trial or experiment is finalized.
Indeed, unique barcoding of up to 20 samples allows one to
combine and subsequently stain, process, and acquire them
as one multiplexed sample. These advantages come with a
cost since CyTOF antibodies are more expensive than fluoro-
chrome-coupled ones, and the instrument used to acquire the
labeled cells is more costly and advanced in the case of mass cy-
tometry. While multi-color flow cytometers are the workhorses of
virtually every lab, mass cytometers are less common and
restricted to high-end core facilities.
Nowadays, spectral analyzers are pushing the boundaries of
flow cytometry. Employing 5 lasers to detect the full emission
spectra over 64 channels, instruments like the Cytek Aurora
are now able to detect 30+ colors in a sensitive, fast, and acces-
sible manner (Ferrer-Font et al., 2020). This is particularly valu-
able for (immuno)metabolism research since it can combine fluo-
rochrome-tagged antibodies against metabolic proteins and
immune lineage markers with fluorescent metabolic tools to pro-
vide an extra layer of metabolic information (Scharping et al.,
2016). For example, expression of the glucose transporter
GLUT1 can be directly related to the uptake of fluorescently
labeled 2NBDG. Similarly, increased expression of proteins
that suggest increased mitochondrial oxidation can be directly
linked to mitochondrial mass and membrane potential using

Cell Metabolism 32, November 3, 2020 11

 ll
fluorescent dyes such as MitoTracker and TMRE (or related
TMRM), respectively. As such, a clear advantage of fluores-
cence-based flow cytometry is the applicability of fluorescent
metabolic dyes, whereas the largest benefits of mass cytometry
are panel size, signal stability, and barcoding possibilities that
allow in-depth single-cell metabolic profiling of immune cells in
patient biopsies obtained during clinical trials.
SPATIALLY RESOLVED IMMUNOMETABOLISM
A major drawback of all approaches listed so far is that they
require cells in suspension and as such lack spatial information.
The spatial aspect is particularly important since specific micro-
environmental factors, including the complex mixture of stimuli,
availability of nutrients, and interactions with neighboring cells,
are key drivers of cellular metabolic profiles in tissues and criti-
cally shape immunity and disease progression. Another draw-
back of conventional single-cell profiling approaches is that the
specific protocols used to digest tissues into single cells induce
cell death in some, but not all, cells and might favor the isolation
of specific immune cell subsets with associated metabolic pro-
files. Together, these limitations underscore the need for the vali-
dation of key observations of immunometabolic features of cells
within their tissue microenvironment, taking into account their
spatial organization and avoiding problems associated with
cell sorting. New methods that instantaneously capture spatial
context and single-cell transcriptome profiles are now being
applied, but their use in immunometabolism research has yet
to be described. Nevertheless, recent publications highlight
how the metabolic configuration of single cells within their micro-
environment can be assessed at the protein level in a highly mul-
tiplexed manner.
MIBI-TOF as Protein-Based Spatial Single-Cell Immune/
Metabolic Analysis
Hartmann et al. took an important step toward spatial metabolic
profiling by transferring their established metabolic CyTOF panel
to the multiplexed ion beam imaging (MIBI-TOF) platform. This
recently developed approach couples single-cell metabolic pro-
files, immune cell phenotypes, and functional states to cell-cell
interactions and location within tissues (Hartmann and Bendall,
2020; Keren et al., 2019). To do so, tissue sections are stained
with heavy-metal-coupled antibodies (that are also used in Cy-
TOF), and then scanned with an ion beam to release the heavy
metals from antibodies recognizing specific metabolic and im-
mune targets. For each acquired pixel, the released ions are
quantified by TOF-MS as in CyTOF, with the important advan-
tage that the obtained high-dimensional data obtained by
MIBI-TOF also contain spatial information.
Staining colorectal carcinoma and control tissue sections with
lineage markers in combination with a broad range of metabolic
antibodies, followed by segmentation of individual cells in the ac-
quired images and subsequent clustering of cells into cell line-
ages, allowed researchers to couple immune cell phenotypes
to metabolic characteristics that are driven by their specific tis-
sue microenvironment (Hartmann et al., 2020). Metabolic profiles
were spatially organized in environmental niches with similar
metabolic characteristics irrespective of cell type. Cells express-
ing high levels of metabolic factors associated with glycolysis,
12 Cell Metabolism 32, November 3, 2020
Perspective
mitochondrial respiration, or amino acid metabolism were often
surrounded by neighboring cells expressing the same metabolic
target. This microenvironment-driven metabolic polarization was
especially apparent when comparing immunometabolic profiles
of cells near the tumor border with the ones positioned farther
away from the boundary. Typically, metabolically suppressed
cells were farther away from the tumor border, while metaboli-
cally active cells resided at the tumor-immune edge. Since the
MIBI-TOF approach can be performed on existing FFPE material
obtained from biobanks, it offers the opportunity to link immuno-
metabolic profiles and locations to clinical outcome and thera-
peutic success. As such, single-cell metabolic profiling is an
important next step to translate immunometabolism research to-
ward clinical diagnostics and therapy. Yet it should be noted that
the presence of a particular enzyme does not always correlate
with its activity, and this aspect should be taken into account
when drawing conclusions from these antibody-based ap-
proaches.
Mapping Metabolism within Microenvironments In Situ
Using Dehydrogenase Activity Assays
Miller et al. previously configured an alternate method to assess
the metabolic configuration of cells in their tissue microenviron-
ment (Miller et al., 2017). Their approach relies on enzyme activ-
ity measurements at saturating substrate and co-factor availabil-
ity in combination with staining of multiple immune cell markers
on consecutive tissue sections. Dehydrogenase activities are
measured for five enzymes catalyzing key steps in major meta-
bolic pathways: G6PD in the PPP, GAPDH in glycolysis, LDH in
lactate fermentation, and IDH and SDH in the TCA cycle. When
these dehydrogenases are active, nitroblue tetrazolium chloride
(NBT) is reduced by NAD(P)H into a strongly colored state that
can be quantified and related to the acquired immune cell marker
expression on consecutive sections. The authors employed this
technique to interrogate the metabolic signatures of macro-
phage and T cell subsets in human cancerous and control colon.
Macrophages producing IL-6 and TNF within the tumor showed
increased glycolytic GADPH activity in comparison with non-in-
flammatory macrophages within the TME, but overall tumor-
associated macrophages appeared metabolically repressed in
comparison to macrophages in healthy tissue. It is therefore
tempting to speculate that an impaired metabolic fitness of mac-
rophages in tumors might prevent their antitumor activities.
Moreover, Tregs showed cancer-specific metabolic features
with suppressed glycolytic GAPDH activity and increased mito-
chondrial SDH compared to Tregs in healthy tissues. Identifying
such tumor-specific metabolic properties of immune cell sub-
sets could aid the development of new therapeutic approaches.
Of note, the activities of these distinct dehydrogenases are
measured at saturated substrate concentrations and are thus
estimates of the optimal in vivo scenario rather than a precise
reflection of their activity in a given context such as a hypoxia
TME in which cells compete for nutrients.
CONCLUSIONS AND FUTURE DIRECTIONS
It is clear that a technological transition toward understanding
immunometabolism at the single-cell level is inevitable, although
not yet complete. Technologies developed for transcriptomic

Perspective
Figure 3. Hypothetical Technology Development Perspective in the Context of Single-Cell Immunometabolism Applications
and proteomic profiling at single-cell resolution demonstrate a
sufficient level of maturity to establish the current frontier in the
field of immunometabolism—initial assessment and focused
probing of the metabolic landscape within clinical samples and
in vivo models of inflammation. As discussed above, specific
metabolic features can be informed by scRNA-seq profiling or
even bulk observations from mixed cell cultures/ex vivo samples
and then dissected through dedicated protein-targeting panels.
For instance, based on the overlap between the recently pub-
lished panels and previous literature, one can compile a core
metabolic panel consisting of 10 metabolic antibodies that is
also applicable in multicolor flow cytometry and a more exten-
sive set panel including an additional 10 antibodies to gain
more detailed and functional insight through CyTOF analysis (Ta-
ble 1).
While methods such as CyTOF and scRNA-seq can be insight-
ful with respect to metabolic remodeling within the diverse cell
subpopulations, the primary data obtained with these methods
remain either indirect or limited in scope. Therefore, the techno-
logical advances that can address these challenges will build up
a foundation for the next generation of immunometabolism
research (Figure 3). Two key directions would be (1) improving
depth and quality of data for the single-cell level profiling of
ex vivo cell suspensions, and (2) improving our ability to resolve
the spatial distribution of metabolic features (e.g., within tissue
slices).
The next generation approaches would include direct single-
cell metabolite profiling (Duncan et al., 2019), which should be
accompanied by the development of robust methods to isolate
individual cells without significant metabolic perturbations (Fig-
ures 3). Such an approach would not only allow direct evaluation
of metabolite levels, but also provide fluxomics at a single-cell
level, when cell mixtures, or even animals and people, are sup-
plemented with labeled substrates. Given the unstable nature
of the metabolome and the fact that isolation of cells affects
the readout, the future of single-cell metabolomics is probably
not in measuring metabolites in isolated cells in suspension,
ll
but rather in MS-based imaging approaches to spatially resolve
the metabolic configuration of single cells within their tissue
microenvironment. For detailed information on emerging tech-
niques such as MALDI-MS for single-cell and even subcellular
analysis, we refer the reader to excellent recent reviews (Gilmore

c
et al., 2019; S
upáková et al., 2020).
Resolving immunometabolism at single-cell resolution is
certainly not the end stage as cells are not homogeneous bags
but rather highly compartmentalized structures with a heteroge-
neous distribution of metabolites with different functions at
different subcellular locations. For example, acetyl-CoA feeds
the TCA cycle within the mitochondria, while fueling fatty acid
and cholesterol biosynthesis in the cytosol and histone acetyla-
tion within the nucleus. As such, improving the subcellular reso-
lution of metabolic readouts will provide an additional layer of
insight into the immunometabolism field and to science in gen-
eral (Alexandrov, 2020; Pareek et al., 2020).
In the context of complementary ‘omics profiling, fusion of
protein and transcriptomic level measurements along the lines
implemented in CITE-seq approaches bears significant promise
for the immunometabolism field (Stoeckius et al., 2017). Specif-
ically, extension of the CITE-seq-based approach to robust
measurement of intracellular proteins would allow for measuring
significantly larger numbers of enzyme and their post-transla-
tional modifications since antibodies are conjugated to nucleic
acid barcodes and therefore 100–200 proteins can feasibly be
measured simultaneously (Hwang et al., 2020; Stoeckius et al.,
2017). This will significantly improve the resolution compared
to the current 40 antibodies possible with CyTOF, especially
considering the fact that almost half of the panel is typically
devoted to cell-type-specific markers.
Furthermore, adopting new data generation techniques will
demand development of a portfolio of novel computational ap-
proaches that could integrate the network-based nature of the
data with corresponding technological outputs in terms of spatial
location or potential temporal aspects of immune responses.
The use of computational algorithms such as SCORPIUS
Cell Metabolism 32, November 3, 2020 13
 ll
(Cannoodt et al., 2016) is promising to deduce a pseudotime
ordering to investigate the presumed temporal sequence of
metabolic rewiring. Understanding the kinetics of phenotypic,
functional, and metabolic changes can reveal which metabolic
events are causal and what are mere bystanders of a particular
immune state. Understanding the spatiotemporal aspect of im-
munometabolism can help to rationally design new therapeutic
approaches as one would have a better understanding of the
drivers of disease versus bystander effects. In particular, target-
ing the metabolic changes that kinetically precede, and thus
likely also regulate, the phenotypic alterations in immune cells
could improve immune cell function and disease outcome.
Moreover, physiological immune response occurs in multicel-
lular environments that are highly organized spatially. Given
that our current understanding of the immunometabolic regula-
tion is obtained using highly controlled in vitro settings (typically
with purified cell types), the design of metabolic interventions for
in vivo phenotypes relies as much on chance as it does on knowl-
edge-based considerations. We envision that understanding the
fundamental principles of multicellular immunometabolic re-
modeling will pave the way to deductively transition from pertur-
bation of key intracellular pathways toward the successful ther-
apies on the whole-body scale. Yet it is clear that additional
analytical tools are required to fully grasp the spatiotemporal
aspect of immunometabolism and for tasks such as metabolic
clustering on the single-cell level.
Altogether, immunometabolism is quickly becoming a sys-
tems-level science where multidisciplinary expertise is required
for making the fullest use of the novel technologies. Therefore,
computational advances would be important not only in terms
of data analysis itself but also in terms of their usability and inter-
pretability by experimental immunologists in order to success-
fully transit to the single-cell era of immunometabolism.
ACKNOWLEDGMENTS
We thank Sjoerd Schetters, Menno de Winther, and Felix Hartmann for insight-
ful feedback; Sanne Verberk, Kyra de Goede, and Luc Magré for proofreading;
and Amanda Swain and Karl Harber for edits and comments on the manu-
script. M.N.A. is supported by R01-AI125618 from National Institute of Allergy
and Infectious Diseases (NIAID). J.V.d.B. received a VENI grant from ZonMW
(91615052), a Netherlands Heart Foundation Junior Postdoctoral grant
(2013T003) and Senior Fellowship (2017T048), an NWO ENW-KLEIN-1 grant
(268), and a CCA PhD grant from Cancer Center Amsterdam.
REFERENCES
Ahl, P.J., Hopkins, R.A., Xiang, W.W., Au, B., Kaliaperumal, N., Fairhurst, A.M.,
and Connolly, J.E. (2020). Met-Flow, a strategy for single-cell metabolic anal-
ysis highlights dynamic changes in immune subpopulations. Commun Biol
3, 305.
Alexandrov, T. (2020). Probing metabolism in time and space. Science 368,
241–242.
Alves, T.C., Pongratz, R.L., Zhao, X., Yarborough, O., Sereda, S., Shirihai, O.,
Cline, G.W., Mason, G., and Kibbey, R.G. (2015). Integrated, step-wise, mass-
isotopomeric flux analysis of the TCA cycle. Cell Metab. 22, 936–947.
Ariss, M.M., Islam, A.B.M.M.K., Critcher, M., Zappia, M.P., and Frolov, M.V.
(2018). Single cell RNA-sequencing identifies a metabolic aspect of apoptosis
in Rbf mutant. Nat. Commun. 9, 5024.
Ayres, J.S. (2020). Immunometabolism of infections. Nat. Rev. Immunol.
20, 79–80.
14 Cell Metabolism 32, November 3, 2020
Perspective
Baardman, J., Verberk, S.G.S., Prange, K.H.M., van Weeghel, M., van der Vel-
den, S., Ryan, D.G., Wu €st, R.C.I., Neele, A.E., Speijer, D., Denis, S.W., et al.
(2018). A defective pentose phosphate pathway reduces inflammatory macro-
phage responses during hypercholesterolemia. Cell Rep. 25, 2044–2052.e5.
Bordbar, A., Feist, A.M., Usaite-Black, R., Woodcock, J., Palsson, B.O., and
Famili, I. (2011). A multi-tissue type genome-scale metabolic network for anal-
ysis of whole-body systems physiology. BMC Syst. Biol. 5, 180.
Bordbar, A., Mo, M.L., Nakayasu, E.S., Schrimpe-Rutledge, A.C., Kim, Y.M.,
Metz, T.O., Jones, M.B., Frank, B.C., Smith, R.D., Peterson, S.N., et al.
(2012). Model-driven multi-omic data analysis elucidates metabolic immuno-
modulators of macrophage activation. Mol. Syst. Biol. 8, 558.
Buck, M.D., Sowell, R.T., Kaech, S.M., and Pearce, E.L. (2017). Metabolic in-
struction of immunity. Cell 169, 570–586.
Cannoodt, R., Saelens, W., Sichien, D., Tavernier, S., Janssens, S., Guilliams,
M., Lambrecht, B., Preter, K.D., and Saeys, Y. (2016). SCORPIUS improves
trajectory inference and identifies novel modules in dendritic cell development.
bioRxiv. https://doi.org/10.1101/079509.
Caputa, G., Castoldi, A., and Pearce, E.J. (2019). Metabolic adaptations of tis-
sue-resident immune cells. Nat. Immunol. 20, 793–801.
Caro-Maldonado, A., Wang, R., Nichols, A.G., Kuraoka, M., Milasta, S., Sun,
L.D., Gavin, A.L., Abel, E.D., Kelsoe, G., Green, D.R., and Rathmell, J.C.
(2014). Metabolic reprogramming is required for antibody production that is
suppressed in anergic but exaggerated in chronically BAFF-exposed B cells.
J. Immunol. 192, 3626–3636.
Chang, C.H., Curtis, J.D., Maggi, L.B., Jr., Faubert, B., Villarino, A.V., O’Sulli-
van, D., Huang, S.C., van der Windt, G.J., Blagih, J., Qiu, J., et al. (2013). Post-
transcriptional control of T cell effector function by aerobic glycolysis. Cell 153,
1239–1251.
De Rosa, V., Galgani, M., Porcellini, A., Colamatteo, A., Santopaolo, M., Zu-
chegna, C., Romano, A., De Simone, S., Procaccini, C., La Rocca, C., et al.
(2015). Glycolysis controls the induction of human regulatory T cells by modu-
lating the expression of FOXP3 exon 2 splicing variants. Nat. Immunol. 16,
1174–1184.
Dennis, E.A., Deems, R.A., Harkewicz, R., Quehenberger, O., Brown, H.A.,
Milne, S.B., Myers, D.S., Glass, C.K., Hardiman, G., Reichart, D., et al.
(2010). A mouse macrophage lipidome. J. Biol. Chem. 285, 39976–39985.
Ding, J., Adiconis, X., Simmons, S.K., Kowalczyk, M.S., Hession, C.C., Marja-
novic, N.D., Hughes, T.K., Wadsworth, M.H., Burks, T., Nguyen, L.T., et al.
(2020). Systematic comparison of single-cell and single-nucleus RNA-
sequencing methods. Nat. Biotechnol. 38, 737–746.
Divakaruni, A.S., Hsieh, W.Y., Minarrieta, L., Duong, T.N., Kim, K.K.O., Des-
ousa, B.R., Andreyev, A.Y., Bowman, C.E., Caradonna, K., Dranka, B.P.,
et al. (2018). Etomoxir inhibits macrophage polarization by disrupting CoA ho-
meostasis. Cell Metab. 28, 490–503.e7.
Duncan, K.D., Fyrestam, J., and Lanekoff, I. (2019). Advances in mass spec-
trometry based single-cell metabolomics. Analyst (Lond.) 144, 782–793.
Everts, B., Amiel, E., Huang, S.C., Smith, A.M., Chang, C.H., Lam, W.Y., Red-
mann, V., Freitas, T.C., Blagih, J., van der Windt, G.J., et al. (2014). TLR-driven
early glycolytic reprogramming via the kinases TBK1-IKKe supports the
anabolic demands of dendritic cell activation. Nat. Immunol. 15, 323–332.
Ferrer-Font, L., Pellefigues, C., Mayer, J.U., Small, S.J., Jaimes, M.C., and
Price, K.M. (2020). Panel design and optimization for high-dimensional immu-
nophenotyping assays using spectral flow cytometry. Curr. Protoc. Cytom.
92, e70.
Galván-Peña, S., Carroll, R.G., Newman, C., Hinchy, E.C., Palsson-McDer-
mott, E., Robinson, E.K., Covarrubias, S., Nadin, A., James, A.M., Haneklaus,
M., et al. (2019). Malonylation of GAPDH is an inflammatory signal in macro-
phages. Nat. Commun. 10, 338.
Geiger, R., Rieckmann, J.C., Wolf, T., Basso, C., Feng, Y., Fuhrer, T., Koga-
deeva, M., Picotti, P., Meissner, F., Mann, M., et al. (2016). L-arginine modu-
lates T cell metabolism and enhances survival and anti-tumor activity. Cell
167, 829–842.e13.
Gilmore, I.S., Heiles, S., and Pieterse, C.L. (2019). Metabolic imaging at the sin-
gle-cell scale: recent advances in mass spectrometry imaging. Annu. Rev.
Anal. Chem. (Palo Alto, Calif.) 12, 201–224.

Perspective
Gubin, M.M., Esaulova, E., Ward, J.P., Malkova, O.N., Runci, D., Wong, P., No-
guchi, T., Arthur, C.D., Meng, W., Alspach, E., et al. (2018). High-dimensional
analysis delineates myeloid and lymphoid compartment remodeling during
successful immune-checkpoint cancer therapy. Cell 175, 1014–1030.e19.
Hartmann, F.J., and Bendall, S.C. (2020). Immune monitoring using mass cy-
tometry and related high-dimensional imaging approaches. Nat. Rev. Rheu-
matol. 16, 87–99.
Hartmann, F.J., Mrdjen, D., McCaffrey, E., Glass, D.R., Greenwald, N.F., Bhar-
adwaj, A., Khair, Z., Verberk, S.G.S., Baranski, A., Baskar, R., et al. (2020). Sin-
gle-cell metabolic profiling of human cytotoxic T cells. Nat. Biotechnol. Pub-
lished online August 31, 2020. https://doi.org/10.1038/s41587-020-0651-8.
Haschemi, A., Kosma, P., Gille, L., Evans, C.R., Burant, C.F., Starkl, P., Knapp,
B., Haas, R., Schmid, J.A., Jandl, C., et al. (2012). The sedoheptulose kinase
CARKL directs macrophage polarization through control of glucose meta-
bolism. Cell Metab. 15, 813–826.
Hayashi, K., Jutabha, P., Endou, H., Sagara, H., and Anzai, N. (2013). LAT1 is a
critical transporter of essential amino acids for immune reactions in activated
human T cells. J. Immunol. 191, 4080–4085.
Hsieh, W.Y., Zhou, Q.D., York, A.G., Williams, K.J., Scumpia, P.O., Kronen-
berger, E.B., Hoi, X.P., Su, B., Chi, X., Bui, V.L., et al. (2020). Toll-like receptors
induce signal-specific reprogramming of the macrophage lipidome. Cell
Metab. 32, 128–143.e5.
Huang, S.C., Everts, B., Ivanova, Y., O’Sullivan, D., Nascimento, M., Smith,
A.M., Beatty, W., Love-Gregory, L., Lam, W.Y., O’Neill, C.M., et al. (2014).
Cell-intrinsic lysosomal lipolysis is essential for alternative activation of macro-
phages. Nat. Immunol. 15, 846–855.
Huang, S.C., Smith, A.M., Everts, B., Colonna, M., Pearce, E.L., Schilling, J.D.,
and Pearce, E.J. (2016). Metabolic reprogramming mediated by the mTORC2-
IRF4 signaling axis is essential for macrophage alternative activation. Immunity
45, 817–830.
Hwang, B., Lee, D.S., Tamaki, W., Sun, Y., Ogorodnikov, A., Hartoularos, G.,
Winters, A., Song, Y.S., Chow, E.D., Spitzer, M.H., et al. (2020). SCITO-seq:
single-cell combinatorial indexed cytometry sequencing. bioRxiv. https://doi.
org/10.1101/2020.03.27.012633.
Jha, A.K., Huang, S.C., Sergushichev, A., Lampropoulou, V., Ivanova, Y., Log-
inicheva, E., Chmielewski, K., Stewart, K.M., Ashall, J., Everts, B., et al. (2015).
Network integration of parallel metabolic and transcriptional data reveals
metabolic modules that regulate macrophage polarization. Immunity 42,
419–430.
Johnson, M.O., Wolf, M.M., Madden, M.Z., Andrejeva, G., Sugiura, A., Contre-
ras, D.C., Maseda, D., Liberti, M.V., Paz, K., Kishton, R.J., et al. (2018). Distinct
regulation of Th17 and Th1 cell differentiation by glutaminase-dependent
metabolism. Cell 175, 1780–1795.e19.
Kanehisa, M., and Goto, S. (2000). KEGG: Kyoto encyclopedia of genes and
genomes. Nucleic Acids Res. 28, 27–30.
Kelly, B., and Pearce, E.L. (2020). Amino assets: how amino acids support im-
munity. Cell Metab. 32, 154–175.
Keren, L., Bosse, M., Thompson, S., Risom, T., Vijayaragavan, K., McCaffrey,
E., Marquez, D., Angoshtari, R., Greenwald, N.F., Fienberg, H., et al. (2019).
MIBI-TOF: a multiplexed imaging platform relates cellular phenotypes and tis-
sue structure. Sci. Adv. 5, x5851.
Ketelhuth, D.F.J., Lutgens, E., Ba €ck, M., Binder, C.J., Van den Bossche, J.,
Daniel, C., Dumitriu, I.E., Hoefer, I., Libby, P., O’Neill, L., et al. (2019). Immuno-
metabolism and atherosclerosis: perspectives and clinical significance: a po-
sition paper from the Working Group on Atherosclerosis and Vascular Biology
of the European Society of Cardiology. Cardiovasc. Res. 115, 1385–1392.
Kimmey, S.C., Borges, L., Baskar, R., and Bendall, S.C. (2019). Parallel anal-
ysis of tri-molecular biosynthesis with cell identity and function in single cells.
Nat. Commun. 10, 1185.
Koelwyn, G.J., Corr, E.M., Erbay, E., and Moore, K.J. (2018). Regulation of
macrophage immunometabolism in atherosclerosis. Nat. Immunol. 19,
526–537.
Krawczyk, C.M., Holowka, T., Sun, J., Blagih, J., Amiel, E., DeBerardinis, R.J.,
Cross, J.R., Jung, E., Thompson, C.B., Jones, R.G., and Pearce, E.J. (2010).
Toll-like receptor-induced changes in glycolytic metabolism regulate dendritic
cell activation. Blood 115, 4742–4749.
ll
Lam, W.Y., Becker, A.M., Kennerly, K.M., Wong, R., Curtis, J.D., Llufrio, E.M.,
McCommis, K.S., Fahrmann, J., Pizzato, H.A., Nunley, R.M., et al. (2016). Mito-
chondrial pyruvate import promotes long-term survival of antibody-secreting
plasma cells. Immunity 45, 60–73.
Lampropoulou, V., Sergushichev, A., Bambouskova, M., Nair, S., Vincent,
E.E., Loginicheva, E., Cervantes-Barragan, L., Ma, X., Huang, S.C., Griss, T.,
et al. (2016). Itaconate links inhibition of succinate dehydrogenase with macro-
phage metabolic remodeling and regulation of inflammation. Cell Metab. 24,
158–166.
Levine, L.S., Hiam, K.J., Marquez, D.M., Tenvooren, I., Contreras, D.C., Rath-
mell, J.C., and Spitzer, M.H. (2020). Single-cell metabolic analysis by mass cy-
tometry reveals distinct transitional states of CD8 T cell differentiation. bioRxiv.
https://doi.org/10.1101/2020.01.21.911545.
Lim, A.R., Rathmell, W.K., and Rathmell, J.C. (2020). The tumor microenviron-
ment as a metabolic barrier to effector T cells and immunotherapy. eLife 9,
e55185.
Llufrio, E.M., Wang, L., Naser, F.J., and Patti, G.J. (2018). Sorting cells alters
their redox state and cellular metabolome. Redox Biol. 16, 381–387.
Ma, E.H., Bantug, G., Griss, T., Condotta, S., Johnson, R.M., Samborska, B.,
Mainolfi, N., Suri, V., Guak, H., Balmer, M.L., et al. (2017). Serine is an essential
metabolite for effector T cell expansion. Cell Metab. 25, 345–357.
Ma, E.H., Verway, M.J., Johnson, R.M., Roy, D.G., Steadman, M., Hayes, S.,
Williams, K.S., Sheldon, R.D., Samborska, B., Kosinski, P.A., et al. (2019).
Metabolic profiling using stable isotope tracing reveals distinct patterns of
glucose utilization by physiologically activated CD8+ T cells. Immunity 51,
856–870.e5.
Makowski, L., Chaib, M., and Rathmell, J.C. (2020). Immunometabolism: from
basic mechanisms to translation. Immunol. Rev. 295, 5–14.
Mazumdar, C., Driggers, E.M., and Turka, L.A. (2020). The untapped opportu-
nity and challenge of immunometabolism: a new paradigm for drug discovery.
Cell Metab. 31, 26–34.
Mereu, E., Lafzi, A., Moutinho, C., Ziegenhain, C., McCarthy, D.J., Álvarez-Var-
€n, D., Lau, J.K., et al. (2020). Benchmarking single-
ela, A., Batlle, E., Sagar, Gru
cell RNA-sequencing protocols for cell atlas projects. Nat. Biotechnol. 38,
747–755.
Michalek, R.D., Gerriets, V.A., Jacobs, S.R., Macintyre, A.N., MacIver, N.J.,
Mason, E.F., Sullivan, S.A., Nichols, A.G., and Rathmell, J.C. (2011). Cutting
edge: distinct glycolytic and lipid oxidative metabolic programs are essential
for effector and regulatory CD4+ T cell subsets. J. Immunol. 186, 3299–3303.
Miller, A., Nagy, C., Knapp, B., Laengle, J., Ponweiser, E., Groeger, M., Starkl,
P., Bergmann, M., Wagner, O., and Haschemi, A. (2017). Exploring metabolic
configurations of single cells within complex tissue microenvironments. Cell
Metab. 26, 788–800.e6.
Miragaia, R.J., Gomes, T., Chomka, A., Jardine, L., Riedel, A., Hegazy, A.N.,
Whibley, N., Tucci, A., Chen, X., Lindeman, I., et al. (2019). Single-cell tran-
scriptomics of regulatory T cells reveals trajectories of tissue adaptation. Im-
munity 50, 493–504.e7.
Moon, J.S., Hisata, S., Park, M.A., DeNicola, G.M., Ryter, S.W., Nakahira, K.,
and Choi, A.M.K. (2015). mTORC1-induced HK1-dependent glycolysis regu-
lates NLRP3 inflammasome activation. Cell Rep. 12, 102–115.
Munder, M., Eichmann, K., and Modolell, M. (1998). Alternative metabolic
states in murine macrophages reflected by the nitric oxide synthase/arginase
balance: competitive regulation by CD4+ T cells correlates with Th1/Th2
phenotype. J. Immunol. 160, 5347–5354.
Nakaya, M., Xiao, Y., Zhou, X., Chang, J.H., Chang, M., Cheng, X., Blonska,
M., Lin, X., and Sun, S.C. (2014). Inflammatory T cell responses rely on amino
acid transporter ASCT2 facilitation of glutamine uptake and mTORC1 kinase
activation. Immunity 40, 692–705.
Nomura, M., Liu, J., Rovira, I.I., Gonzalez-Hurtado, E., Lee, J., Wolfgang, M.J.,
and Finkel, T. (2016). Fatty acid oxidation in macrophage polarization. Nat. Im-
munol. 17, 216–217.
O’Neill, L.A.J., and Artyomov, M.N. (2019). Itaconate: the poster child of meta-
bolic reprogramming in macrophage function. Nat. Rev. Immunol. 19,
273–281.
Cell Metabolism 32, November 3, 2020 15
 ll
O’Neill, L.A., Kishton, R.J., and Rathmell, J. (2016). A guide to immunometab-
olism for immunologists. Nat. Rev. Immunol. 16, 553–565.
Orth, J.D., Thiele, I., and Palsson, B.O. (2010). What is flux balance analysis?
Nat. Biotechnol. 28, 245–248.
Palsson-McDermott, E.M., Curtis, A.M., Goel, G., Lauterbach, M.A., Sheedy,
F.J., Gleeson, L.E., van den Bosch, M.W., Quinn, S.R., Domingo-Fernandez,
R., Johnston, D.G., et al. (2015). Pyruvate kinase M2 regulates Hif-1a activity
and IL-1b induction and is a critical determinant of the warburg effect in
LPS-activated macrophages. Cell Metab. 21, 65–80.
Pareek, V., Tian, H., Winograd, N., and Benkovic, S.J. (2020). Metabolomics
and mass spectrometry imaging reveal channeled de novo purine synthesis
in cells. Science 368, 283–290.
Pelgrom, L.R., van der Ham, A.J., and Everts, B. (2016). Analysis of TLR-
induced metabolic changes in dendritic cells using the Seahorse XF(e)96
extracellular flux analyzer. Methods Mol. Biol. 1390, 273–285.
Poznanski, S.M., and Ashkar, A.A. (2019). What defines NK cell functional fate:
phenotype or metabolism? Front. Immunol. 10, 1414.
Raud, B., Roy, D.G., Divakaruni, A.S., Tarasenko, T.N., Franke, R., Ma, E.H.,
Samborska, B., Hsieh, W.Y., Wong, A.H., Stu €ve, P., et al. (2018). Etomoxir ac-
tions on regulatory and memory T cells are independent of Cpt1a-mediated
fatty acid oxidation. Cell Metab. 28, 504–515.e7.
Rodriguez, A.E., Ducker, G.S., Billingham, L.K., Martinez, C.A., Mainolfi, N.,
Suri, V., Friedman, A., Manfredi, M.G., Weinberg, S.E., Rabinowitz, J.D., and
Chandel, N.S. (2019). Serine metabolism supports macrophage IL-1b produc-
tion. Cell Metab. 29, 1003–1011.e4.
Scharping, N.E., Menk, A.V., Moreci, R.S., Whetstone, R.D., Dadey, R.E., Wat-
kins, S.C., Ferris, R.L., and Delgoffe, G.M. (2016). The tumor microenvironment
represses T cell mitochondrial biogenesis to drive intratumoral T cell metabolic
insufficiency and dysfunction. Immunity 45, 374–388.

c
S
upáková, K., Balluff, B., Tressler, C., Adelaja, T., Heeren, R.M.A., Glunde, K.,
and Ertaylan, G. (2020). Cellular resolution in clinical MALDI mass spectrom-
etry imaging: the latest advancements and current challenges. Clin. Chem.
Lab. Med. 58, 914–929.
Slavov, N. (2020). Unpicking the proteome in single cells. Science 367,
512–513.
Spitzer, M.H., and Nolan, G.P. (2016). Mass cytometry: single cells, many fea-
tures. Cell 165, 780–791.
Stoeckius, M., Hafemeister, C., Stephenson, W., Houck-Loomis, B., Chatto-
padhyay, P.K., Swerdlow, H., Satija, R., and Smibert, P. (2017). Simultaneous
epitope and transcriptome measurement in single cells. Nat. Methods 14,
865–868.
Swain, A., Bambouskova, M., Kim, H., Andhey, P.S., Duncan, D., Auclair, K.,
Chubukov, V., Simons, D.M., Roddy, T.P., Stewart, K.M., and Artyomov,
M.N. (2020). Comparative evaluation of itaconate and its derivatives reveals
divergent inflammasome and type I interferon regulation in macrophages.
Nat Metab 2, 594–602.
Tabas, I., and Bornfeldt, K.E. (2020). Intracellular and intercellular aspects of
macrophage immunometabolism in atherosclerosis. Circ. Res. 126,
1209–1227.
Taguchi, K., Motohashi, H., and Yamamoto, M. (2011). Molecular mechanisms
of the Keap1–Nrf2 pathway in stress response and cancer evolution. Genes
Cells 16, 123–140.
16 Cell Metabolism 32, November 3, 2020
Perspective
Turbitt, W.J., Buchta Rosean, C., Weber, K.S., and Norian, L.A. (2020). Obesity
and CD8 T cell metabolism: implications for anti-tumor immunity and cancer
immunotherapy outcomes. Immunol. Rev. 295, 203–219.
Van den Bossche, J., and Saraber, D.L. (2018). Metabolic regulation of macro-
phages in tissues. Cell. Immunol. 330, 54–59.
Van den Bossche, J., and van der Windt, G.J.W. (2018). Fatty acid oxidation in
macrophages and T cells: time for reassessment? Cell Metab. 28, 538–540.
Van den Bossche, J., Lamers, W.H., Koehler, E.S., Geuns, J.M., Alhonen, L.,
Uimari, A., Pirnes-Karhu, S., Van Overmeire, E., Morias, Y., Brys, L., et al.
(2012). Pivotal Advance: Arginase-1-independent polyamine production stim-
ulates the expression of IL-4-induced alternatively activated macrophage
markers while inhibiting LPS-induced expression of inflammatory genes.
J. Leukoc. Biol. 91, 685–699.
Van den Bossche, J., Baardman, J., and de Winther, M.P. (2015). Metabolic
characterization of polarized M1 and M2 bone marrow-derived macrophages
using real-time extracellular flux analysis. J. Vis. Exp. 105.
Van den Bossche, J., Baardman, J., Otto, N.A., van der Velden, S., Neele, A.E.,
van den Berg, S.M., Luque-Martin, R., Chen, H.J., Boshuizen, M.C., Ahmed,
M., et al. (2016). Mitochondrial dysfunction prevents repolarization of inflam-
matory macrophages. Cell Rep. 17, 684–696.
Van den Bossche, J., O’Neill, L.A., and Menon, D. (2017). Macrophage immu-
nometabolism: where are we (going)? Trends Immunol. 38, 395–406.
van der Windt, G.J., Everts, B., Chang, C.H., Curtis, J.D., Freitas, T.C., Amiel,
E., Pearce, E.J., and Pearce, E.L. (2012). Mitochondrial respiratory capacity is
a critical regulator of CD8+ T cell memory development. Immunity 36, 68–78.
van der Windt, G.J., O’Sullivan, D., Everts, B., Huang, S.C., Buck, M.D., Curtis,
J.D., Chang, C.H., Smith, A.M., Ai, T., Faubert, B., et al. (2013). CD8 memory
T cells have a bioenergetic advantage that underlies their rapid recall ability.
Proc. Natl. Acad. Sci. USA 110, 14336–14341.
van der Windt, G.J.W., Chang, C.H., and Pearce, E.L. (2016). Measuring bio-
energetics in T cells using a Seahorse extracellular flux analyzer. Curr. Protoc.
Immunol. 113, 1–10, 14.
Vats, D., Mukundan, L., Odegaard, J.I., Zhang, L., Smith, K.L., Morel, C.R.,
Wagner, R.A., Greaves, D.R., Murray, P.J., and Chawla, A. (2006). Oxidative
metabolism and PGC-1beta attenuate macrophage-mediated inflammation.
Cell Metab. 4, 13–24.
Wagner, A., Wang, C., DeTomaso, D., Avila-Pacheco, J., Zaghouani, S., Fess-
ler, J., Akama-Garren, E., Pierce, K., Ron-Harel, N., Douglas, V.P., et al. (2020).
In silico modeling of metabolic state in single Th17 cells reveals novel regula-
tors of inflammation and autoimmunity. bioRxiv. https://doi.org/10.1101/2020.
01.23.912717.
Wang, R., Dillon, C.P., Shi, L.Z., Milasta, S., Carter, R., Finkelstein, D., McCor-
mick, L.L., Fitzgerald, P., Chi, H., Munger, J., and Green, D.R. (2011). The tran-
scription factor Myc controls metabolic reprogramming upon T lymphocyte
activation. Immunity 35, 871–882.
Xiao, Z., Dai, Z., and Locasale, J.W. (2019). Metabolic landscape of the tumor
microenvironment at single cell resolution. Nat. Commun. 10, 3763.
Zhang, Y., Kim, M.S., Nguyen, E., and Taylor, D.M. (2020). Modeling metabolic
variation with single-cell expression data. bioRxiv. https://doi.org/10.1101/
2020.01.28.923680.
Molecular Cell

Perspective

Single-Cell RNA Sequencing in Cancer:

Lessons Learned and Emerging Challenges
Mario L. Suvà1,2,* and Itay Tirosh3,*
1Department of Pathology and Center for Cancer Research, Massachusetts General Hospital and Harvard Medical School, Boston,

MA 02114, USA
2Broad Institute of Harvard and MIT, Cambridge, MA 02142, USA
3Department of Molecular Cell Biology, Weizmann Institute of Science, Rehovot 761001, Israel

*Correspondence: suva.mario@mgh.harvard.edu (M.L.S.), itay.tirosh@weizmann.ac.il (I.T.)

https://doi.org/10.1016/j.molcel.2019.05.003

Bulk genomic analyses and expression profiling of clinical specimens have shaped much of our understand-
ing of cancer in patients. However, human tumors are intricate ecosystems composed of diverse cells,
including malignant, immune, and stromal subsets, whose precise characterization is masked by bulk
genomic methods. Single-cell genomic techniques have emerged as powerful approaches to dissect human
tumors at the resolution of individual cells, providing a compelling approach to deciphering cancer biology.
Here, we discuss some of the common themes emerging from initial studies of single-cell RNA sequencing in
cancer and then highlight challenges in cancer biology for which emerging single-cell genomics methods
may provide a compelling approach.

Although cancer is a genetic disease, malignancies are associ-
ated with aberrations in gene expression related to both cancer
cell-intrinsic and extrinsic factors. Genomic profiling of biological
samples has revolutionized our understanding of cancer in pa-
tients. Yet most datasets and analyses of gene expression are
dominated by bulk profiling, in which an entire biological sample
(e.g., tumor) is profiled as a single entity, thus averaging the
expression profiles of the constituent cells. Expression intra-tu-
mor heterogeneity (eITH) governs many critical facets of tumor
biology that are driven by subsets of tumor cells, such as tumor
growth, metastasis, and resistance to treatment. eITH of malig-
nant cells is governed by at least three determinants: (1) genetic
heterogeneity, due to subclonal mutations that arise during tu-
mor evolution, drives diverse cellular programs; (2) epigenetic
and developmental programs, such as those of tissue stem cells
and their differentiated progeny, ascribe cancer cells with an
(often aberrant) cell identity and in some cases enable cellular
plasticity; and (3) extrinsic and spatial factors that determine ox-
ygen, nutrient availability, and cell-cell interactions. Single-cell
RNA sequencing (scRNA-seq) techniques measure the tran-
scriptional output of cells directly in tumor samples; the
measured cellular programs represent an integration of its ge-
netic, epigenetic, and environmental cues and determine cancer
cells fitness, behavior, and response to therapies. As such,
scRNA-seq studies place cellular biology at the center of cancer
biology and offer both different resolution and perspective to
bulk genomic studies. We have recently reviewed initial studies
of single-cell expression profiling in human cancer (Tirosh and
Suvà, 2019). Here, we focus on emerging themes and on future
challenges for the field.
Cell Types, Cell States, and Malignant and Non-
malignant Cells
Analysis of the expression diversity within tumors (or other tis-
sues) typically reveals two layers: highly distinct clusters that
may be considered as ‘‘cell types’’ (e.g., malignant cells,
T cells, and fibroblasts) and further diversity within each of those
clusters that may be considered as ‘‘cell states,’’ such as pro-
gression along the cell cycle, different metabolic states, and
other dynamic programs. Cell types and states may then be an-
notated based on the identity of preferentially expressed genes
(in each cluster or subcluster) along with genetic classification,
including large-scale copy number alterations (CNAs), point mu-
tations, and fusion proteins, that can be inferred from scRNA-
seq data and help to resolve malignant from non-malignant cells
(Filbin et al., 2018; Jerby-Arnon et al., 2018; Patel et al., 2014;
Puram et al., 2017; Tirosh et al., 2016a, 2016b; Venteicher
et al., 2017). An emerging theme from recent scRNA-seq studies
is that malignant cells tend to cluster in their expression profiles
primarily by patient sample, and non-malignant cells cluster in
their expression profiles by cell type, somewhat independently
of the patient of origin (Filbin et al., 2018; Jerby-Arnon et al.,
2018; Tirosh et al., 2016a, 2016b; Venteicher et al., 2017). This
indicates that (1) inter-tumor heterogeneity is typically larger for
malignant cells than for any particular type of non-malignant
cells. (2) For malignant cells, inter-tumor heterogeneity is much
larger than intra-tumor heterogeneity. These results highlight
the considerable degree of inter-tumor heterogeneity and might
be misinterpreted as reflecting a limited role of intra-tumor het-
erogeneity of cancer cells. However, the power of single-cell
methods is that they now further allow us to interrogate the pat-
terns of intra-tumor heterogeneity, which have been difficult to
assess with previous approaches and are often continuous
rather than discrete.
Recurrent Patterns of eITH among Cancer Cells
What then are the patterns of heterogeneity observed among the
malignant cells in an individual tumor? First of all, the level of any
individual gene varies between cells, partially due to the stochas-
tic and noisy nature of gene expression regulation. In addition to

Molecular Cell 75, July 11, 2019 ª 2019 Elsevier Inc. 7


Figure 1. Recurrent Patterns of eITH among Malignant Cells
(A–D) IDH mutant glioma (A; Tirosh et al., 2016b; Venteicher et al., 2017), H3-K27M glioma (B; Filbin et al., 2018), melanoma (C; Tirosh et al., 2016a), and head and
neck cancer (D; Puram et al., 2017). Images represent a scatter-plot in which cells (dots) are scored by the expression of particular programs identified in that
cancer type (x and y axes). In each case, a similar pattern of eITH was identified across multiple patients. Plots were either reproduced (A–C) or generated from the
data (D) of the original publication. Also highlighted in red are the associations between certain subpopulations and important clinical features, as described in the
main text and the original publications.
this stochasticity, however, sets of genes vary coherently be-
tween cells, reflecting different cellular states that might have a
more significant biological implication. For example, concomi-
tant upregulation of dozens of cell-cycle-related genes indicates
that certain cells have entered the cell cycle and others were not
cycling at the moment in which they were profiled (Tirosh et al.,
2016a, 2016b). Similarly, concomitant upregulation of many
genes associated with hypoxia and/or stress responses indicate
that certain cells respond to their immediate microenvironment
and to the lack of nutrients or oxygen by mounting such re-
sponses (Patel et al., 2014; Tirosh et al., 2016a, 2016b). These
two processes—cell cycle and stress responses—reflect com-
mon patterns of cellular heterogeneity within tumors, which are
observed within different patients and across different can-
cer types.
In addition to those ‘‘generic’’ patterns of heterogeneity, there
are also patterns that are context specific and hence may be
more informative about the specific biology of a given tumor
type. Multiple previous studies found that apart from cell cycle
and stress there are particular intra-tumor expression programs
(i.e., sets of genes that coherently vary among malignant cells in
the same tumor) that are found within multiple patients of a
particular cancer type, but not in other cancer types (Figure 1;
Tirosh and Suvà, 2019). Strikingly, in each of those cases, these
were large-scale expression programs that were uncovered by
unbiased analysis, and examination of the associated genes
suggested that they impact on crucial aspects of tumor progres-
sion, including drug resistance, metastasis, and self-renewal. In
melanoma, those programs highlighted distinct MITF and AXL
expression levels of cells within the same tumor, with suggested
implications for resistance to BRAF inhibition (Rambow et al.,
2018; Shaffer et al., 2017; Tirosh et al., 2016a). More recently,
a cancer cell state in melanoma that promotes immune evasion
and resistance to immune checkpoint inhibitors was identified by
scRNA-seq analysis of 33 melanoma samples (Jerby-Arnon
et al., 2018). In head and neck cancer, malignant cells differed
in their expression of epithelial differentiation programs and in
the expression of epithelial-to-mesenchymal transition pro-
grams, with important implications for metastasis (Puram et al.,
8 Molecular Cell 75, July 11, 2019
Molecular Cell
Perspective
2017). In glioblastoma, IDH mutant gliomas, and histone H3-
K27M mutant gliomas, eITH patterns were dominated by glial
differentiation and neurodevelopmental programs. These pro-
grams highlighted putative differentiation hierarchies, with
stem-like cells that were linked to self-renewal and increased
aggressiveness (Filbin et al., 2018; Patel et al., 2014; Tirosh
et al., 2016b; Venteicher et al., 2017).
These initial results suggest two important conclusions. First,
the patterns of eITH are of considerable significance for the
maintenance and progression of tumors, warranting further anal-
ysis and future approaches to target particular tumor subpopu-
lations. Second, the consistency of these patterns across pa-
tients of the same cancer type or subtype suggests that these
reflect fundamental properties of the corresponding cells from
which the cancer originates. For example, gliomas likely origi-
nate from neural stem-like cells, which have an intrinsic capacity
to differentiate toward astrocytes and oligodendrocytes; thus,
the resulting cancer cells display an aberrant differentiation to-
ward astrocyte-like and oligodendrocyte-like cells that dominate
the cellular diversity within gliomas. Similarly, epithelial cells
have the intrinsic capacity to activate mesenchymal programs
that might normally be required during development and wound
healing; thus, the resulting cancer cells may also activate those
programs in response to cues, such as transforming growth fac-
tor b (TGF-b), produced by surrounding fibroblasts.
Thus, these patterns might be thought of as reflecting the
‘‘epigenetic landscape’’ of the cancer cells: only particular pro-
grams may be accessible to cells in each cancer type, based
on the nature of the cell of origin or the impact of oncogenic
transformation. Accordingly, cells may dynamically activate
those ‘‘accessible’’ programs, resulting in their variability within
an individual tumor. This model suggests that it will be important
to define such programs of variability in each tumor type (or sub-
type) and subsequently examine its impact on the functional
properties of the cancer cells and their vulnerabilities.
Integrating Bulk with Single-Cell Tumor Profiles
These initial studies provide a proof of concept for the utility of
single-cell expression profiling of tumors. However, it is also
Molecular Cell
Perspective
recurrence
and MRD
3 4
frozen
samples
throughput
microenvironmental cues
and cell-cell interactions
2 spatial
profiling
scRNA-seq
tumor profiling
multi-omics
1 N-of-1
studies
genetic vs.
non-genetic
mechanisms
7 exceptional
responders
important to realize the limitations of this approach. First,
scRNA-seq provides only a partial sampling of the transcriptome
of cells, with a bias for more highly expressed genes over mid-to-
lowly expressed ones. Second, scRNA-seq is very sensitive to
sample quality and as such is not suited for the profiling of
sub-optimally preserved or handled clinical specimen. Third,
high cost limits the ability to profile large cohorts of tumors,
and only few to a few dozens of samples are typically profiled
in each study. Given these challenges with single-cell ap-
proaches and the comprehensive bulk databases that already
exist, such as The Cancer Genome Atlas (TCGA), we anticipate
that single-cell datasets will not supplant bulk tumor profiling
but instead will be integrated with existing and new bulk profiles
to advance our understanding of tumor biology.
In the past, bulk profiles were analyzed in a somewhat naive
way, ignoring the fact that each bulk profile reflects a composite
of many distinct populations and sometimes misinterpreting sig-
nals of the tumor microenvironment as those of the cancer cells.
Single-cell genomics, along with advances in cancer immuno-
therapies, have emphasized the significance of considering not
just the cancer cells but also the various non-cancer populations.
As a result, many recent methods and studies have focused on
inferring the composition of tumors through deconvolution of
their bulk profiles (Aran et al., 2017; Newman et al., 2015; Puram
et al., 2017; Tirosh et al., 2016a). Such methods typically require
a prior definition of potential tumor components (i.e., cell types),
which can now be derived directly and accurately from scRNA-
seq studies. In the next few years, we expect this trend to
expand and that it will become standard to analyze bulk profiles
in light of their measured or inferred composition. For example, if
certain tumors have high expression of extracellular matrix
(ECM) genes, the naive possibility of an aberrant ECM expres-
Figure 2. Questions in Cancer Biology
rare tumor
subpopulations Addressed by Emerging Directions with
scRNA-Seq
See main text for further description of each of
the biological questions and the emerging ap-
increased
proaches to address it. MRD, minimal residual
disease.
cell-of-origin
and dev. pathways
5
HCA sion by the cancer cells would need to
be contrasted with the alternative option
of abundant ECM-producing fibroblasts.
More generally, any subpopulation of
cells that is identified within tumors
through single-cell approaches could
large cohorts
model systems then be investigated for its estimated
abundance and associations across
large cohorts of bulk samples, thereby
partially circumventing the limited size
6 of single-cell cohorts.
significance of
subpopulations
Emerging Challenges and
Directions in Single-Cell Cancer
Analysis
In the next few years, the approach of
investigating tumors through single-cell
RNA-seq will be further integrated into cancer research and
will be used to address an increasing number of questions about
tumor biology (Figure 2). Below, we briefly describe some of the
main questions that are likely to be addressed and point to
emerging technologies in the area of scRNA-seq that would
facilitate these studies.
Distinguishing the Contributions of Genetic versus Non-
genetic Mechanisms: Single-Cell Multi-omics Analyses
Genetic heterogeneity is a prominent mechanism that generates
diversity between and within tumors. However, an increasing
amount of evidence points to the significance of non-genetic
mechanisms. For example, in the context of eITH, we have
recently shown that cellular hierarchies in IDH mutant and his-
tone H3-K27M mutant glioma appear to be largely independent
of genetics (Filbin et al., 2018; Tirosh et al., 2016b; Venteicher
et al., 2017). Thus, an important challenge is to map both cell
state and genetic state in the same cells and begin to distinguish
between these mechanisms. Notably, therapeutic approaches
for dealing with genetic versus non-genetic ITH may be quite
distinct, underscoring the significance of this question. To
date, most scRNA-seq studies either ignored the genetic states
of cells or had a partial capacity to evaluate genetic states, by (1)
detecting mutations in the RNA with limited sensitivity, (2) tar-
geted amplification of individual mutations, or (3) inferring
CNAs (Filbin et al., 2018; Giustacchini et al., 2017; Tirosh and
Suvà, 2019). However, multiple technologies are being devel-
oped (Macaulay et al., 2015, 2016, 2017) and improved to enable
such joint profiling of cell state and genetics, and these are ex-
pected to provide a more integrated view in the future (Nam
et al., 2018; Rodriguez-Meira et al., 2019; Velten et al., 2018).
For example, a recent study in acute myelogenous leukemia
Molecular Cell 75, July 11, 2019 9

(AML) highlighted that cell type composition correlated with pro-
totypic genetic alterations (van Galen et al., 2019). More gener-
ally, ‘‘multi-OMICs’’ approaches are being developed to
combine the measurement of mRNA levels with various other
features, such as protein levels, DNA methylation, chromatin
accessibility, clonal expansion, etc. These approaches will pro-
vide the ability to identify mechanisms that generate eITH and
advance scRNA-seq toward more detailed mechanisms.
Cell-Cell Interactions and Spatial Location: Spatially
Resolved Single-Cell Technologies
Each tumor is analogous to an ecosystem, and thus, the state of
each cell may be highly dependent on its exact spatial location
and interaction with adjacent cells. These complex interactions
and their impact on cancer cell biology remain poorly under-
stood. Coupling transcriptomes with spatial information will
dramatically improve the ability to predict and further charac-
terize such interactions, and accordingly, many approaches
are being developed to couple scRNA-seq with spatial informa-
tion. To date, most scRNA-seq studies have measured the state
of cells without preserving any information on their location.
Other studies have measured or inferred the spatial location of
cells but typically characterized the cells by either multiplexed
markers (Chen et al., 2015) or in lower (non-cellular) resolution
(Chen et al., 2017) or relied on a stereotypical tissue structure,
which is a more effective strategy for normal than for tumor tis-
sues (Halpern et al., 2017; Satija et al., 2015). Continuous im-
provements in spatially resolved technologies for tumor profiling,
as well as the integration of data from disparate technologies,
will dramatically advance these directions over the next few
years and improve our understanding of cell-cell interactions
and spatial effects.
Apart from spatial profiling, the study of cell-cell interactions
within tumors will advance through increased focus on the
expression of ligands and receptors as central mediators of
cellular communication. Databases of ligand-receptor com-
plexes, coupled with a definition of tumor cell subpopulations
by scRNA-seq data, highlight potential cell-cell interactions, in
which one population produces a ligand that signals to another
population expressing the corresponding receptor (Puram
et al., 2017; Vento-Tormo et al., 2018). The complexity of the tu-
mor microenvironment implies that the number of such potential
interactions would be large, warranting follow-up studies that will
determine the significance of individual interactions.
Residual Disease and Recurrences: Leveraging Single-
Nucleus Profiling (Frozen)
In many cancer types, initial therapeutic responses are typically
followed by the recurrence of tumors with increased aggressive-
ness and drug resistance, highlighting the significance of both
the residual disease and its evolution and growth toward estab-
lishing the recurrent tumors. Previous studies compared primary
and recurrent tumors and revealed various genetic causes of
drug resistance. Future studies will extend this approach to com-
parison by scRNA-seq and provide a more detailed view of
recurrent tumors and their difference from the corresponding pri-
mary tumors (Brady et al., 2017; Kim et al., 2018). An important
challenge is that current protocols require fresh tumor samples
10 Molecular Cell 75, July 11, 2019
Molecular Cell
Perspective
for scRNA-seq, creating a severe bottleneck for processing of
matched primary and recurrent samples given the time until
recurrence. However, multiple approaches may enable profiling
of frozen samples and thereby open the door to characterizing
collections of samples that are stored in hospitals and labs,
including longitudinal cohorts. For example, frozen or fixed sam-
ples may be analyzed by isolation and sequencing of single
nuclei (instead of single cells), as recently demonstrated (Gao
et al., 2017; Habib et al., 2016, 2017). In addition to recurrent tu-
mors, scRNA-seq provides a compelling approach to directly
characterize the rare residual cells that survive through treat-
ments and ultimately underlie tumor recurrence. Such rare cells,
typically referred to as minimal residual disease (MRD), may be
profiled in animal models (Rambow et al., 2018) and compared
to both primary and recurrent tumors, providing additional in-
sights to their capacity to evade treatments, remain dormant,
and finally give rise to recurrent tumors.
Tumor Subpopulations: Their Detection, Functional
Significance, and Origin
Tumor progression is an evolutionary process, and hence, only
few cells are sufficient for a new phenotype to gradually emerge
through selection. Consequently, critical subpopulations, such
as cancer stem cells, drug-resistant cells, and migratory cells
that underlie metastasis, might reflect a very rare subpopulation
(e.g., less than 0.1%) that escapes detection by most methods,
including current scRNA-seq approaches. Importantly, although
previous scRNA-seq studies were performed with platforms that
measure limited cell numbers (e.g., tens to hundreds), more
recent and ongoing studies now leverage droplet microfluidics
or other technologies (Gierahn et al., 2017; Jerby- Arnon et al.,
2018; Cao et al., 2017) that enable an order of magnitude
more cells in each batch (e.g., thousands), and future technolo-
gies are likely to further expand this capacity into tens of thou-
sands to even millions of cells, which would help uncover very
rare subpopulations.
Once unique subpopulations have been uncovered, their func-
tional and clinical significance remains difficult to ascertain. Two
main approaches enable further analysis of such subpopula-
tions. First, studies of large cohorts would uncover statistical as-
sociations between the presence and/or frequency of subpopu-
lations and clinical features, such as survival, metastasis, and
drug responses. Given the cost and labor associated with
scRNA-seq of large cohorts, we anticipate that many of these
analyses will leverage available bulk RNA-seq datasets and
use computational approaches for their deconvolution and the
inference of tumor composition, as described above. However,
whether these would be based on single-cell or bulk RNA-seq
profiles, such correlative studies will require mechanistic
follow-ups. Thus, a second approach will rely on the detailed
characterization of such subpopulations by scRNA-seq to iden-
tify markers that would enable isolation of these subpopulations
and further functional studies in animal and culture models.
As described above, recent scRNA-seq studies uncovered
remarkable similarities between tumor expression profiles and
those of normal developmental cell types, raising specific hy-
potheses about the origin of cancers and the ongoing differenti-
ation processes that occur within them. Through extensive
Molecular Cell
Perspective
scRNA-seq of tumors, of normal tissues, and of developmental
processes, for example, through the human cell atlas (HCA)
initiative, the number of such observations is expected to
dramatically increase over the next few years, including refine-
ment of many previous observations. For example, scRNA-seq
of H3-K27M midline gliomas (Filbin et al., 2018) highlighted the
similarity of malignant cells to oligodendrocytic precursor cells
(OPCs), supporting previous literature that suggested that
OPCs are a candidate cell of origin for this tumor type (Gibson
et al., 2018). A caveat with these inferences is that scRNA-seq
of an established tumor cannot identify the cell in which the mu-
tation first occurred (‘‘cell of mutation’’), which can be distinct
from the cell that gives rise to the tumor. Additionally, compari-
sons between malignant cells and normal cells not only highlight
similarities but also provide a more detailed view of the differ-
ences between tumor cells and their presumed cells of origin
and thereby increase our understanding of the oncogenic
process.
Unique Tumor Phenotypes and Exceptional
Responders: Single-Cell Analysis in N = 1 Cohorts
Although therapeutic decisions are based on statistical patterns
in large cohorts, outliers are quite common and present impor-
tant conundrums. For example, ‘‘exceptional responders’’ illus-
trate our limited understanding of treatment response and the
theoretical possibility to improve patient stratifications (Ferreri
et al., 2010). In the past, it has been difficult to understand the
unique phenotypes of such cases, which were studied primarily
by bulk genetic analysis. However, scRNA-seq, along with the
associated technological improvements described above (e.g.,
spatial profiling and multi-OMIC techniques), enable character-
ization of tumor ecosystems at unprecedented depth and
breadth. We propose that these would be ideal to begin to iden-
tify the basis for unique phenotypes and accordingly that N-of-1
scRNA-seq studies may be particularly informative. Because
extreme phenotypes are often detected late, this approach will
also likely depend on analysis of frozen samples that have
been archived.
Concluding Remarks
Tumor heterogeneity has puzzled cancer biologists and clini-
cians for a long time. As genomic tools were lacking to accu-
rately measure individual cells in a tumor, we relied on average
measurements or single markers, masking many critical aspects
of ITH and hindering our understanding of tumor biology. The
precise characterization of single-cell programs in tumors places
cellular biology at the center of cancer biology. By measuring the
transcriptional output as a function of both cell type and ge-
netics, these efforts lay the foundation for renewed understand-
ing of cancer initiation and evolution. The comprehensive
profiling of single-cell programs in cancer also sheds light on tu-
mor classification based on bulk measurements by enabling the
deconvolution of signal and highlighting unanticipated patterns
of ITH. Although the increased understanding of ITH might at first
seem daunting, common cellular states and programs are
emerging and are offering new candidate vulnerabilities that
could be exploited for therapeutics. The initial studies discussed
here are the beginning of an exciting era of single-cell genomics
of human cancer; they are paving the way for many more discov-
eries that will come from improved technologies, resolution,
computational methods, and broader clinical settings.
ACKNOWLEDGMENTS
We thank Ania Hupalowska for help with graphic design. This work was sup-
ported by grants from the Howard Goodman Fellowship at MGH (M.L.S.),
the Merkin Institute Fellowship at the Broad Institute of MIT and Harvard
(M.L.S.), the Sontag Foundation (M.L.S.), the Zuckerman STEM Leadership
Program (I.T.), the Human Frontiers Science Program (I.T.), the Mexican
Friends New Generation (I.T.), the Benoziyo Endowment Fund for the
Advancement of Science (I.T.), and start-up funds from the MGH Department
of Pathology (M.L.S.) and the Weizmann Institute of Science (I.T.). I.T. is an
incumbent of the Dr. Celia Zwillenberg-Fridman and Dr. Lutz Zwillenberg
Career Development Chair.
REFERENCES
Aran, D., Hu, Z., and Butte, A.J. (2017). xCell: digitally portraying the tissue
cellular heterogeneity landscape. Genome Biol. 18, 220.
Brady, S.W., McQuerry, J.A., Qiao, Y., Piccolo, S.R., Shrestha, G., Jenkins,
D.F., Layer, R.M., Pedersen, B.S., Miller, R.H., Esch, A., et al. (2017).
Combating subclonal evolution of resistant cancer phenotypes. Nat. Commun.
8, 1231.
Cao, J., Packer, J.S., Ramani, V., Cusanovich, D.A., Huynh, C., Daza, R., Qiu,
X., Lee, C., Furlan, S.N., Steemers, F.J., et al. (2017). Comprehensive single-
cell transcriptional profiling of a multicellular organism. Science 357, 661–667.
Chen, K.H., Boettiger, A.N., Moffitt, J.R., Wang, S., and Zhuang, X. (2015). RNA
imaging. Spatially resolved, highly multiplexed RNA profiling in single cells.
Science 348, aaa6090.
Chen, J., Suo, S., Tam, P.P., Han, J.J., Peng, G., and Jing, N. (2017). Spatial
transcriptomic analysis of cryosectioned tissue samples with Geo-seq. Nat.
Protoc. 12, 566–580.
Ferreri, A.J., Illerhaus, G., Zucca, E., and Cavalli, F.; International Extranodal
Lymphoma Study Group (2010). Flows and flaws in primary central nervous
system lymphoma. Nat. Rev. Clin. Oncol. 7, 472.
Filbin, M.G., Tirosh, I., Hovestadt, V., Shaw, M.L., Escalante, L.E., Mathewson,
N.D., Neftel, C., Frank, N., Pelton, K., Hebert, C.M., et al. (2018). Develop-
mental and oncogenic programs in H3K27M gliomas dissected by single-
cell RNA-seq. Science 360, 331–335.
Gao, R., Kim, C., Sei, E., Foukakis, T., Crosetto, N., Chan, L.K., Srinivasan, M.,
Zhang, H., Meric-Bernstam, F., and Navin, N. (2017). Nanogrid single-nucleus
RNA sequencing reveals phenotypic diversity in breast cancer. Nat. Commun.
8, 228.
Gibson, E.M., Geraghty, A.C., and Monje, M. (2018). Bad wrap: myelin and
myelin plasticity in health and disease. Dev. Neurobiol. 78, 123–135.
Gierahn, T.M., Wadsworth, M.H., 2nd, Hughes, T.K., Bryson, B.D., Butler, A.,
Satija, R., Fortune, S., Love, J.C., and Shalek, A.K. (2017). Seq-Well: portable,
low-cost RNA sequencing of single cells at high throughput. Nat. Methods 14,
395–398.
Giustacchini, A., Thongjuea, S., Barkas, N., Woll, P.S., Povinelli, B.J., Booth,
C.A.G., Sopp, P., Norfo, R., Rodriguez-Meira, A., Ashley, N., et al. (2017). Sin-
gle-cell transcriptomics uncovers distinct molecular signatures of stem cells in
chronic myeloid leukemia. Nat. Med. 23, 692–702.
Habib, N., Li, Y., Heidenreich, M., Swiech, L., Avraham-Davidi, I., Trombetta,
J.J., Hession, C., Zhang, F., and Regev, A. (2016). Div-seq: single-nucleus
RNA-seq reveals dynamics of rare adult newborn neurons. Science 353,
925–928.
Habib, N., Avraham-Davidi, I., Basu, A., Burks, T., Shekhar, K., Hofree, M.,
Choudhury, S.R., Aguet, F., Gelfand, E., Ardlie, K., et al. (2017). Massively par-
allel single-nucleus RNA-seq with DroNc-seq. Nat. Methods 14, 955–958.
Halpern, K.B., Shenhav, R., Matcovitch-Natan, O., Toth, B., Lemze, D., Go-
lan, M., Massasa, E.E., Baydatch, S., Landen, S., Moor, A.E., et al. (2017).
Molecular Cell 75, July 11, 2019 11

Single-cell spatial reconstruction reveals global division of labour in the
mammalian liver. Nature 542, 352–356.
Jerby-Arnon, L., Shah, P., Cuoco, M.S., Rodman, C., Su, M.J., Melms, J.C.,
Leeson, R., Kanodia, A., Mei, S., Lin, J.R., et al. (2018). A cancer cell program
promotes T cell exclusion and resistance to checkpoint blockade. Cell 175,
984–997.e24.
Kim, C., Gao, R., Sei, E., Brandt, R., Hartman, J., Hatschek, T., Crosetto, N.,
Foukakis, T., and Navin, N.E. (2018). Chemoresistance evolution in triple-
negative breast cancer delineated by single-cell sequencing. Cell 173, 879–
893.e13.
Macaulay, I.C., Haerty, W., Kumar, P., Li, Y.I., Hu, T.X., Teng, M.J., Goolam,
M., Saurat, N., Coupland, P., Shirley, L.M., et al. (2015). G&T-seq: parallel
sequencing of single-cell genomes and transcriptomes. Nat. Methods 12,
519–522.
Macaulay, I.C., Teng, M.J., Haerty, W., Kumar, P., Ponting, C.P., and Voet, T.
(2016). Separation and parallel sequencing of the genomes and transcrip-
tomes of single cells using G&T-seq. Nat. Protoc 11, 2081–2103.
Macaulay, I.C., Ponting, C.P., and Voet, T. (2017). Single-cell multiomics: mul-
tiple measurements from single cells. Trends Genet. 33, 155–168.
Nam, A.S., Kim, K.-T., Chaligne, R., Izzo, F., Ang, C., Abu-Zeinah, G., Omans,
N.D., Taylor, J., Pastore, A., Alonso, A., et al. (2018). High throughput droplet
single-cell Genotyping of Transcriptomes (GoT) reveals the cell identity depen-
dency of the impact of somatic mutations. bioRxiv. https://doi.org/10.1101/
444687.
Newman, A.M., Liu, C.L., Green, M.R., Gentles, A.J., Feng, W., Xu, Y., Hoang,
C.D., Diehn, M., and Alizadeh, A.A. (2015). Robust enumeration of cell subsets
from tissue expression profiles. Nat. Methods 12, 453–457.
Patel, A.P., Tirosh, I., Trombetta, J.J., Shalek, A.K., Gillespie, S.M., Wakimoto,
H., Cahill, D.P., Nahed, B.V., Curry, W.T., Martuza, R.L., et al. (2014). Single-
cell RNA-seq highlights intratumoral heterogeneity in primary glioblastoma.
Science 344, 1396–1401.
Puram, S.V., Tirosh, I., Parikh, A.S., Patel, A.P., Yizhak, K., Gillespie, S.,
Rodman, C., Luo, C.L., Mroz, E.A., Emerick, K.S., et al. (2017). Single-cell tran-
scriptomic analysis of primary and metastatic tumor ecosystems in head and
neck cancer. Cell 171, 1611–1624.e24.
Rambow, F., Rogiers, A., Marin-Bejar, O., Aibar, S., Femel, J., Dewaele, M.,
Karras, P., Brown, D., Chang, Y.H., Debiec-Rychter, M., et al. (2018). Toward
minimal residual disease-directed therapy in melanoma. Cell 174, 843–
855.e19.
12 Molecular Cell 75, July 11, 2019
Molecular Cell
Perspective
Rodriguez-Meira, A., Buck, G., Clark, S.-A., Povinelli, B.J., Alcolea, V., Louka,
E., McGowan, S., Hamblin, A., Sousos, N., Barkas, N., et al. (2019). Unraveling
intratumoral heterogeneity through high-sensitivity single-cell mutational anal-
ysis and parallel RNA sequencing. Mol. Cell 73, 1292–1305.E8.
Satija, R., Farrell, J.A., Gennert, D., Schier, A.F., and Regev, A. (2015). Spatial
reconstruction of single-cell gene expression data. Nat. Biotechnol 33,
495–502.
Shaffer, S.M., Dunagin, M.C., Torborg, S.R., Torre, E.A., Emert, B., Krepler, C.,
Beqiri, M., Sproesser, K., Brafford, P.A., Xiao, M., et al. (2017). Rare cell vari-
ability and drug-induced reprogramming as a mode of cancer drug resistance.
Nature 546, 431–435.
Tirosh, I., and Suvà, M.L. (2019). Deciphering Human Tumor Biology by Single-
Cell Expression Profiling. Annu. Rev. Cancer Biol 3, 151–166.
Tirosh, I., Izar, B., Prakadan, S.M., Wadsworth, M.H., 2nd, Treacy, D., Trom-
betta, J.J., Rotem, A., Rodman, C., Lian, C., Murphy, G., et al. (2016a). Dissect-
ing the multicellular ecosystem of metastatic melanoma by single-cell RNA-
seq. Science 352, 189–196.
Tirosh, I., Venteicher, A.S., Hebert, C., Escalante, L.E., Patel, A.P., Yizhak, K.,
Fisher, J.M., Rodman, C., Mount, C., Filbin, M.G., et al. (2016b). Single-cell
RNA-seq supports a developmental hierarchy in human oligodendroglioma.
Nature 539, 309–313.
van Galen, P., Hovestadt, V., Wadsworth Ii, M.H., Hughes, T.K., Griffin, G.K.,
Battaglia, S., Verga, J.A., Stephansky, J., Pastika, T.J., Lombardi Story, J.,
et al. (2019). Single-cell RNA-seq reveals AML hierarchies relevant to disease
progression and immunity. Cell 176, 1265–1281.e24.
Velten, L., Story, B.A., Hernandez-Malmierca, P., Milbank, J., Paulsen, M.,
Lutz, C., Nowak, D., Jann, J.-C., Pabst, C., Boch, T., et al. (2018). MutaSeq re-
veals the transcriptomic consequences of clonal evolution in acute myeloid
leukemia. bioRxiv. https://doi.org/10.1101/500108.
Venteicher, A.S., Tirosh, I., Hebert, C., Yizhak, K., Neftel, C., Filbin, M.G., Hov-
estadt, V., Escalante, L.E., Shaw, M.L., Rodman, C., et al. (2017). Decoupling
genetics, lineages, and microenvironment in IDH-mutant gliomas by single-
cell RNA-seq. Science 355, eaai8478.
Vento-Tormo, R., Efremova, M., Botting, R.A., Turco, M.Y., Vento-Tormo, M.,
Meyer, K.B., Park, J.E., Stephenson, E., Polanski, K., Goncalves, A., et al.
(2018). Single-cell reconstruction of the early maternal-fetal interface in
humans. Nature 563, 347–353.
Trends in Cell Biology

Review

Subcellular Chemical Imaging: New Avenues in

Cell Biology
Johan Decelle,1,* Giulia Veronesi,2,3 Benoit Gallet,4 Hryhoriy Stryhanyuk,5 Pietro Benettoni,5 Matthias Schmidt,5
Rémi Tucoulou,3 Melissa Passarelli,6 Sylvain Bohic,3,7 Peta Clode,8,9 and Niculina Musat5

To better understand the physiology and acclimation capability of the cell, one of Highlights
the great challenges of the future is to access the interior of a cell and unveil its Chemical imaging instruments have
chemical landscape (composition and distribution of elements and molecules). substantially improved in sensitivity
and spatial resolution and have become
Chemical imaging has greatly improved in sensitivity and spatial resolution to
the method of choice to visualize and
visualize and quantify nutrients, metabolites, toxic elements, and drugs in single quantify elements, isotopes, and mole-
cells at the subcellular level. This review aims to present the current potential of cules at the subcellular level in various
these emerging imaging technologies and to guide biologists towards a strategy disciplines, such as cell biology, eco-
physiology, and toxicology.
for interrogating biological processes at the nanoscale. We also describe various
solutions to combine multiple imaging techniques in a correlative way and pro- Since chemical imaging provides low
vide perspectives and future directions for integrative subcellular imaging across contrast, it is often combined with elec-
different disciplines. tron microscopy in a correlative way to
unambiguously identify the localization
of an element or molecule in the cell.
New Avenues for the Subcellular Exploration of the Cell
The advent of electron microscopy (EM) in the mid-20th century was a formidable tool for the de- Different complementary chemical imag-
ing instruments (e.g., SIMS, X-ray fluo-
tailed exploration of a cell’s structure at nanoscale resolution. Nowadays, a key challenge in cell
rescence) can be combined to correlate
biology is to understand the activity and function of organelles and cellular compartments and elemental, isotopic, and molecular infor-
their role in the metabolism and physiology of a cell. Omics bulk analyses (e.g., transcriptomics, mation from a region of the cell or from
metabolomics) have greatly improved our understanding of cellular mechanisms but provide the entire cell in two or three dimensions.
only averaged information on extracted molecules from numerous lysed cells. Hence, spatial
New sample preparation methods and
information at the subcellular level is a missing dimension to fully interpret the phenotypic state analytical instruments enable indirect or
of a cell and assess heterogeneity in a population. Chemical imaging (see Glossary) techniques direct correlative subcellular imaging
are now able to reveal the chemical landscape of cells (i.e., the composition and distribution of from cells in their close-to-native state.

elements and molecules) at the subcellular level without the need to add or genetically encode
fluorescent labels. Probing the elemental and molecular composition of organelles and subcellular
structures can reveal fundamental information about the function and physiology of a cell in re- 1
Cell and Plant Physiology Laboratory,
sponse to different conditions. The subcellular distribution of some elements (e.g., the macronu- Université Grenoble Alpes, CNRS, CEA,
INRAE, IRIG, 38000, Grenoble, France
trients N, P, and S), which are essential building blocks of biomolecules (e.g., DNA, proteins, 2
Chemistry and Biology of Metals
lipids), can reflect the metabolic roles and needs of organelles [1]. Trace metals (e.g., Fe, Cu, Laboratory, Université Grenoble Alpes,
Zn) play a fundamental role in various biochemical functions of the cell and their homeostasis CNRS, CEA, IRIG, Grenoble, France
3
ESRF – The European Synchrotron,
and compartmentalization need to be tightly controlled to avoid cell death and severe patholo- Grenoble, France
gies. More particularly, metals are key players in parasitic and viral infections, cancer cells, and 4
Institut de Biologie Structurale,
neurodegenerative diseases [2]. In the biomedical field, the increasing human exposure to Université Grenoble Alpes, CNRS, CEA,
Grenoble, France
exogenous compounds (e.g., metal-based nanoparticles, toxic elements) and the use of thera- 5
Helmholtz Centre for Environmental
peutic drugs call for imaging techniques to visualize their fate in tissues and cells and to assess Research – UFZ, Department of Isotope
their toxicity and impact on the homeostasis of native elements [3,4]. In addition to elements, Biogeochemistry, Leipzig, Germany
6
École Polytechnique Fédérale de
the localization of metabolites (e.g., sugars, lipids) in cells is essential to fully understand metabolic Lausanne (EPFL), Laboratory for
processes. Therefore, subcellular mapping of elements and metabolites is becoming indispensable Biological Geochemistry,
to investigate the physiology and metabolism of healthy and diseased cell types, to understand Lausanne, Switzerland
7
INSERM – UA7 – Synchrotron Radiation
cellular interactions in tissues or with beneficial cells (e.g., symbioses) and pathogens (e.g., viral for Biomedicine, STROBE, Université
or bacterial infection), and their adaptive response to abiotic stresses. Grenoble Alpes, Grenoble, France

Trends in Cell Biology, March 2020, Vol. 30, No. 3 https://doi.org/10.1016/j.tcb.2019.12.007 173
© 2020 Elsevier Ltd. All rights reserved.
Trends in Cell Biology

8
Recent technological progress in chemical imaging has substantially improved the sensitivity and Centre for Microscopy Characterisation
and Analysis, The University of Western
spatial resolution, allowing the disentangling of cellular compartments in a single cell. However, Australia, Crawley, WA, Australia
9
the multiplicity of these complex imaging techniques requires guidance for nonspecialists. An UWA School of Biological Sciences, The
overview of the chemical imaging techniques currently available is therefore needed to help University of Western Australia, Crawley,
WA, Australia
biologists integrate the subcellular scale in their studies while being aware of its potential and
limitations. Each chemical imaging platform presents experimental specifications that make
them more sensitive to some elements or molecules, so different platforms need to be combined *Correspondence:
to have a comprehensive view of the chemical landscape of a cell. Moreover, since chemical johan.decelle@univ-grenoble-alpes.fr
imaging generally provides limited information on cell ultrastructure, EM is often required to inter- (J. Decelle).

pret the intracellular localization of elements and molecules. Correlation between light microscopy
and EM (CLEM) is well established [5,6], but correlation between EM and chemical imaging is less
developed. Bridging the data acquired with different high-resolution imaging strategies is the next
challenge and will make correlative subcellular imaging a new, powerful research tool towards
integrative cell biology.

This review aims to present the potential and limitations of state-of-the-art chemical imaging
techniques for nonspecialists who seek to obtain chemical information at the subcellular level.
We aim to guide biologists to the appropriate imaging technique and associated sample prepa-
ration to visualize and quantify elements or biomolecules in cells. We also summarize the new
developments in correlative subcellular imaging (Figure 1, Key Figure), highlight the role of such
combinatory techniques in disentangling the biochemical processes of a cell, and discuss future
challenges and directions in the field.

Potential and Limitations of Subcellular Chemical Imaging Platforms and

Required Sample Preparation
Multiple chemical imaging instruments are capable of visualizing the molecular, elemental, and
isotopic composition of a cell with high lateral resolution [7]. These microscopes are generally
equipped with a focused high-energy primary beam of electrons, protons, photons, or ions that
scans the surface of the sample and obtain quantitative information in a spatially
resolved manner (Box 1). The instrument and experimental setup need to be carefully chosen
according to the research question and the target elements or molecules of interest. Here we
focus on the key methodologies that are routinely used to provide subcellular information.
Lower-resolution technologies (e.g., MALDI, LA-ICP-MS) are not discussed in detail here
(see the review in [8]).

X-ray fluorescence (XRF) microscopy relies on the excitation of core electrons of atoms that
leads to X-ray emissions, which are specific to elements in the sample (Box 1). The primary probe
determines the technique: electrons in energy dispersive X-ray spectrometry [scanning/transmission
EM (S/TEM)-EDS]; protons in particle-induced X-ray emission (PIXE); and synchrotron-generated
photons in synchrotron XRF (S-XRF) imaging. These analytical techniques can be used to visualize
and quantify the distribution of macronutrients (e.g., P, S), key trace elements (e.g., Mn, Fe, Cu, Zn,
Se), toxic heavy metals (e.g., Hg, Pb), and pharmacological compounds (e.g., organometallic
compounds based on Pt, Ir, Os, or Ru). S/TEM-EDS can provide the highest spatial resolution
(subnanometer in TEM) with sensitivity of ~1000 ppm (1 mg g−1) [9]. Compared with this, PIXE is
less spatially resolved (submicron) but is more sensitive (ppm range) [7]. However, overall, S-XRF
provides arguably the best combination of high-spatial-resolution capabilities (down to a few
tens of nanometers) and high sensitivity (sub-ppm) to light and heavy elements (Figure 2 and
Box 1) [10,11]. S-XRF has allowed the mapping and quantification of metals such as Fe, Zn, and
Cu in microalgal and human cells [12–14] as well as silica, drugs, organometallic molecules,
and titanium oxide nanoparticles in cancer cells [2,15–19]. In combination with XRF imaging,

174 Trends in Cell Biology, March 2020, Vol. 30 , No. 3

Trends in Cell Biology

Key Figure Glossary

Outline of Correlative Multimodal Subcellular Imaging Workflow Including Chemical imaging: spatial
characterization of the chemical
Electron and Chemical Imaging composition of a sample (isotopes,
elements, molecules). This can be
achieved by multiple high-resolution
imaging platforms using different
physical processes to interrogate
subcellular information (e.g. XRF,
ionization).
Chemical landscape: composition
and distribution patterns of elements,
isotopes, and molecules in a sample
(e.g., cell). Its visualization cannot be
obtained with a single imaging platform
but different techniques need to be used
in a correlative way.
Correlative EM and chemical
imaging: workflow to prepare samples,
obtain micrographs from different
complementary imaging platforms, and
overlap morphological and chemical
(elements/molecules) information from
the same region of a specimen.
Hybrid SIMS: [Bi]n (n = 1, 3, 5, 7) or [Ar]
n (n ~ 1000) gas-cluster ion source used
for the analysis of large molecular ions.
The OrbiTrap analyzer provides a MRP
of ~10E5 that enables precise
compound identification.
Multimodal imaging: microscopy
observations of the same sample using
two or more imaging platforms to obtain
complementary morphological and
chemical information (e.g., light and EM,
nanoSIMS and S-XRF).
NanoSIMS: single-atomic Cs+ or O-
primary ions are used for both sputtering
and analysis; the ionized material is then
analyzed by a Mattauch–Herzog mass
spectrometer that allows the parallel
detection of a maximum of seven
masses. The high energy of the primary
ions causes strong fragmentation of
molecules down to single-atomic ions,
allowing quantitation of changes in
isotopic composition.
Secondary ion mass spectrometry
(SIMS): secondary ions are sputtered
away from the topmost layer of a sample
by a focused primary ion beam and
analyzed in a mass spectrometer.
NanoSIMS, ToF-SIMS, and hybrid SIMS
are SIMS imaging techniques with
primary ion beams of different sources
and energies and with different mass
spectrometers to probe elements,
isotopes, and small molecules.
Time-of-flight (ToF)-SIMS: pulsed [Bi]
n (n = 1, 3, 5, 7) or [Ar]n (n ~ 1000) cluster
Trends in Cell Biology
ion sources are mainly used for analysis
(See figure legend at the bottom of the next page.) while Ar[n] clusters, Cs+ or O- can be

Trends in Cell Biology, March 2020, Vol. 30, No. 3 175

Trends in Cell Biology

X-ray absorption spectroscopy (XAS) can be performed to reveal the chemical speciation of a used as sputtering sources. The use of
cluster ions for analysis reduces the
target element. XAS has disclosed the chemical transformations of indium-based nanocrystals
fragmentation on impact, leading to the
and of osmium-based anticancer drugs in cancer cells [20,21]. The modulations of the preservation of molecular species. The
XAS spectra also allowed mapping of the distribution of the different chemical states of ToF mass spectrometer allows the
S (e.g., sulfate esters, inorganic sulfate) in a biological tissue, to understand their role in cell simultaneous detection of all masses.
Vitreous cell: frozen hydrated cell with
differentiation [22]. amorphous ice (i.e., without crystals that
can alter the ultrastructure and chemical
Secondary ion mass spectrometry (SIMS) instruments are based on the analysis of the mass composition of cells). Vitreous cells can
of elements and molecules. Secondary ions are sputtered away from the topmost layer of a be obtained using high-pressure
freezing or plunge-freezing machines.
sample by a focused primary ion beam and analyzed in a mass spectrometer (Box 1). NanoSIMS X-ray fluorescence (XRF): physical
is a SIMS instrument particularly suitable for probing macronutrients and metals in cells at a lateral process comprising the emission of X-
resolution down to 50 nm (Figure 2) [23–26]. For instance, P, S, Ca, Fe, Zn, Mn, and Cu have been rays from a specimen following the
excitation of core electrons of atoms;
mapped in cells [12,25,27,28]. Morphological features of the cell can be revealed by a secondary
analysis of the emitted X-rays allows the
electron signal (only in negative extraction mode) and from different secondary ions, such as identification of the elemental content of
cyanide (12C14N−) and phosphorous (31P−) showing the overall shape and internal compartments the specimen. The excitation of
of the cell and the nucleus, respectively. The high mass-resolving power of nanoSIMS can also electrons can be achieved by a beam of
electrons, protons, or photons.
unveil the isotopic composition of a cell (i.e., being able to distinguish between 12C15N and
13 14 X-ray phase-contrast tomography:
C N) and so is highly suitable for stable isotope probing (SIP) [29,30]. SIP-nanoSIMS allows tomographic technique sensitive to the
quantitation of metabolic activities at the subcellular level (e.g., 50–100 nm), such as C and N refraction of X-rays in matter, leading to
assimilation [31–34]. This technique has been used to understand nutrient exchange between phase variations of the X-rays
depending on the sample’s electron
cells in symbiotic, pathogenic, and virus–host interactions [35–40] and the localization of drug
density, and particularly adapted to
compounds in human cells [41] (Figure 2). reveal weakly absorbing features like
those present in biological samples.
With time-of-flight (ToF)-SIMS (Box 1), molecular information can be obtained since the ion
probe (polyatomic or gas cluster) is less destructive than in nanoSIMS (monoatomic). The softer
ionization conditions compared with nanoSIMS allow spatially resolved analysis of large molecular
fragment species in the range 1 to ~1000 Da with a typical lateral resolution of 100 nm to 5 μm.
Different analysis modes exist: spectrometry mode to obtain high mass resolution, imaging mode
to obtain high lateral resolution, and delayed-extraction mode to combine high mass resolution
(10 000 MRP) with high lateral resolution (400 nm) [42]. Delayed extraction is now widely used
to image organic samples and recently was shown to achieve a 108-nm lateral resolution to
visualize single particles in algal biofilms [43,44]. ToF-SIMS is a useful method for studying
small molecules [45–47] and lipids in cells, especially in lipid-related diseases, such as cancer,
Duchenne muscular dystrophy, and atherosclerosis (Figure 2) [48,49].

Seeing is believing, but what we see critically depends on the sample preparation. Sample
preparation is one of the most fundamental steps and should aim to preserve cells as close as
possible to their native state – the Holy Grail in cell biology. The ideal method is one that fixes
and conserves both the ultrastructure of the cell and its native chemical composition (Box 2).
However, sample preparation is highly specific to both the sample and the instrument used
and compromises must be made at each experiment (Table 1).

Figure 1. Individual cells can be isolated from a population in a tissue, culture, or the environment and observed in vivo using
light/fluorescence microscopy for dynamic and functional imaging. After sample preparation (fixation and sectioning/milling),
a cell can be analyzed by different high-resolution imaging platforms in a correlative way. Electron microscopy can unveil
detailed ultrastructure of the cell while chemical imaging platforms [nano-secondary ion mass spectrometry (nanoSIMS),
X-ray fluorescence microscopy, time-of-flight (ToF)-SIMS, hybrid SIMS] enable the visualization and quantification of
elements, isotopes, and molecules at the subcellular level. Finally, image processing allows the correlation between
multimodal micrographs that contain complementary information on the cell. This workflow still requires some
methodological developments at various steps, from sample preparation to image processing, to further understand the
metabolism and physiology of a cell at the nanoscale in its close-to-native state.

176 Trends in Cell Biology, March 2020, Vol. 30 , No. 3

Trends in Cell Biology

Box 1. Principles of Subcellular Chemical Imaging Techniques

XRF relies on the photoelectric effect occurring when an X-ray photon is absorbed by an atom: a photoelectron is ejected from a core orbital of the excited atom leaving a
vacancy, which is then filled by another electron from a higher orbital. This is followed by the emission of a fluorescent X-ray photon whose energy is characteristic of the
element. XRF acquisition is often performed by scanning the sample and energy dispersive detectors collect and measure the emitted fluorescence spectrum in each
pixel, unveiling the elemental composition. The beam size determines the spatial resolution, from 10 nm to 1 μm. For a thin sample, the intensity of the fluorescence is
proportional to the concentration of the elements, which can be calculated using thin standards of known concentration (Figure I).

In SIMS, on the impact of a focused ion beam, sample material is sputtered and about 1% of ejected material is ionized. In nanoSIMS, these ions are extracted in negative
or positive modes into a mass spectrometer and separated according to their mass-to-charge ratio (m/z) in the magnetic sector of the mass analyzer. Simultaneous
detection of up to seven secondary ion species (monoatomic ions or small molecular fragments of up to four to six atoms) can be achieved. The ion beam can scan
a predefined surface area of the cell, therefore providing color-coded cartography of ion counts per pixel. Note that the charge compensation (i.e., compensating the
buildup of charge on nonconductive surfaces) is available only in negative extraction mode. Therefore, coating the sample surface with a conductive metal (~10 nm)
is mandatory to overcome this limitation in positive extraction mode.

In ToF-SIMS, the ejection and ionization of material relies on a lower ion beam fluence and less destructive cluster ion sources compared with nanoSIMS. With new
cluster ion sources (Bin, Arn, Aun, C60), ToF-SIMS provides the possibility of minimizing molecular damage while maximizing molecular ion yields. Small organic molecules
in the range of 1 to ~1000 Da can be detected with a lateral resolution of around a micron on biological material. The pulsed operation mode of the primary ion gun and its
45° mounting geometry allows charge compensation in both extraction polarities while employing the same primary ion species. However, the 45° geometry can cause
a shadowing and lateral displacement effect on depth profiling or when analyzing a surface with a pronounced topography.

Trends in Cell Biology

Figure I. Principles of Subcellular Chemical Imaging Techniques.

Correlated Subcellular Imaging towards Integrative Cell Biology

Correlation between Morphology and Chemical Imaging
Since chemical imaging provides little morphological information, combination with light microscopy/
EM is required to unambiguously elucidate the localization of elements and molecules in a cell.
However, the challenge is the trade-off between ultrastructure and chemistry preservation of the
cell during the sample preparation and the ability to transfer and analyze the same cells on different

Trends in Cell Biology, March 2020, Vol. 30 , No. 3 177

Trends in Cell Biology

Trends in Cell Biology

Figure 2. The Potential of Chemical Imaging to Unveil the Chemical Landscape of a Cell: Composition and Distribution of Elements, Isotopes, and
Molecules at the Nanoscale. (A–C) Nano-secondary ion mass spectrometry (nanoSIMS) images showing the distribution of the macronutrients nitrogen [(A)
12
C14N−], phosphorous [(B) 31P−], and sulfur [(C) 32S−] inside a microalgal cell. (D) NanoSIMS image showing the uptake of 13C incorporated in proteins (13C14N) in cells
after incubation in 13C-labeled bicarbonate [stable isotope probing (SIP)-nanoSIMS]. (E) NanoSIMS image acquired on macrophages treated with the drug iodine-
containing amiodarone. The overlay of 31P− (blue) and 127I− (purple) secondary ion maps provides morphological information (localization of the nucleus) and shows the
specific localization of the drug within the lysosomes. Reproduced, with permission, from [84]. (F) Specific labeling of proteins for correlated fluorescence microscopy
and nanoSIMS using FluorLink–nanobody anti-GFP and direct immunostaining strategies. NanoSIMS image of 19F/12C14N ratio shows the presence of the targeted
protein in specific cellular areas. Reproduced, with permission, from [65]. (G) Visualization of antibiotic in cells. Overlay of nanoSIMS and electron microscopy images
showing the accumulation of the antibiotic in lipid droplets (LDs). The bromine-containing antibiotic (bedaquiline) can be detected and semiquantified by nanoSIMS
(Figure legend continued at the bottom of the next page.)

178 Trends in Cell Biology, March 2020, Vol. 30 , No. 3

Trends in Cell Biology

Box 2. Sample Preparation for Subcellular Imaging

The sample preparation workflow depends on the sample, imaging platform, and targeted compound (e.g., molecule, element, isotope) (Figure I). Because of the high-
vacuum conditions in most EM and chemical imaging instruments, live cells cannot be directly analyzed but generally need to be fixed, dehydrated, and sectioned/milled
to access subcellular structures (Table 1). For fixation, the use of aldehydes at room temperature can alter the ultrastructure and chemistry of the cell since diffusible
mobile ions and small molecules (e.g., free amino acids, lipids) can be redistributed and/or washed out from the cell. Therefore, rapid freezing methods (high-pressure
freezing and plunge freezing) are superior in preserving native-state cell structure and chemistry at a fast rate (a scale of milliseconds compared with minutes for chemical
fixation). Then, freeze substitution is a dehydration step where the ice of the vitreous cell is slowly replaced by an organic solvent at a very low temperature (from −90°C
to −30°C) with retention of elements and metabolites possible, provided the substitution protocol is chosen carefully [13,50]. Chemical fixatives, such as osmium, tannic
acid, or aldehydes, can be used during this step to stabilize cellular structures [51]. A resin-embedding step is then necessary to obtain thin sections and access the
cell’s interior. Cryodehydration can also be performed by freeze drying, but this method can cause ultrastructural modification and elemental redistribution. However,
freeze drying is particularly powerful for drying ultrathin frozen-hydrated sections for room-temperature analysis. After sample preparation, chemical preservation of
the cell can be assessed by visualizing the most diffusible elements (K+, Na+, Ca2+, Cl−) that move rapidly across membranes and within the cytoplasm, thus
representing a relevant rule-of-thumb criterion for chemical preservation [52]. Overall, it remains difficult to be certain that the chemical environment in a cell is a true
representation of normal physiology, and each step of the preparation can be debatable. We recommend more methodological development and comparisons in
the future to optimize sample preparation and assess putative artefacts. This is a challenge since access to cutting-edge microscopes is generally difficult for that meth-
odological purpose. Compared with freeze-substituted and resin-embedded cells, analysis of frozen-hydrated cells (or vitreous cells) is obviously superior for chemical
preservation (Table 1). However, this leads to many challenges that need to be tackled in the future (further discussed in this review): (i) cryosectioning; (ii) the need for a
cryotransfer system and a cryostage in the imaging platform; (iii) difficulty in undertaking correlative studies across different platforms; and (iv) inherent lack of contrast for
sample visualization.

Trends in Cell Biology

Figure I. Sample Preparation for Subcellular Imaging.

through the 79Br ions (red signal). Reproduced, with permission, from [41]. (H) Time-of-flight (ToF)-SIMS images showing accumulation of phosphates (red; PO3−) in biofilm
of algal cells (green; CN−) growing in cotton (blue; CH4S−). Reproduced, with permission, from [47]. (I) Synchrotron X-ray fluorescence image showing the distribution of
indium phosphide-based nanocrystals in a frozen section of Hydra vulgaris. Nanocrystals detected by indium X-ray fluorescence (in red) are mainly internalized in the
ectoderm layer (ect). The natural macronutrients phosphorous (blue) and sulfur (green) provide the morphological context. Reproduced, with permission, from [21].
(J) Synchrotron X-ray fluorescence image showing the presence of silver nanowires in a fibroblast cell (S in green, Ag in red). Yellow regions indicate colocalization of
Ag and S inside the cell. Reproduced, with permission, from [4]. Copyright 2019 National Academy of Sciences. (K) NanoSIMS image showing the distribution of Ca
(40Ca+) and Fe (56Fe+) in a microalga. Calcium mapping unveils the overall morphology of the cells, with high concentration in the nucleus (nucl). Iron is mostly
contained in the plastids (pla). (L,M) Synchrotron X-ray fluorescence images showing the subcellular distribution and quantification of the trace metals Fe (red), Co
(blue), and Os (green) in microalgal cells. Pla, plastid of microalgal cell; Cyt, cytoplasm; Nucl, nucleus; LD, lipid droplet; ect, ectoderm; end, endoderm.

Trends in Cell Biology, March 2020, Vol. 30 , No. 3 179

Trends in Cell Biology

Table 1. Detailed Procedures for Sample Preparation for EM and Chemical Imaging and Considerations (Pros and Cons) for Each Step
Sample preparation Strategy Pros Cons
step
Fixation of cell Chemical fixation Easy to use in the field or for pathogens (human Ultrastructure and chemical composition can be
parasites) greatly modified
Cryofixation under high Excellent preservation of the ultrastructure and Thickness of the sample must be less than 200 μm,
pressure chemistry of the cell requires bulky laboratory-based equipment
Cryofixation with plunge Can be done in the field or the laboratory Maximum sample thickness to maintain vitreous
freezing in a liquid cryogen ice formation is ~5 μm, some ice crystal
formation in thicker samples
Dehydration Chemical dehydration at Can be done in the field or the laboratory Structural and chemical preservation are not
room temperature guaranteed
Freeze drying No use of chemicals or solvents, ideal for drying Likelihood of movement of target ions (particularly
ultrathin frozen-hydrated sections for diffusible elements) or metabolites, especially in
room-temperature analysis highly vacuolated tissues
Freeze substitution Dehydration at very low temperature allows good Use of solvents can extract materials of interest,
structural and chemical preservation long process (days to weeks)
Use of chemical Osmium Membranes are fixed and osmium provides contrast Highly toxic, interferes with some molecules and
fixative during the for EM investigation and structural information in XRF elements for XRF and ToF-SIMS analysis
freeze substitution
Aldehydes Proteins are fixed, maintaining structural preservation Toxic, no contrast for EM investigation, cannot
be prepared as anhydrous (therefore loss of any
water-soluble material)
Acrolein Crosslinks at low temperatures making it highly suited Hazardous material
for use with freeze substitution, can be anhydrous
Resin embedding Plastic epoxy resin Good structural preservation and contrast Often contains Cl, requires solvents for good
infiltration
Methacrylate resins Preservation of antigenicity, low viscosity ideal for Poor stability under an ion beam, usually require
difficult-to-embed samples O-free environment to cure
Sectioning Wet sectioning Easy to collect the sections Highly diffusible molecules can be washed out
Dry sectioning Avoidance of water/liquids allowing retainment of Difficult to cut sections thinner than 500 nm, difficult
water-soluble ions and molecules to obtain flat/uncompressed sections for analysis
Cryoanalysis Frozen-hydrated cells The best structural and chemical preservation close Lack of contrast and structural information,
to the native state, whole cells or cross sections of correlative approaches across platforms
cells can be analyzed currently difficult, cryosectioning (microtomy or
cryo-FIB) highly specialized

imaging platforms. Some imaging techniques, such as SIMS, are destructive, meaning that morpho-
logical imaging must generally take place beforehand. Here we propose various strategies that can
be adopted to analyze the same cellular region of interest with multimodal imaging (Figure 1).

Organelles can be labeled and observed with fluorescence microscopy before the entire cell is
subjected to chemical imaging. For instance, the accumulation of Mn in the Golgi apparatus of
dopaminergic cells was revealed using GFPs targeting the organelle followed by S-XRF imaging
under cryogenic conditions [53–55]. More recently, correlation between super-resolution stimu-
lated emission depletion microscopy of proteins and S-XRF imaging of trace metals was per-
formed with 40-nm spatial resolution on neurons [56].

To obtain a high-resolution cellular context, it is also possible to analyze cell sections in S/TEM
followed by S-XRF, providing an unambiguous spatial origin of elements in subcellular compart-
ments (Figures 3 and 4) [57]. Osmium tetroxide (OsO4) can be used to fix and stain cellular com-
partments (Table 1), providing morphological contrast not only on EM, but also on S-XRF, where
Os fluorescence reveals the ultrastructure of the cell (Figures 2 and 3; [12]). The drawback is that

180 Trends in Cell Biology, March 2020, Vol. 30 , No. 3

Trends in Cell Biology

Trends in Cell Biology

Figure 3. Examples of Correlated Electron Microscopy (EM) and Chemical Imaging. (A) Correlation between EM
(left images) and nano-secondary ion mass spectrometry (nanoSIMS) (middle images) showing the sulfur (32S−; upper image)
and nitrogen (12C14N−; lower image) content of a microalgal cell. Right images show the overlay of macronutrient mapping
and ultrastructure obtained from consecutive sections or the same section. Image courtesy of Charlotte Lekieffre.
(B) Correlation between scanning EM (SEM) and synchrotron X-ray fluorescence microscopy (S-XRF). SEM observation
has been performed on the same cell section after S-XRF analysis. S-XRF mapping of phosphorous and iron (green) and
sulfur (red) unveils numerous hotspots (one example highlighted by the yellow circle) where sulfur and iron are colocalized
at high concentration in the cell. (C) Transmission EM (TEM) image (left) with corresponding energy-filtered TEM (EFTEM)
Fe map (right) showing aggregated ferritin molecules in a cell. By courtesy of Jeremy Shaw and David Keays. Bar,
100 nm. (D) Correlation between SEM, time-of-flight (ToF)-SIMS, and nanoSIMS showing the cell ultrastructure and
(Figure legend continued at the bottom of the next page.)

Trends in Cell Biology, March 2020, Vol. 30 , No. 3 181

Trends in Cell Biology

Trends in Cell Biology

Figure 4. How to Visualize Metals in Cells. Flowchart to guide users when probing metals in a single cell at high
resolution and sensitivity.

distribution of sulfur molecules (CHSN−, S−, HS−) and sulfur (32S−), respectively. These multimodal images were acquired
from consecutive thin sections. (E) EM hybrid SIMS image showing the 3D distribution in a single cell of the drug
amiodarone (m/z 646; green) and biomolecules at m/z 157 (purple) and m/z 184 (green) (upper image). Lower image
obtained from the hybrid SIMS instrument shows that C24:1 sulfatides (m/z 888.62; green) are localized to the corpus
collosum. The DNA base adenine (red; m/z 134.05), a nuclear marker, shows that neurons are densely packed in the
pyramidal layer and sparsely packed in the striatum oriens, where phosphoinositol is located (m/z 241.01; blue). Adapted,
with permission, from [75]. (F) Direct correlation between X-ray phase-contrast and X-ray fluorescence tomography on a
malaria-infected cell. The 3D mass density volume is obtained after tomographic reconstruction (on the left). Subsequent
X-ray fluorescence scanning measurements were performed on the same sample showing the 3D mass concentration
volumes of iron, sulfur, and phosphorous (on the right). Reproduced, with permission, from [70].

182 Trends in Cell Biology, March 2020, Vol. 30 , No. 3

Trends in Cell Biology

the XRF emission lines of Os interfere with those of phosphorous and some trace metals (e.g., Cu,
Zn), increasing their detection limit. To obtain structural and elemental/isotopic information from a
single cellular region, it is possible to perform TEM followed by nanoSIMS (Figure 3). Usually it
would require the use of a specific TEM grid with coordinates or fiducial markers to find the
same regions of interest in the two instruments (Figures 3 and 4) [35,37,58,59]. With the recent
advent of sensitive backscatter detectors in modern SEM, it is also possible to acquire structural
information from sections using SEM before nanoSIMS and S-XRF analyses, on the same sample
or on consecutive sections [12,41,60].

Correlation between Fluorescence and NanoSIMS Using Element Labeling

The coupling between fluorescence microscopy and nanoSIMS can be a powerful approach
to unveil the functional identity of a cell and organelles. This relies on targeted probing of
DNA, RNA, and proteins by coupling fluorescent dyes to elements usually absent from the
cell’s natural composition, such as halogens (fluorine, bromine), gold, and boron. A specific
exogenous element (detectable by nanoSIMS) can be associated with/linked to fluorescent
dyes, antibodies, or nanobodies, allowing correlative microscopy between fluorescence
in situ hybridization (FISH), immunocytochemistry, or click chemistry approaches and
nanoSIMS [23,61,62]. For example, the nucleotide analog bromodeoxyuridine (BrdU), which
is incorporated into replicating DNA during cell division, can be detected by fluorescence
immunohistochemistry but also as bromine ions (e.g., 79Br, 81Br) by nanoSIMS in the same
cells [63]. Fluorine (19F) labeling of proteins using a 19F-azide probe [64] or by conjugation to
nanobodies (e.g., GFP-like proteins and antibodies) can enable correlated fluorescence
and nanoSIMS imaging [65] (Figure 2). Similarly, 197Au can be conjugated to an antibody
to recognize cellular actin and synaptophysin proteins by coupled immunofluorescence
and nanoSIMS [66], while antibodies tagged with isotopically pure elemental metal reporters
(i.e., lanthanides) have been utilized to image protein expression in human breast tumor tissue
sections [67]. Recently, boron linked to proteins or to nanobodies binding to proteins has been
used for simultaneous protein identification and elemental mapping by correlative fluorescence
microscopy and nanoSIMS [68]. These studies suggest that exogenous elements of small
mass and size are suitable for probing DNA and proteins in the complex cellular environment
and can be used in correlated nanoSIMS studies.

Correlation between More Than Two Subcellular Imaging Platforms

The combination of multiple subcellular imaging platforms can provide a comprehensive view of
the ultrastructure, concentration, and distribution of elements (macronutrients and metals) and
isotopic ratios, and molecules from a single region of interest in the cell (Figure 3). This can be per-
formed by analyzing different consecutive resin sections from the same cell on different platforms
[12,13,57]. This flexible strategy allows one to choose the required thickness and support
(e.g., wafer, grid, Si window) for each section, and detailed ultrastructure can be obtained with
EM. Thus, the morphological and various chemical features can be superimposed. However,
collecting serial sections of different thicknesses to target the same cellular region or organelle is
a challenge.

Integrated Correlative Instruments

A correlative approach can be facilitated when morphological and chemical information from exactly
the same area can be acquired in the same instrument. One of the best examples is TEM-EDS and
EFTEM. TEM offers the highest possible resolution for both imaging and element analysis. Provided
that samples are thin (usually ~150 nm at 200 kV) and the elements of interest can be preserved,
simultaneous structural and elemental information can be obtained at the nanoscale (Figure 3)
[69]. However, molecular information is not available and absolute quantitation can be difficult.

Trends in Cell Biology, March 2020, Vol. 30, No. 3 183

Trends in Cell Biology

Using synchrotron X-ray nanoprobes, the combination of X-ray phase-contrast tomography and
XRF microscopy can provide the morphological information and quantification of elements,
respectively (Figure 3). This has been recently performed on a freeze-dried human phagocytic cell
[16] and human red blood cells infected with the malaria parasite Plasmodium falciparum [70]. An
alternative to phase-contrast imaging is ptychography [71], which can be combined with XRF
tomography to obtain the 3D localization of elements in cells, such as bacteria in [72].

New instruments that offer simultaneous morphological and molecular information are emerging
but are yet to be applied to biological specimens. For instance, SIMS in helium-ion microscopy
(HIM-SIMS) can combine high-resolution morphological images with elemental and isotopic
maps from SIMS [73]. In contrast to EDS on SEM, HIM-SIMS provides better detection limits
for elements (including the very light ones) and differentiation between isotopes. However,
using HIM-SIMS for stable-isotope labeling experiments will require a significant improvement
of the mass resolution. Overall, the combination of high-resolution secondary electron images
and mass-separated sputtered ion distributions has high potential to answer open questions in
cell biology.

Image Processing for Multimodal Correlative Imaging

On their acquisition on different imaging platforms, micrographs need to be processed to corre-
late or even overlap information from the same cellular region. Multimodal 2D microscopy requires
registration algorithms that can analyze the same region but at different fields of view and resolu-
tions from images acquired with different excitation probes (e.g., photons, electrons, ions) as well
as different detectors. Compared with monomodal image processing, more sophisticated
approaches are needed for multimodal images since the pixel intensities of features are not com-
parable or do not even occur in both images. The additional challenge in multimodal data sets is
that the shape of the object might differ because of preparation steps or the different probing
depths of the microscopes. Distortions may also be introduced when subsequent sections,
representing different depth layers of the sample, are used for correlative imaging. In this context,
the ImageJ-based software Correlia has recently been developed for the registration of 2D–2D
multimodal microscopy data sets (available on request).

Perspectives and Future Challenges for Multimodal Subcellular Imaging

Subcellular Mapping of Metabolites in Cells
Visualization of the compartmentalized distribution of metabolites in cells is one of the most
promising research avenues in biology. Recently, laser-based ionization mass spectrometry
techniques have been successfully used in single-cell metabolomics profiling experiments,
but the lateral resolution precludes imaging of subcellular chemistry [74]. Mapping metabolites
at subcellular level can be now envisioned with the revolutionary hybrid SIMS instrument (3D
OrbiSIMS, IONTOF) that combines the speed and high-lateral-resolution imaging capabilities of
ToF-SIMS with the unprecedented mass resolution (240K), mass accuracy (sub-ppm), dynamic
range (S/N ratio ~ 105), and MS/MS capabilities of the Orbitrap Q-Exactive [75]. The instrument
has been used to map the distribution of lipids and neurotransmitters in the hippocampal region
of the mouse brain at the cellular and subcellular level. The 3D imaging capabilities were used to
visualize the accumulation of amiodarone in single lung macrophage cells and to assess its toxic
effect by correlating drug concentration with the levels of phospholipids and cholesterol [75].

Future Improvements Needed to Probe Cells in Their Native State and in Four Dimensions
The future of chemical imaging largely relies in the development of cryoanalyses and associated
correlative workflows to allow multiscale chemical analysis of cells in their native state across mul-
tiple instruments. Currently, cryoanalyses present limitations with regard to the suitability of

184 Trends in Cell Biology, March 2020, Vol. 30 , No. 3

Trends in Cell Biology

samples for cryopreservation and their transfer between different platforms without ice contami- Outstanding Questions
nation. In most cases, frozen cells must be sectioned/milled to visualize the interior. Future expan- • Is it possible to preserve the native
sion in this area will require improvements in the preparation of quality cryosections (e.g., cryo- physiological state of a cell from
sample preparation to imaging?
FIB-SEM), the development of cryoenabled instrument platforms (e.g., nanoSIMS), and the ability
How can structural and chemical
to transfer samples seamlessly between these. Additional challenges associated with preservation of the cell be carefully
cryoimaging are the low contrast of vitrified cells and possible devitrification under irradiation assessed and artifacts identified to
from the probe (e.g., ions, X-rays). avoid misleading results?
• Can we resolve the chemical landscape
(composition and distribution of
Chemical information in 2D is insufficient to fully describe the compartmentalization of an element/
elements and molecules) of a cell
molecule throughout an entire cell, especially when information comes from a single thin section with high spatial resolution, in three
(60–300 nm thickness) of a cell. Techniques to acquire 3D structural information at the subcellular dimensions, and over a relevant
level are readily available (e.g., FIB-SEM) but it is difficult to couple these with analytical informa- temporal scale? What would be the
analytical workflow to implement for
tion. Synchrotron-based coherent X-ray scattering techniques such as holotomography or 4D subcellular imaging?
ptychography have both demonstrated constant improvements towards the ultrastructural 3D • What are the methodological and
characterization of a cell architecture [76]. We therefore foresee a bright future in the coupling technological strategies to analyze
between these techniques and XRF tomography for 3D elemental imaging. the same subcellular region of
interest across different subcellular
imaging techniques?
The downside of high-resolution chemical imaging is the low throughput, which precludes
• How can we include cryoanalysis in the
robust statistical analyses. We can expect that ultrahigh-speed scanning strategy and highly workflow of correlative subcellular
efficient detection will be the next steps to foster on the instrumentation side. For example, imaging, from sample preparation and
extremely brilliant X-ray synchrotron sources and high-rate scanning strategies have become obtaining vitreous cells to transfer and
imaging across different platforms?
reality with the upgrade of synchrotrons such as the European Synchrotron Radiation Facility Can it be applied to different types of
(ESRF) [77]. cells from a tissue or isolated in culture
or the environment?
Finally, future developments need to integrate a biologically relevant temporal dimension into the • How can the throughput of chemical
imaging techniques be increased to
correlative imaging workflow towards 4D imaging. Because probing elements and molecules of
observe large numbers of biological
live cells at the nanoscale level remain challenging, there is a need to capture multiple snapshots samples and perform statistical
of the phenotypic state of a cell over time to follow and better understand dynamic cellular pro- comparisons?
cesses following exposure to abiotic or biotic stress. A single snapshot may provide a biased • Can we enhance the mass resolution
vision of the phenotypic response of a cell if imaging analysis occurs on an inappropriate timescale. and spatial resolution of SIMS imaging
instruments to visualize a large number
Temporal resolution can be obtained by coupling with live imaging (fluorescence and super- of different metabolites in the cell?
resolution microscopy) and the development of microfluidic devices [78] or rapid cryofixation
strategies [79]. For instance, dynamic fluorescence light microscopy can be rapidly followed by
cryoimmobilization in few seconds for CLEM studies thanks to new tools (e.g., CryoCapsule)
and could be extended to chemical imaging in the future [80]. In addition, recent advances in
super-resolution fluorescence microscopy [e.g., stimulated emission depletion (STED) microscopy]
[81] and the future development of new fluorescent, element-specific probes will potentially provide
a dynamic view of labile elements (e.g., Ca2+, Fe2+/3+, Zn2+) in live cells at 10–50 nm resolution [82].
Correlation with chemical imaging will open new perspectives bridging temporal and spatial
resolution [83].

Thus, the development of time-resolved 3D subcellular imaging under cryoconditions will be a

breakthrough in cell biology that will capture dynamic subcellular processes in a native-state cell.

Concluding Remarks
Subcellular chemical imaging techniques are constantly improving and becoming ever-more-
powerful tools for quantitative visualization of elements, isotopes, and molecules in cells. How-
ever, untangling their complex requirements and capabilities is a vital step in ensuring that
researchers can apply such methods to outstanding research questions and problems (see
Outstanding Questions). With this, appropriate sample preparation and suitable imaging

Trends in Cell Biology, March 2020, Vol. 30 , No. 3 185

Trends in Cell Biology

platforms need to be selected according to the sample, spatial resolution, and targeted ele-
ments/molecules. In this review, we have outlined the principles of key analytical instrumentation,
discussed strategies for sample preparation, and highlighted the potential for correlative EM and
chemical imaging to accumulate structural and chemical information from a single region of a cell.
Correlated morphological and chemical imaging has the potential to spur a rapid expansion in
various fields, such as cell biology, biomedicine, ecophysiology, pharmacology, toxicology, and
biogeochemistry. In the near future, we do not foresee a technique that would encompass all of
the capabilities to explore the chemical/molecular/isotopic composition and ultrastructure of a
cell. Thus, the development of integrative studies and dedicated analytical correlative workflows,
from sample preparation to multimodal imaging and image processing, will be a major contribution
towards a full comprehension of the physiology of the cell at the subcellular level.

Acknowledgments
J.D. was supported by the LabEx GRAL (ANR-10-LABX-49-01), financed within the University Grenoble Alpes graduate
school (Ecoles Universitaires de Recherche) CBH-EUR-GS (ANR-17-EURE-0003), and Défi X-Life grant from CNRS. The
authors acknowledge the support and use of resources from Instruct (a Landmark ESFRI project) and the ESRF for pro-
viding beamtime. We are thankful for the use of the analytical facilities of the Centre for Chemical Microscopy (ProVIS) at
UFZ Leipzig, which is supported by European Regional Development Funds (EFRE – Europe Funds Saxony) and the Helmholtz
Association. This research is also supported by EMBRC-France, whose French state funds are managed by the ANR within the
Investments of the Future program under reference ANR-10-INBS-02. We also thank Yannick Schwab for critically reading the
manuscript and suggesting improvements. We are grateful to colleagues for sharing microscopy images.

References
1. Salt, D.E. et al. (2008) Ionomics and the study of the plant
ionome. Annu. Rev. Plant Biol. 59, 709–733
2. Hackett, M.J. et al. (2019) Elemental characterisation of the py-
ramidal neuron layer within the rat and mouse hippocampus.
Metallomics 11, 151–165
3. Veronesi, G. et al. (2016) Visualization, quantification and coordi-
nation of Ag+ ions released from silver nanoparticles in hepato-
cytes. Nanoscale 8, 17012–17021
4. Lehmann, S.G. et al. (2019) Crumpling of silver nanowires by
endolysosomes strongly reduces toxicity. Proc. Natl. Acad.
Sci. U. S. A. 116, 14893–14898
5. de Boer, P. et al. (2015) Correlated light and electron microscopy:
ultrastructure lights up! Nat. Methods 12, 503–513
6. Karreman, M.A. et al. (2016) Intravital correlative microscopy:
imaging life at the nanoscale. Trends Cell Biol. 26, 848–863
7. da Cunha, M.M.L. et al. (2016) Overview of chemical imaging
methods to address biological questions. Micron 84, 23–36
8. Baker, T.C. et al. (2017) Recent advancements in matrix-assisted
laser desorption/ionization mass spectrometry imaging. Curr.
Opin. Biotechnol. 43, 62–69
9. Lauwers, M. et al. (2013) An iron-rich organelle in the cuticular
plate of avian hair cells. Curr. Biol. 23, 924–929
10. Cotte, M. et al. (2017) The ID21 X-ray and infrared microscopy
beamline at the ESRF: status and recent applications to artistic
materials. J. Anal. At. Spectrom. 32, 477–493
11. Pushie, M.J. et al. (2014) Elemental and chemically specific X-ray
fluorescence imaging of biological systems. Chem. Rev. 114,
8499–8541
12. Decelle, J. et al. (2019) Algal remodeling in a ubiquitous planktonic
photosymbiosis. Curr. Biol. 29, 968–978.e4
13. Kashiv, Y. et al. (2016) Imaging trace element distributions in single
organelles and subcellular features. Sci. Rep. 6, 21437
14. Adams, M.S. et al. (2016) Copper uptake, intracellular localization,
and speciation in marine microalgae measured by synchrotron ra-
diation X- ray fluorescence and absorption microspectroscopy.
Environ. Sci. Technol. 50, 8827–8839
15. Ciccotosto, G.D. et al. (2014) Quantitation and localization of
intracellular redox active metals by X-ray fluorescence micros-
copy in cortical neurons derived from APP and APLP2 knockout
tissue. Metallomics 6, 1894–1904
16. Gramaccioni, C. et al. (2018) Nanoscale quantification of intra-
cellular element concentration by X-ray fluorescence microscopy
combined with X-ray phase contrast nanotomography. Appl.
Phys. Lett. 112, 053701
17. Brown, K. et al. (2018) Intracellular in situ labeling of TiO2 nano-
particles for fluorescence microscopy detection. Nano Res. 11,
464–476
18. Fus, F. et al. (2019) Intracellular localization of an osmocenyl-
tamoxifen derivative in breast cancer cells revealed by synchro-
tron radiation X-ray fluorescence nanoimaging. Angew. Chem.
Int. Ed. Engl. 58, 3461–3465
19. Kapishnikov, S. et al. (2019) Mode of action of quinoline antima-
larial drugs in red blood cells infected by Plasmodium falciparum
revealed in vivo. Proc. Natl. Acad. Sci. U. S. A. 116,
22946–22952
20. Sanchez-Cano, C. et al. (2019) Nanofocused synchrotron X-ray
absorption studies of the intracellular redox state of an organo-
metallic complex in cancer cells. Chem. Commun. 55,
7065–7068
21. Veronesi, G. et al. (2019) In vivo biotransformations of indium
phosphide quantum dots revealed by X-ray microspectroscopy.
ACS Appl. Mater. Interfaces 11, 35630–35640
22. Hackett, M.J. et al. (2016) Chemical biology in the embryo: in situ
imaging of sulfur biochemistry in normal and proteoglycan-
deficient cartilage matrix. Biochemistry 55, 2441–2451
23. Gyngard, F. and Steinhauser, M.L. (2019) Biological explorations
with nanoscale secondary ion mass spectrometry. J. Anal. At.
Spectrom. 34, 1534–1545
24. Agüi-Gonzalez, P. et al. (2019) SIMS imaging in neurobiology
and cell biology. J. Anal. At. Spectrom. 34, 1355–1368
25. Malherbe, J. et al. (2016) A new RF plasma oxygen primary ion
source on nanoSIMS for improved lateral resolution and detec-
tion of electropositive elements at single cell level. Anal. Chem.
88, 7130–7136
26. Nuñez, J. et al. (2018) NanoSIMS for biological applications:
current practices and analyses. Biointerphases 13, 03B301
27. Weng, N. et al. (2017) In situ subcellular imaging of copper
and zinc in contaminated oysters revealed by nanoscale
secondary ion mass spectrometry. Environ. Sci. Technol.
51, 14426–14435
28. Tsednee, M. et al. (2019) Manganese co-localizes with calcium
and phosphorus in Chlamydomonas acidocalcisomes and is
mobilized in Mn-deficient conditions. J. Biol. Chem. 294,
17626–17641

186 Trends in Cell Biology, March 2020, Vol. 30 , No. 3

Trends in Cell Biology
29. Mayali, X. (2020) NanoSIMS: microscale quantification of
biogeochemical activity with large-scale impacts. Annu. Rev.
Mar. Sci. 12, 449–467
30. Musat, N. et al. (2016) Tracking microbial interactions with
nanoSIMS. Curr. Opin. Biotechnol. 41, 114–121
31. Stryhanyuk, H. et al. (2018) Calculation of single cell assimilation
rates from SIP-nanoSIMS-derived isotope ratios: a comprehen-
sive approach. Front. Microbiol. 9, 2342
32. Berthelot, H. et al. (2018) NanoSIMS single cell analyses reveal
the contrasting nitrogen sources for small phytoplankton. ISME
J. 13, 651–662
33. Terrado, R. et al. (2017) Autotrophic and heterotrophic acquisi-
tion of carbon and nitrogen by a mixotrophic chrysophyte
established through stable isotope analysis. ISME J. 11,
2022–2034
34. He, C. et al. (2018) NanoSIMS imaging reveals unexpected
heterogeneity in nutrient uptake by brown adipocytes. Biochem.
Biophys. Res. Commun. 504, 899–902
35. Lekieffre, C. et al. (2018) Inorganic carbon and nitrogen assimilation
in cellular compartments of a benthic kleptoplastic foraminifer. Sci.
Rep. 8, 10140
36. Volland, J.M. et al. (2018) NanoSIMS and tissue autoradiography
reveal symbiont carbon fixation and organic carbon transfer to
giant ciliate host. ISME J. 12, 714–727
37. Gibbin, E. et al. (2019) Vibrio coralliilyticus infection triggers a be-
havioural response and perturbs nutritional exchange and tissue
integrity in a symbiotic coral. ISME J. 13, 989–1003
38. Martínez-Pérez, C. et al. (2016) The small unicellular diazotrophic
symbiont, UCYN-A, is a key player in the marine nitrogen cycle.
Nat. Microbiol. 1, 16163
39. Pasulka, A.L. et al. (2018) Interrogating marine virus–host inter-
actions and elemental transfer with BONCAT and nanoSIMS-
based methods. Environ. Microbiol. 20, 671–692
40. Worrich, A. et al. (2017) Mycelium-mediated transfer of water
and nutrients stimulates bacterial activity in dry and oligotrophic
environments. Nat. Commun. 8, 15472
41. Greenwood, D.J. et al. (2019) Subcellular antibiotic visualization re-
veals a dynamic drug reservoir in infected macrophages. Science
364, 1279–1282
42. Vanbellingen, Q.P. et al. (2015) Time-of-flight secondary ion
mass spectrometry imaging of biological samples with delayed
extraction for high mass and high spatial resolutions. Rapid
Commun. Mass Spectrom. 29, 1187–1195
43. Benettoni, P. et al. (2019) Identification of nanoparticles and their
localization in algal biofilm by 3D-imaging secondary ion mass
spectrometry. J. Anal. At. Spectrom. 34, 1098–1108
44. Henss, A. et al. (2018) High resolution imaging and 3D analysis
of Ag nanoparticles in cells with ToF-SIMS and delayed extrac-
tion. Biointerphases 13, 03B410
45. Raina, J.-B. et al. (2017) Subcellular tracking reveals the location
of dimethylsulfoniopropionate in microalgae and visualises its up-
take by marine bacteria. Elife 6, e23008
46. Dowlatshahi Pour, M. et al. (2019) Mass spectrometry imaging
as a novel approach to measure hippocampal zinc. J. Anal. At.
Spectrom. 34, 1581–1587
47. Osorio, J.H.M. et al. (2019) Investigation of architecture
development and phosphate distribution in Chlorella biofilm
by complementary microscopy techniques. FEMS Microbiol.
Ecol. 95, fiz029
48. Sämfors, S. et al. (2019) Localised lipid accumulation detected in
infarcted mouse heart tissue using ToF-SIMS. Int. J. Mass
Spectrom. 437, 77–86
49. Passarelli, M.K. et al. (2013) Single-cell lipidomics: characterizing
and imaging lipids on the surface of individual Aplysia californica
neurons with cluster secondary ion mass spectrometry. Anal.
Chem. 85, 2231–2238
50. Moore, K.L. et al. (2014) Combined nanoSIMS and synchrotron
X-ray fluorescence reveal distinct cellular and subcellular
distribution patterns of trace elements in rice tissues. New
Phytol. 201, 104–115
51. Zhou, J. et al. (2011) Reproducibility and quantitation
of amplicon sequencing-based detection. ISME J. 5,
1303–1313
52. Perrin, L. et al. (2015) Evaluation of sample preparation methods
for single cell quantitative elemental imaging using proton or
synchrotron radiation focused beams. J. Anal. At. Spectrom.
30, 2525–2532
53. Das, S. et al. (2019) Manganese mapping using a fluorescent
Mn2+ sensor and nanosynchrotron X-ray fluorescence reveals
the role of the Golgi apparatus as a manganese storage site.
Inorg. Chem. 58, 13724–13732
54. Carmona, A. et al. (2019) Mapping chemical elements and iron
oxidation states in the substantia nigra of 6-hydroxydopamine le-
sioned rats using correlative immunohistochemistry with proton
and synchrotron micro-analysis. Front. Neurosci. 13, 1014
55. Carmona, A. et al. (2014) Environmental manganese com-
pounds accumulate as Mn(II) within the Golgi apparatus of
dopamine cells: relationship between speciation, subcellular dis-
tribution, and cytotoxicity. Metallomics 6, 822
56. Domart, F. et al. (2019) Correlating STED and synchrotron XRF
nano-imaging unveils the co-segregation of metals and cytoskele-
ton proteins in dendrites. bioRxiv Published online October 21,
2019. https://www.biorxiv.org/content/10.1101/810754v1
57. Sanchez-Cano, C. et al. (2017) Synchrotron X-ray fluorescence
nanoprobe reveals target sites for organo-osmium complex in
human ovarian cancer cells. Chemistry 23, 2512–2516
58. Clode, P.L. et al. (2009) In situ mapping of nutrient uptake in the
rhizosphere using nanoscale secondary ion mass spectrometry.
Plant Physiol. 151, 1751–1757
59. Kopp, C. et al. (2015) Subcellular investigation of photosynthesis-
driven carbon assimilation in the symbiotic reef coral Pocillopora
damicornis. MBio 6, e02299-14
60. Hu, X. et al. (2019) Release of cholesterol-rich particles from the
macrophage plasma membrane during movement of filopodia
and lamellipodia. Elife 8, e50231
61. Musat, N. et al. (2012) Detecting metabolic activities in single
cells, with emphasis on nanoSIMS. FEMS Microbiol. Rev. 36,
486–511
62. Saka, S.K. et al. (2014) Correlated optical and isotopic
nanoscopy. Nat. Commun. 5, 3664
63. Lau, K.H. et al. (2010) Development of a new bimodal imaging
methodology: a combination of fluorescence microscopy and
high-resolution secondary ion mass spectrometry. J. Microsc.
240, 21–31
64. Vreja, I.C. et al. (2015) Secondary-ion mass spectrometry of
genetically encoded targets. Angew. Chem. Int. Ed. Engl. 54,
5784–5788
65. Kabatas, S. et al. (2019) Fluorinated nanobodies for targeted
molecular imaging of biological samples using nanoscale
secondary ion mass spectrometry. J. Anal. At. Spectrom. 34,
1083–1087
66. Thiery-Lavenant, G. et al. (2014) Detection of immunolabels with
multi-isotope imaging mass spectrometry. Surf. Interface Anal.
46, 147–149
67. Angelo, M. et al. (2014) Multiplexed ion beam imaging of human
breast tumors. Nat. Med. 20, 436–442
68. Kabatas, S. et al. (2019) Boron-containing probes for non-
optical high-resolution imaging of biological samples. Angew.
Chem. Int. Ed. Engl. 58, 3438–3443
69. Ismagulova, T. et al. (2018) A new simple method for quantifica-
tion and locating P and N reserves in microalgal cells based on
energy-filtered transmission electron microscopy (EFTEM) ele-
mental maps. PLoS One 13, e0208830
70. Yang, Y. et al. (2019) Three-dimensional correlative imaging of a
malaria-infected cell with a hard X-ray nanoprobe. Anal. Chem.
91, 6549–6554
71. Dierolf, M. et al. (2010) Ptychographic X-ray computed tomogra-
phy at the nanoscale. Nature 467, 436–439
72. Victor, T.W. et al. (2018) X-ray fluorescence nanotomography
of single bacteria with a sub-15 nm beam. Sci. Rep. 8,
13415
73. Wirtz, T. et al. (2019) Imaging and analytics on the helium ion
microscope. Annu. Rev. Anal. Chemi. 12, 523–543
74. Baumeister, T.U.H. et al. (2019) Live single-cell metabolo-
mics with matrix-free laser/desorption ionization mass spec-
trometry to address microalgal physiology. Front. Plant Sci.
10, 172
75. Passarelli, M.K. et al. (2017) The 3D OrbiSIMS – label-free meta-
bolic imaging with subcellular lateral resolution and high mass-
resolving power. Nat. Methods 14, 1175–1183
Trends in Cell Biology, March 2020, Vol. 30 , No. 3 187
Trends in Cell Biology
76. Deng, J. et al. (2018) Correlative 3D X-ray fluorescence and
ptychographic tomography of frozen-hydrated green algae.
Sci. Adv. 4, eaau4548
77. Deng, J. et al. (2019) The Velociprobe: an ultrafast hard X-ray
nanoprobe for high-resolution ptychographic imaging. Rev. Sci.
Instrum. 90
78. Gibbin, E. et al. (2018) Using nanoSIMS coupled with
microfluidics to visualize the early stages of coral infection by
Vibrio coralliilyticus. BMC Microbiol. 18, 39
79. Kontziampasis, D. et al. (2019) A cryo-EM grid preparation
device for time-resolved structural studies. IUCrJ 6, 1024–1031
80. Heiligenstein, X. et al. (2014) The CryoCapsule: simplifying
correlative light to electron microscopy. Traffic 15, 700–716
188 Trends in Cell Biology, March 2020, Vol. 30 , No. 3
81. Sigal, Y.M. et al. (2018) Visualizing and discovering cellular struc-
tures with super-resolution microscopy. Science 361, 880–887
82. Ackerman, C.M. et al. (2017) Analytical methods for imaging
metals in biology: from transition metal metabolism to transition
metal signaling. Anal. Chem. 89, 22–41
83. Bernhardt, M. et al. (2018) Correlative microscopy approach for
biology using X-ray holography, X-ray scanning diffraction and
STED microscopy. Nat. Commun. 9, 3641
84. Jiang, H. et al. (2017) High-resolution sub-cellular imaging
by correlative nanoSIMS and electron microscopy of amioda-
rone internalisation by lung macrophages as evidence
for drug-induced phospholipidosis. Chem. Commun. 53,
1506–1509
Cell Metabolism

Review

GPIHBP1 and Lipoprotein Lipase,

Partners in Plasma Triglyceride Metabolism
Stephen G. Young,1,2,* Loren G. Fong,1,* Anne P. Beigneux,1 Christopher M. Allan,1 Cuiwen He,1 Haibo Jiang,1,3
Katsuyuki Nakajima,4 Muthuraman Meiyappan,5 Gabriel Birrane,6 and Michael Ploug7,8,*
1Department of Medicine, David Geffen School of Medicine, University of California, Los Angeles, Los Angeles, CA 90095, USA
2Department of Human Genetics, David Geffen School of Medicine, University of California, Los Angeles, Los Angeles, CA 90095, USA
3School of Molecular Sciences, University of Western Australia, Crawley 6009, Australia
4Department of Clinical Laboratory Medicine, Gunma University Graduate School of Department of Medicine, Maebashi,

Gunma 371-0805, Japan

5Discovery Therapeutics, Takeda Pharmaceutical Company Ltd., Cambridge, MA 02142, USA
6Division of Experimental Medicine, Beth Israel Deaconess Medical Center, Boston, MA 02215, USA
7Finsen Laboratory, Rigshospitalet, Copenhagen DK–2200, Denmark
8Biotech Research and Innovation Centre (BRIC), University of Copenhagen, Copenhagen DK–2200, Denmark

*Correspondence: sgyoung@mednet.ucla.edu (S.G.Y.), lfong@mednet.ucla.edu (L.G.F.), m-ploug@finsenlab.dk (M.P.)

https://doi.org/10.1016/j.cmet.2019.05.023

Lipoprotein lipase (LPL), identified in the 1950s, has been studied intensively by biochemists, physiologists,
and clinical investigators. These efforts uncovered a central role for LPL in plasma triglyceride metabolism
and identified LPL mutations as a cause of hypertriglyceridemia. By the 1990s, with an outline for plasma
triglyceride metabolism established, interest in triglyceride metabolism waned. In recent years, however, in-
terest in plasma triglyceride metabolism has awakened, in part because of the discovery of new molecules
governing triglyceride metabolism. One such protein—and the focus of this review—is GPIHBP1, a protein of
capillary endothelial cells. GPIHBP1 is LPL’s essential partner: it binds LPL and transports it to the capillary
lumen; it is essential for lipoprotein margination along capillaries, allowing lipolysis to proceed; and it pre-
serves LPL’s structure and activity. Recently, GPIHBP1 was the key to solving the structure of LPL. These
developments have transformed the models for intravascular triglyceride metabolism.

The identification of lipoprotein lipase (LPL) as an intravascular
triglyceride hydrolase (Korn, 1955; Korn and Quigley, 1955)
was a momentous discovery, one that suggested a mechanism
by which triglycerides in the bloodstream are converted to fatty
acids for utilization by vital tissues (Havel, 2010). This discovery
triggered interest in triglyceride metabolism by scientists around
the globe. Soon thereafter, a deficiency of LPL was shown to
cause familial chylomicronemia (Havel and Gordon, 1960), the
first example of an inborn error in lipoprotein metabolism (Havel,
2010). LPL-mediated processing of triglyceride-rich lipoproteins
was shown to be important for utilization of dietary lipids, for
generating atherogenic remnant lipoproteins, and for contrib-
uting to the biogenesis of high-density lipoproteins (HDLs)
(Havel, 2010; Havel and Kane, 2001; Rader, 2006; Rye and
Barter, 2014). The observation that LPL could be released into
the bloodstream with an injection of heparin (Korn, 1955), com-
bined with the fact that LPL binds to heparan sulfate proteogly-
cans (HSPGs) in vitro (Lookene et al., 1997b; Simionescu and
Simionescu, 1986; Vijayagopal et al., 1980), led to the proposal
that the LPL inside blood vessels is bound to HSPGs within the
glycocalyx lining of blood vessels (Blanchette-Mackie et al.,
1989; Olivecrona and Bengtsson-Olivecrona, 1993). LPL was
shown to be activated by apolipoprotein (apo) CII (Bengtsson
and Olivecrona, 1979; Breckenridge et al., 1978; Havel et al.,
1970; LaRosa et al., 1970), and several lines of evidence sug-
gested that LPL is active as a homodimer (Garfinkel et al.,
1983; Iverius and Ostlund-Lindqvist, 1976; Kobayashi et al.,
2002; Lookene and Bengtsson-Olivecrona, 1993; Lookene
et al., 1994; Olivecrona and Bengtsson-Olivecrona, 1990; Olive-
crona et al., 1985; Osborne et al., 1985; Peterson et al., 2002;
Vannier and Ailhaud, 1989; Wong et al., 1994, 1997). The cloning
of the LPL cDNA in the 1980s (Enerba €ck et al., 1987; Kirchgess-
ner et al., 1987; Wion et al., 1987) triggered studies on LPL
sequences required for substrate binding, catalytic activity,
and heparin binding (Emmerich et al., 1992; Lookene et al.,
1996, 1997a, 1997b; Ma et al., 1994; Sendak et al., 1998), and
also opened the door to understanding LPL mutations underly-
ing the familial chylomicronemia syndrome (Mailly et al., 1997;
Peterson et al., 2002). Together, these efforts created an outline
for intravascular lipoprotein processing—one that was taught to
a generation of students and practicing physicians. The outline
changed little over several decades, contributing to a percep-
tion, occasionally voiced at conferences during the 1990s, that
the field of plasma triglyceride metabolism was approaching
maturity.
For the scientists who were actively working on plasma triglyc-
eride metabolism, any suggestion that the field was close to
maturity would have made little sense. After all, no one under-
stood how LPL, an enzyme that is secreted by adipocytes and
myocytes, reached its site of action in the capillary lumen; the
regulation of LPL activity was not understood; the structure of
LPL remained elusive; and many cases of chylomicronemia,
both genetic and acquired, were unexplained. Also, some fea-
tures within the outline of triglyceride metabolism needed more
investigation. For example, it was unclear why the LPL secreted
by parenchymal cells would end up being attached to HSPGs in
the lumen of blood vessels, given that HSPGs are abundant
within the interstitial spaces around parenchymal cells. Also,

Cell Metabolism 30, July 2, 2019 ª 2019 Elsevier Inc. 51

 Cell Metabolism

Review

Figure 1. A Cartoon Representation of GPIHBP1’s Functions in Lipoprotein Lipase (LPL)-Mediated Lipoprotein Processing
GPIHBP1 serves four important functions in plasma triglyceride processing. The first two are illustrated in the left panel. First, GPIHBP1 on the abluminal plasma
membrane of capillary endothelial cells captures LPL (yellow) from heparin-sulfate proteoglycan (HSPG)-binding sites (green, tree-like structures) within the
subendothelial spaces (Davies et al., 2010; Kristensen et al., 2018). GPIHBP1’s intrinsically disordered acidic domain (green), which projects from GPIHBP1’s
three-fingered LU domain (blue), is important for capturing LPL and bringing it into high-affinity interactions with GPIHBP1’s LU domain (Kristensen et al., 2018).
Second, GPIHBP1 stabilizes the structure of LPL and protects it from unfolding, even in the face of physiologic inhibitor proteins such as ANGPTL4 (Kristensen
et al., 2018; Mysling et al., 2016a, 2016b). The preservation of LPL structure is illustrated by changing the color of LPL from yellow to purple. The middle panel
depicts the third function of GPIHBP1 in LPL biology, which is to transport LPL to its site of action in the capillary lumen (Davies et al., 2010). The movement of
GPIHBP1 across endothelial cells is bidirectional (Davies et al., 2012); thus, GPIHBP1 is able to return to the abluminal surface of endothelial cells, pick up more
LPL, and then replenish LPL stores along the capillary lumen. The right panel depicts the fourth function of GPIHBP1, which is to mediate the margination of
triglyceride-rich lipoproteins (TRLs) along the luminal surface of capillaries. GPIHBP1-bound LPL is required for TRL margination (Goulbourne et al., 2014),
allowing triglyceride hydrolysis to proceed.

the idea that LPL is only active as a homodimer needed more
study. Finally, the mechanisms by which specific apolipopro-
teins affected the efficiency of triglyceride hydrolysis were
incompletely understood.
Over the past decade, there has been considerable progress
in understanding plasma triglyceride metabolism. Much of that
progress has centered around GPIHBP1, a GPI-anchored pro-
tein of capillary endothelial cells (Beigneux et al., 2007).
GPIHBP1 is a dedicated partner for LPL: it captures LPL from
within the subendothelial spaces and shuttles it to the capillary
lumen (Davies et al., 2012, 2010); it is essential for the margin-
ation of lipoproteins along capillaries, allowing LPL-mediated tri-
glyceride processing to proceed (Goulbourne et al., 2014); and it
stabilizes the structure and catalytic activity of LPL (Mysling
et al., 2016a, 2016b). Recently, GPIHBP1 proved to be crucial
for solving the structure of LPL (Birrane et al., 2019; Horton,
2019; Arora et al., 2019). Here, we review recent insights into
the functions of GPIHBP1 and LPL in intravascular lipolysis,
including insights derived from the structures of those proteins.
GPIHBP1 and Plasma Triglyceride Metabolism
GPIHBP1 was initially identified during a screen of a mouse liver
cDNA library for clones that render CHO cells capable of binding
HDL (Ioka et al., 2003). The screen uncovered SR-B1, long
recognized as an HDL-binding protein (Acton et al., 1996), and
GPIHBP1, a GPI-anchored Ly6/uPAR (LU) protein with a long
stretch of negatively charged amino acids at its amino terminus
(Ioka et al., 2003). To this day, it is unclear why GPIHBP1 ap-
peared in that screen, given that Gin and colleagues found little
evidence of HDL, apo-AI, or apo-E binding to GPIHBP1-ex-
pressing cells (Gin et al., 2011). Our best guess is that the HDL
used in the initial screen contained small amounts of LPL (which
binds to GPIHBP1 avidly) or was enriched in apo-AV (a ‘‘heparin-
binding’’ apolipoprotein that has been reported to bind to the
negatively charged domain at the amino terminus of GPIHBP1)
(Gin et al., 2008, 2011).
The first clue that GPIHBP1 was important for plasma triglyc-
eride metabolism was finding severe hypertriglyceridemia in
Gpihbp1 knockout mice (Gpihbp1 ⁄ ) (Beigneux et al., 2007;
Weinstein et al., 2010a, 2010b). That discovery, combined with
the finding that GPIHBP1 binds LPL and is expressed by endo-
thelial cells, suggested that GPIHBP1 could be the binding site
for LPL on endothelial cells (i.e., the ‘‘platform’’ for LPL-mediated
lipoprotein processing) (Beigneux et al., 2007).
Subsequent studies showed that GPIHBP1’s function is more
substantial than simply serving as a binding site for LPL within
capillaries. GPIHBP1 is a partner for LPL in four ways (Figure 1).
First, the GPIHBP1 on the abluminal surface of capillary
endothelial cells is important for capturing LPL from within
the subendothelial spaces (Davies et al., 2010; Kristensen
et al., 2018). Second, GPIHBP1 binding stabilizes the structure
and activity of LPL (Kristensen et al., 2018; Mysling et al.,
2016a, 2016b). Third, GPIHBP1 shuttles LPL across capillary
endothelial cells to its site of action in the capillary lumen (Davies
et al., 2010). Fourth, GPIHBP1-bound LPL is critical for the
margination of lipoproteins in the bloodstream (Goulbourne
et al., 2014), allowing LPL-mediated lipoprotein processing to
proceed.
GPIHBP1’s function in transporting LPL across capillaries was
evident from the distribution of LPL in the tissues of wild-type
and Gpihbp1 ⁄ mice (Davies et al., 2010). In wild-type mice,
most of the LPL in tissues is bound to GPIHBP1 on capillary
endothelial cells (Figure 2A), where it is distributed evenly be-
tween the luminal and abluminal plasma membranes (Davies
et al., 2012, 2010). In Gpihbp1 ⁄ mice, LPL never reaches the
capillary lumen (Davies et al., 2010). Instead, the LPL in
Gpihbp1 ⁄ mice remains stranded within the interstitial spaces,
presumably bound to HSPGs in close proximity to the surface of

52 Cell Metabolism 30, July 2, 2019

Cell Metabolism
Review
cells (both parenchymal cells and capillary endothelial cells)
(Davies et al., 2010) (Figure 2A).
By immunohistochemistry, GPIHBP1 is expressed solely in
capillary endothelial cells and is absent in endothelial cells of ve-
nules, arterioles, and larger blood vessels (Beigneux et al., 2007;
Davies et al., 2010) (Figure 2B). GPIHBP1 is found in capillaries
of all peripheral tissues, with particularly high levels in the heart
and adipose tissue (where LPL-mediated processing of lipopro-
teins is robust), but it is absent from capillaries of the brain
(Beigneux, 2010; Beigneux et al., 2007; Davies et al., 2010; Olafsen
et al., 2010), which uses glucose for fuel (Mergenthaler et al., 2013).
Consistent with immunohistochemistry studies, Gpihbp1 tran-
scripts are absent from capillary endothelial cells of the brain, as
judged by single-cell RNA sequencing (RNA-seq) studies (He
et al., 2018b; Vanlandewijck et al., 2018). Interestingly, high levels
of GPIHBP1 expression are observed in capillaries of the lung, a
tissue where LPL expression is low (Olafsen et al., 2010). GPIHBP1
in lung capillaries appears to play a role in scavenging LPL that
escapes into the plasma compartment (Goulbourne et al., 2014),
but the physiologic importance of that role is unclear. GPIHBP1
deficiency does not appear to affect pulmonary function.
Why GPIHBP1 expression is restricted to the endothelial cells
of the capillaries of peripheral tissues and why GPIHBP1 is ex-
pressed at higher levels in some tissues than others are poorly
understood. An enhancer element was recently identified
3.6 kb upstream of exon 1 of Gpihbp1 (Allan et al., 2019). De-
leting that enhancer reduced Gpihbp1 transcripts in heart and
brown adipose tissue by 50% but nearly eliminated transcripts
in liver and kidney (Allan et al., 2019).
By immunohistochemistry, the tissue distribution of LPL in pe-
ripheral tissues mirrors that of GPIHBP1. Even though LPL is
synthesized and secreted by parenchymal cells, the vast major-
ity of the LPL in tissues is bound to GPIHBP1 on capillary endo-
thelial cells (Beigneux et al., 2007; Davies et al., 2010).
Figure 2. Localization of GPIHBP1 and
Lipoprotein Lipase (LPL) in Brown Adipose
Tissue (BAT)
(A) Mislocalization of LPL in the BAT of Gpihbp1 ⁄
mice. Sections of BAT from a Gpihbp1+/+ and
Gpihbp1 ⁄ mouse were stained with antibodies
against CD31 (green), GPIHBP1 (magenta), and
LPL (red), and imaged by confocal fluorescence
microscopy. DNA was stained with DAPI (blue).
Scale bar, 50 mm.
(B) GPIHBP1 is expressed in endothelial cells of
capillaries, but not in endothelial cells of larger
blood vessels. Sections of BAT were stained with
antibodies against CD31 (green) and GPIHBP1
(red). GPIHBP1 expression in endothelial cells
decreases as capillaries merge into a venule
(arrow). Scale bar, 50 mm.
GPIHBP1’s transport function was
established by studies showing that
GPIHBP1 transports a GPIHBP1-specific
monoclonal antibody from one side of an
endothelial cell to the other (Davies
et al., 2010). The movement of GPIHBP1
across endothelial cells is bidirectional
(Davies et al., 2012). When wild-type
mice are injected intravenously with a flu-
orescently labeled GPIHBP1-specific monoclonal antibody, the
antibody initially binds to GPIHBP1 within the capillary lumen
but then quickly appears on the abluminal surface of capillaries.
Conversely, when the antibody is injected into tissues, it first
associates with GPIHBP1 on the outside of capillaries but then
quickly appears in the capillary lumen. Antibody transport did
not occur in Gpihbp1 ⁄ mice. GPIHBP1 appears to be a long-
lived protein (Olafsen et al., 2010), implying that GPIHBP1 could
make multiple trips across endothelial cells, with each trip
providing an opportunity to capture LPL and bring it to the capil-
lary lumen.
The intravascular processing of triglyceride-rich lipoproteins
by LPL requires that the lipoproteins in the bloodstream stop
(marginate) along capillaries. For years, lipoprotein margination
was assumed to involve electrostatic interactions between
positively charged apolipoproteins on lipoproteins and the nega-
tively charged HSPGs that line the luminal surface of endothelial
cells (de Beer et al., 1999; Lookene et al., 1997b; Wang and
Eckel, 2009). More recent studies revealed that lipoprotein
margination is utterly dependent on GPIHBP1—and more spe-
cifically on the GPIHBP1-LPL complex (Goulbourne et al.,
2014). GPIHBP1 expression alone is not sufficient for lipoprotein
margination along capillaries; lipoprotein margination is negli-
gible in lung capillaries, where GPIHBP1 is abundant but LPL
expression is low. However, when the GPIHBP1 in lung capil-
laries was artificially loaded with exogenous bovine LPL, lipopro-
tein margination along lung capillaries was robust (Goulbourne
et al., 2014).
Fluorescently labeled lipoproteins, when injected intrave-
nously into wild-type mice, rapidly marginate along capillaries
of the heart, co-localizing with LPL and GPIHBP1. In contrast,
lipoprotein margination is negligible along capillaries in hearts
of Gpihbp1 ⁄ mice, which are devoid of GPIHBP1-LPL com-
plexes (Goulbourne et al., 2014). In vitro studies indicated that
Cell Metabolism 30, July 2, 2019 53

lipoprotein binding to the GPIHBP1-LPL complex depends on a
tryptophan-rich loop within the carboxy-terminal domain of LPL
(Goulbourne et al., 2014).
Because the GPIHBP1-LPL complex is anchored to the
plasma membrane of endothelial cells and because the luminal
surface of endothelial cells is covered by an HSPG-rich glycoca-
lyx, it was initially unclear how the GPIHBP1-LPL complex would
be able to participate in lipoprotein margination. However, elec-
tron microscopy (EM) studies revealed that the glycocalyx on the
surface of capillaries is patchy, with tufts of glycocalyx interrup-
ted by ‘‘meadows’’ where the plasma membrane is exposed
(Goulbourne et al., 2014). EM tomography suggested that lipo-
proteins bind to capillaries within the gaps in the glycocalyx
(Goulbourne et al., 2014).
Relevance of GPIHBP1 Expression to Uptake of
Lipoprotein-Derived Nutrients by Tissues
The turnover of triglyceride-rich lipoproteins in the bloodstream
is quite rapid (Havel, 2010). Recent NanoSIMS imaging studies
revealed that the movement of the fatty acid products of lipolysis
across endothelial cells and into surrounding parenchymal cells
is also rapid (He et al., 2018a). After administering [2H]triglycer-
ide-enriched lipoproteins (2H-TRLs) to wild-type mice by an
intravenous injection, it was possible to visualize the margination
of 2H-TRLs in heart capillaries as early as 30 s after the injection
(He et al., 2018a). As early as 2 min after the injection, substantial
amounts of deuterium had already entered the mitochondria and
cytosolic lipid droplets of cardiomyocytes (He et al., 2018a)
(Figure 3A). Interestingly, deuterium enrichment in capillary
endothelial cells and cardiomyocytes was similar (i.e., there
was little evidence that endothelial cells play a gatekeeper role
in the movement of fatty acids into cardiomyocytes). When iso-
lated hearts from wild-type mice were perfused with 2H-TRLs,
similar findings were observed—rapid margination of 2H-TRLs
along capillaries and rapid uptake of fatty acids into cytosolic
lipid droplets (He et al., 2018a). When isolated hearts from
Gpihbp1 ⁄ mice were perfused with 2H-TRLs, the margination
of 2H-TRLs was absent and 2H enrichment in the cytosolic tri-
glyceride droplets of cardiomyocytes was markedly reduced
(He et al., 2018a).
54 Cell Metabolism 30, July 2, 2019
Cell Metabolism
Review
Figure 3. NanoSIMS Imaging to Examine
Intravascular Lipolysis
(A) 2H/1H ratio image of a capillary in a mouse heart
2 min after an intravenous injection of 2H-TRLs
(triglyceride-rich lipoproteins enriched in 2H-tri-
glycerides). The 2H/1H image shows 2H-TRLs that
have marginated along the luminal surface of capil-
lary endothelial cells and entry of 2H-enriched lipids
into endothelial cells as well as the mitochondria and
lipid droplets of cardiomyocytes. The 2H/1H ratio is
multiplied by 10,000; the range spans from slightly
higher than 2H natural abundance to approximately
six times natural abundance. Scale bar, 1 mm.
(B) Composite 12C14N and 32S /12C14N images of
mouse heart capillaries. The signal from 12C14N
secondary ions (blue), reflecting 14N distribution,
was robust in both capillary endothelial cells (EC)
and cardiomyocytes. The 32S /12C14N ratio (green)
was set from 0.05 to 0.06. Pixels with a 32S /12C14N
ratio above 0.05 are shown, revealing features rich
in 32S, relative to 14N, on the surface of endothelial
cells and cardiomyocytes. Scale bar, 1 mm.
Properties of GPIHBP1 Required for LPL Binding
Mature GPIHBP1 contains two key domains, a disordered
amino-terminal acidic domain and an LU domain containing 10
cysteines, all arranged in a characteristic spacing pattern and
all disulfide bonded (Beigneux et al., 2011; Mysling et al.,
2016a). The five disulfide bonds stabilize the assembly of three
loops and form the hallmark three-fingered scaffold of the LU
domain (Leth et al., 2019). The acidic domain of human GPIHBP1
is impressive, with 21 acidic amino acids (glutamates or aspar-
tates) in 26 consecutive residues. A sulfated tyrosine, located
in the middle of the acidic domain (Kristensen et al., 2018),
adds to the negative charge. The GPIHBP1 in every mammalian
species contains an acidic domain, but the precise sequence
and length of the domain is variable. For example, the acidic
domain in opossum GPIHBP1 contains 32 aspartates or gluta-
mates in 39 consecutive residues, including a stretch of 23
consecutive aspartates (Fong et al., 2016). Like all GPI-anchored
proteins, GPIHBP1 contains a carboxy-terminal hydrophobic
signal peptide that triggers the addition of a GPI anchor within
the endoplasmic reticulum (Ferguson et al., 2009).
GPIHBP1’s LU domain is required for the stability of LPL bind-
ing, contributing to a long half-life for the GPIHBP1-LPL complex
(t½  35 s). When GPIHBP1’s LU domain is replaced with an LU
domain from another member of the LU family (Gin et al., 2008) or
when any of the 10 cysteines in GPIHBP1’s LU domain is
mutated (Beigneux et al., 2009b), stable interactions with LPL
are abolished. Aside from the 10 cysteines, alanine-scanning
mutagenesis of GPIHBP1’s LU domain uncovered 12 additional
residues required for LPL binding, with the majority located in
conserved segments within the second finger of the LU domain
(amino acids 89–110) (Beigneux et al., 2011). Many mutations in
GPIHBP1’s LU domain interfered with disulfide bonding and pro-
moted the formation of disulfide-linked GPIHBP1 dimers and
multimers (Beigneux et al., 2015; Plengpanich et al., 2014), but
there was one noteworthy exception. Mutations in GPIHBP1
Trp-109 abolished LPL binding without interfering with disulfide
bonding or promoting protein multimerization, suggesting that
Trp-109 plays a direct role in the GPIHBP1-LPL binding interface
(Beigneux et al., 2015). Consistent with that idea, hydrogen-
deuterium exchange mass spectrometry (HDX-MS) studies
Cell Metabolism
Review
showed that the binding of LPL to GPIHBP1 protects GPIHBP1
residues 104–128 from solvent exposure and reduces deuterium
uptake into GPIHBP1 (Mysling et al., 2016a).
Studies by Gin and coworkers (Gin et al., 2012) revealed that
GPIHBP1 binds to the carboxy-terminal domain of LPL (residues
325–475). Consistent with those findings, an LPL-specific mono-
clonal antibody with an epitope within the carboxy-terminal
domain of LPL, 88B8, blocks LPL binding to GPIHBP1 (Allan
et al., 2016). Also, two different missense mutations within the
carboxy-terminal domain of LPL, C445Y and E448K, first identi-
fied in patients with severe hypertriglyceridemia (Henderson
et al., 1996, 1998), abolish LPL binding to GPIHBP1 (Voss
et al., 2011). Finally, HDX-MS studies revealed that when
GPIHBP1 is bound to LPL, residues 429–446 in LPL’s carboxy-
terminal domain are protected from solvent exposure and deute-
rium uptake, implying that those residues are involved in
GPIHBP1 binding (Mysling et al., 2016a).
Deciphering Functions for GPIHBP1’s Intrinsically
Disordered Acidic Domain
To explore the contributions of GPIHBP1’s LU and acidic
domains to the formation of the GPIHBP1-LPL complex, the ki-
netics of LPL binding to full-length GPIHBP1 and ‘‘Dacidic-
GPIHBP1’’ (a GPIHBP1 mutant lacking the amino-terminal acidic
domain) were analyzed by surface plasmon resonance (SPR). In
these studies, the two different GPIHBP1 proteins were passed
over a sensor chip coated with LPL (which had been tethered
to the sensor chip with the LPL-specific monoclonal antibody
5D2) (Kristensen et al., 2018; Mysling et al., 2016a). Both full-
length GPIHBP1 and Dacidic-GPIHBP1 bound LPL with nano-
molar affinities, but the binding of full-length GPIHBP1 was
stronger. Interestingly, the stability of the GPIHBP1-LPL
complex, as judged by the dissociation rate constant (koff), was
nearly identical for full-length GPIHBP1 and Dacidic-GPIHBP1
(2 3 10 2 s 1) (Mysling et al., 2016a). However, the association
rate constants (kon) for Dacidic-GPIHBP1 (0.1 3 106 M 1s 1)
and full-length GPIHBP1 (2.6 3 107 M 1s 1) were very different,
highlighting a key role for GPIHBP1’s acidic domain in
accelerating the encounter rate and promoting the formation
of the GPIHBP1-LPL complex (Kristensen et al., 2018). Further-
more, the kon for full-length GPIHBP1-LPL binding was sensitive
to the ionic strength of the buffer and was as high as
3 3 108 M 1s 1 under physiological conditions, underscoring
the importance of electrostatic steering in the initial interaction
between LPL and full-length GPIHBP1 (Kristensen et al., 2018).
Under physiologic conditions, GPIHBP1’s disordered acidic
domain increases the velocity of LPL binding by >2,500-fold.
These observations are likely relevant to the ability of GPIHBP1
to capture LPL from within the interstitial spaces along the baso-
lateral surface of capillaries.
Kristensen and coworkers recently investigated GPIHBP1-
LPL interactions in more detail (Kristensen et al., 2018). They es-
tablished that the stoichiometry of GPIHBP1 binding to LPL is 1:1
(Kristensen et al., 2018). Remarkably, a peptide corresponding
to GPIHBP1’s acidic domain also bound to LPL with 1:1 stoichi-
ometry. They also found, by SPR, that the sulfated tyrosine (Tyr-
38) in GPIHBP1’s acidic domain contributes to the strength of
LPL binding. When Tyr-38 was replaced with Phe, the KD for
GPIHBP1-LPL binding increased 3-fold (Kristensen et al.,
2018). Finally, they documented, by SPR, large differences in
the abilities of full-length GPIHBP1 and Dacidic-GPIHBP1 to
dislodge LPL when it was bound to a sensor chip containing im-
mobilized short-chain heparin moieties. Full-length GPIHBP1
dislodged heparin-bound LPL efficiently, whereas Dacidic-
GPIHBP1 did not (despite the fact that Dacidic-GPIHBP1 was
able to bind to the LPL) (Kristensen et al., 2018).
The fact that full-length GPIHBP1 effectively captures LPL
from heparin is probably relevant to LPL trafficking to capillaries.
As noted earlier, the LPL in Gpihbp1 ⁄ mice is stranded within
the interstitial spaces, bound to HSPGs near the surface of cells
(Davies et al., 2010). The interactions between LPL and HSPGs
within the interstitium are transient (i.e., with a high koff), but the
abundance of HSPG-binding sites means that LPL in Gpihbp1 ⁄
tissues is retained within the interstitium rather than escaping
into the lymphatic circulation. The same interstitial HSPG-bind-
ing sites exist in wild-type mice; however, the dynamic nature
of LPL-HSPG interactions allows LPL to move to more stable
GPIHBP1-binding sites on capillaries. The ability of GPIHBP1
to recruit LPL away from interstitial HSPG-binding sites has
been documented experimentally. When GPIHBP1-coated
agarose beads were implanted into the brown adipose tissue
of Gpihbp1 ⁄ mice, HSPG-bound LPL within the interstitial
spaces moved to the GPIHBP1-coated beads, depleting LPL
stores within the interstitium (Allan et al., 2017b).
Recent immunohistochemical studies on LPL localization in
tissues of Gpihbp1 ⁄ mice yielded further insights into the traf-
ficking of LPL within the interstitium (Kristensen et al., 2018). As
expected, the LPL in the heart and skeletal muscle of Gpihbp1 ⁄
mice was mislocalized to the interstitium. Interestingly, however,
the amount of LPL adjacent to endothelial cells was greater than
the amount adjacent to myocytes (Kristensen et al., 2018). Thus,
even in the absence of GPIHBP1, the LPL within the interstitium
appears to associate preferentially with HSPGs on the outside of
capillary endothelial cells. We propose that ‘‘directed diffusion’’
(Duchesne et al., 2012) along an HSPG electrostatic gradient is
responsible for the flow of LPL from the HSPGs near myocytes
to HSPGs on the outer surface of endothelial cells. The same
directional LPL movement presumably exists in tissues of wild-
type mice, generating a pool of LPL for capture by GPIHBP1
on endothelial cells. Small-angle X-ray scattering (SAXS) studies
showed that GPIHBP1’s disordered acidic domain, as it projects
from the LU domain, occupies a large mushroom-shaped space
with a diameter of 112 Å (Kristensen et al., 2018). In tissues,
GPIHBP1’s acidic domain likely projects at least 100 Å from
the basolateral plasma membrane of endothelial cells and prob-
ably functions in a fly casting-like mechanism (Shoemaker et al.,
2000) to engage the LPL that is bound in a dynamic fashion to
HSPGs and drive it into a stable complex with GPIHBP1’s LU
domain.
The explanation for the preferential association of interstitial
LPL with endothelial cells in Gpihbp1 ⁄ mice is unknown, but
the existence of HSPGs on endothelial cells is well documented.
Collagen XVIII, a basement membrane HSPG, is found on capil-
lary endothelial cells, as judged by immunohistochemistry
(Bishop et al., 2010). Also, NanoSIMS analyses of 32S distribution
in tissues appear consistent with the existence of HSPGs adja-
cent to cardiomyocytes and endothelial cells (Figure 3B). 32S is
found in methionines and cysteines of all cellular proteins, but
Cell Metabolism 30, July 2, 2019 55

the amount of 32S (relative to 14N) is greater within the interstitial
spaces near the surface of cardiomyocytes and endothelial cells
than in the 14N-rich cytoplasm of those cells. Larger amounts
of 32S, relative to 14N, are also observed in the capillary lumen,
likely because of HSPGs in the glycocalyx of endothelial cells
(Figure 3B).
GPIHBP1 Stabilizes the Structure and Activity of LPL,
even in the Presence of Endogenous LPL Inhibitors
For years, biochemists and physiologists have noted that cata-
lytic activity disappears rapidly when purified preparations of
LPL are incubated at room temperature. HDX-MS studies by
Mysling and coworkers (Mysling et al., 2016a) provided a likely
explanation. When LPL is incubated at 25 C with deuterated
water (2H2O), deuterium uptake into specific regions of LPL re-
sults in bimodal isotope envelopes, reflecting progressive un-
folding of those regions (Figure 4A). Bimodal isotope envelopes
were observed in peptides corresponding to LPL’s amino-termi-
nal hydrolase domain, including in a peptide spanning LPL’s cat-
alytic triad (Figure 4B), but they were nearly absent in peptides
from LPL’s carboxy-terminal domain, which has a more stable
conformation. The appearance of bimodal deuterium signatures
in the amino-terminal domain was accompanied by a parallel fall
in LPL catalytic activity (Mysling et al., 2016a). The binding of
GPIHBP1 to LPL markedly reduced both the unfolding of LPL’s
amino-terminal domain and the loss of catalytic activity, demon-
strating that GPIHBP1 preserves the structural integrity and cat-
alytic activity of LPL. Interestingly, the protective effect of
GPIHBP1 was largely due to GPIHBP1’s intrinsically disordered
acidic domain (Mysling et al., 2016a). First, the preservation
of LPL structure and catalytic activity was much greater with
full-length GPIHBP1 than with Dacidic-GPIHBP1 (Mysling
et al., 2016a). Second, a synthetic peptide corresponding to
GPIHBP1’s acidic domain reduced both LPL unfolding and the
loss of triglyceride hydrolase activity.
56 Cell Metabolism 30, July 2, 2019
Cell Metabolism
Review
Figure 4. Inherent Instability of LPL, as
Judged by HDX-MS Studies
Bovine LPL was incubated at 25 C in deuterium
oxide.
(A) The emergence of a bimodal isotope envelope
of deuterium uptake in a peptic peptide covering
the catalytic triad of LPL (from Mysling et al.,
2016a). Bimodality of the isotope envelope is a
hallmark of protein unfolding (Mysling et al., 2016a,
2016b). Substantial portions of the amino-terminal
catalytic domain of LPL (NTD) exhibited the
bimodal isotope envelope, while peptides from
the carboxy-terminal lipid-binding domain (CTD)
exhibit a unimodal exchange pattern, reflecting
greater stability.
(B) Position of the catalytic triad peptic peptide
(green) in the crystal structure of human LPL
(modified from the crystal structure by Birrane and
coworkers [Birrane et al., 2019]). The structure of
human LPL is shown as a cartoon representation,
with a helices in red and b strands in cyan.
The location of the three catalytic triad residues
(S159, D183, and H268) are noted. The structure
was visualized with PyMol (Schrödinger) using
PDB: 6E7K.
The spontaneous unfolding of LPL, as judged by HDX-MS
studies, is more than an in vitro biochemical curiosity. In
follow-up studies, Mysling and coworkers (Mysling et al.,
2016b) showed that ANGPTL4, a physiologic inhibitor of LPL, in-
activates LPL by catalyzing LPL unfolding. Spontaneous unfold-
ing of LPL, as judged by HDX-MS, is relatively slow, with 7%
unfolding after 5 min at 25 C and 29% unfolding after 30 min.
In contrast, incubating LPL with sub-stoichiometric amounts of
ANGPTL4 (a 10:1 LPL to ANGPTL4 molar ratio) resulted in
70% unfolding of LPL in 5 min. Mysling and coworkers also
tested the capacity of an ANGPTL4 polymorphic variant,
ANGPTL4E40K, to catalyze LPL unfolding (Mysling et al.,
2016b). The ANGPTL4E40K variant, present in 3% of Caucasians,
is associated with lower plasma triglyceride levels and reduced
risk of coronary disease (Dewey et al., 2016; Helgadottir et al.,
2016; Myocardial Infarction Genetics and CARDIoGRAM
Exome Consortia Investigators et al., 2016). The capacity of
ANGPTL4E40K to catalyze LPL unfolding was 60% lower than
that of wild-type ANGPTL4. Consistent with those findings,
LPL incubated with ANGPTL4E40K retained 40%–50% more cat-
alytic activity than LPL incubated with wild-type ANGPTL4. The
reduced capacity of ANGPTL4E40K to inactivate LPL, which is
almost certainly due to instability of a functionally important
a-helix in ANGPTL4 (Mysling et al., 2016b), presumably explains
the reduced plasma triglyceride levels in ANGPTL4E40K carriers.
The fact that LPL activity is regulated by an inhibitory protein
(i.e., ANGPTL4) that promotes unfolding is, to the best of our
knowledge, a novel mechanism for controlling the activity of an
enzyme. Mysling and colleagues (Mysling et al., 2016b) pro-
posed that ANGPTL4 interacts transiently with LPL and pro-
motes an unstable conformation that leads inexorably to enzyme
inactivation. Another group reported that inactivation of LPL by
ANGPTL4 is reversible when the inactivation of LPL is performed
in the presence of deoxycholate (Gutgsell et al., 2019; Lafferty
et al., 2013), a detergent that stabilizes LPL activity (Bengtsson
Cell Metabolism
Review
and Olivecrona, 1979). The significance of this finding is unclear,
in part because the relevance of deoxycholate-stabilized LPL to
the normal physiology of intravascular lipolysis is open to ques-
tion. Also, the LPL activity in those studies was assessed with a
soluble substrate rather than with an emulsified triglyceride sub-
strate. Finally, the ANGPTL4 proved to be a relatively inefficient
inhibitor under the conditions that were studied, with a Ki of
0.86–1.7 mM (Gutgsell et al., 2019; Lafferty et al., 2013).
In the studies by Mysling and coworkers, full-length GPIHBP1
protected LPL from ANGPTL4-mediated unfolding, but there
was little protection by Dacidic-GPIHBP1 or an acidic domain
peptide (Mysling et al., 2016b). Of note, the inactivation of LPL
by ANGPTL4 was not reversible; the loss of LPL activity with
ANGPTL4 could not be reversed by adding GPIHBP1 to the incu-
bation mixture (Mysling et al., 2016b). Mysling and coworkers
proposed that the anchoring of GPIHBP1’s LU domain to LPL re-
sults in optimal positioning of GPIHBP1’s acidic domain, allow-
ing it to stabilize LPL structure and activity (Mysling et al.,
2016b). From the standpoint of plasma triglyceride physiology,
the ability of GPIHBP1 to preserve LPL structure and activity
(even in the presence of inhibitory proteins) probably serves to
focus catalytically active LPL where it needs to be for lipoprotein
processing (i.e., in the capillary lumen). It seems likely that at
least some of the regulation of LPL by ANGPTL4 occurs before
LPL reaches the capillary lumen. Kersten and collaborators
showed that ANGPTL4 can inactivate LPL in the secretory
pathway of adipocytes (Dijk et al., 2016). Also, the LPL in the
interstitial spaces, as it moves toward endothelial cells, could
be susceptible to inhibition by ANGPTL4. Nilsson and coworkers
suggested that the interstitial spaces may be a site for
ANGPTL4-mediated regulation of LPL (Nilsson et al., 2012).
ANGPTL3, another physiologic inhibitor of LPL, also catalyzes
LPL unfolding, but it is less potent than ANGPTL4—at least un-
der the experimental conditions that were tested (Mysling
et al., 2016b). ANGPTL3-catalyzed LPL unfolding is also in-
hibited by the binding of GPIHBP1. In the case of ANGPTL3,
several studies have demonstrated that the functional inhibitory
unit is actually a complex of ANGPTL3 and ANGPTL8 (Chi et al.,
2015; Haller et al., 2017; Quagliarini et al., 2012) and that this
complex is as efficient as ANGPTL4 in inactivating LPL (Kovrov
et al., 2019).
Familial Chylomicronemia from GPIHBP1 Mutations
Soon after the discovery of chylomicronemia in Gpihbp1 ⁄
mice, GPIHBP1 mutations were identified in patients with familial
chylomicronemia syndrome (Ariza et al., 2016; Beigneux et al.,
2009a; Charrière et al., 2011; Coca-Prieto et al., 2011; Franssen
et al., 2010; Gonzaga-Jauregui et al., 2014; Olivecrona et al.,
2010; Plengpanich et al., 2014; Rabacchi et al., 2016; Rios
et al., 2012; Yamamoto et al., 2013). The chylomicronemia that
results from GPIHBP1 mutations presents in infancy (Rios
et al., 2012) and persists lifelong, with plasma triglycerides often
exceeding 2,000 mg/dL. Two mutant GPIHBP1 alleles are
required for disease; the plasma triglyceride levels in heterozy-
gotes are normal. A variety of different GPIHBP1 mutations
have been uncovered but the majority has involved one of the
cysteines in the LU domain (e.g., C65Y, C65S, C68Y, C68G,
C68R, C83R, and C89F) or a residue in close proximity to a
cysteine (e.g., Q115P and T111P). In one kindred, chylomicrone-
mia was caused by the introduction of an unpaired cysteine into
the LU domain (S107C) (Plengpanich et al., 2014). Chylomicro-
nemia was also observed in the setting of a T80K mutation,
which interferes with N-linked glycosylation and trafficking of
GPIHBP1 to the plasma membrane (Ariza et al., 2016). A
G175R substitution has been observed in several patients with
hypertriglyceridemia (Beigneux et al., 2017; Charrière et al.,
2011; Chyzhyk et al., 2019) but is unlikely to be causative.
G175 is located in the signal peptide that is normally replaced
by the GPI anchor (i.e., it is not present in mature GPIHBP1).
The G175R substitution has no significant effect on the addition
of the GPI anchor (Beigneux et al., 2017). No one has yet uncov-
ered a GPIHBP1 mutation that results in the deletion of all or part
of GPIHBP1’s acidic domain.
Low levels of LPL in the pre- and post-heparin plasma are a
hallmark of GPIHBP1 deficiency and presumably reflect reduced
amounts of LPL within capillaries (Beigneux et al., 2009a; Olive-
crona et al., 2010; Plengpanich et al., 2014). Low levels of
GPIHBP1 in the plasma are another hallmark (Beigneux et al.,
2017). Heterozygous carriers of GPIHBP1 mutations have
half-normal plasma levels of GPIHBP1 (Beigneux et al., 2017;
Miyashita et al., 2018b).
GPIHBP1 missense mutations causing chylomicronemia
abolish the capacity of GPIHBP1 to bind LPL (Franssen et al.,
2010; Olivecrona et al., 2010; Plengpanich et al., 2014). In cell
transfection studies, missense mutations involving cysteine
residues (e.g., C65Y, C65S, and C68G) interfere with disulfide
bonding and result in the appearance of disulfide-linked
GPIHBP1 dimers and multimers on the surface of cells (Beigneux
et al., 2015). The GPIHBP1 dimers and multimers have no capac-
ity to bind LPL (Beigneux et al., 2015; Plengpanich et al., 2014).
The fact that dimeric and multimeric GPIHBP1 proteins were
able to reach the surface of the transfected cells was somewhat
surprising, given that a cysteine mutation in CD59 (another LU
protein) abolished CD59 trafficking to the plasma membrane
(Nevo et al., 2013). Given that observation, it seemed possible
that the appearance of GPIHBP1 dimers and multimers on the
surface of transfected cells was simply an artifact of protein
overexpression. To explore that possibility, Allan and coworkers
(Allan et al., 2017a) created knockin mice harboring a cysteine
substitution in the LU domain (C63Y, corresponding to a muta-
tion observed in humans; Franssen et al., 2010). Homozygotes
exhibited severe chylomicronemia, even on a low-fat chow
diet. GPIHBP1-C63Y reached the surface of endothelial cells
and could be easily detectable by immunohistochemistry, but
the amount was reduced by 70% (Allan et al., 2017a). No LPL
was bound to GPIHBP1-C63Y in the capillary lumen, demon-
strating that the mutant protein had no capacity to bind LPL.
Interestingly, most of the GPIHBP1-C63Y on the surface of
endothelial cells in vivo was monomeric; only small amounts of
disulfide-linked dimers were detected. Thus, the capillary endo-
thelial cells in vivo appear to be more efficient than transfected
CHO cells in removing misfolded GPIHBP1 proteins (including
dimers and multimers), thereby limiting the amounts of GPIHBP1
that reach the surface of cells.
Chylomicronemia from GPIHBP1 Autoantibodies
Chylomicronemia occasionally appears spontaneously, and
some of these ‘‘acquired’’ cases are caused by GPIHBP1
Cell Metabolism 30, July 2, 2019 57

autoantibodies (‘‘GPIHBP1 autoantibody syndrome’’) (Beigneux
et al., 2017; Eguchi et al., 2019; Hu et al., 2017a; Kersten, 2017).
This disorder was discovered fortuitously while characterizing a
monoclonal antibody-based ELISA designed to measure plasma
levels of GPIHBP1 (Beigneux et al., 2017; Miyashita et al.,
2018a). In those studies, two plasma samples, both from pa-
tients with chylomicronemia, had abnormally low GPIHBP1
levels, and the GPIHBP1 levels remained low even after
‘‘spiking’’ the plasma samples with recombinant GPIHBP1. The
latter finding strongly suggested ‘‘immunoassay interference,’’
which can be caused by autoantibodies. Additional studies re-
vealed that the plasma from the two chylomicronemia patients
did indeed contain GPIHBP1 autoantibodies (Beigneux et al.,
2017). Of note, the autoantibodies abolished the capacity of
GPIHBP1 to bind LPL. Patients with the GPIHBP1 autoantibody
syndrome invariably have low plasma levels of LPL, consistent
with the inability of GPIHBP1 to bind LPL and transport it to
the capillary lumen (Beigneux et al., 2017). Subsequent studies
revealed that GPIHBP1 autoantibodies are directed against
GPIHBP1’s LU domain, not the acidic domain (Kristensen
et al., 2018).
The plasma triglyceride levels in patients with the GPIHBP1
autoantibody syndrome are extremely high, typically exceeding
1,500 mg/dL, and many of the patients had experienced an
episode of acute pancreatitis. In some cases, GPIHBP1 autoan-
tibodies were the sole manifestation of autoimmune disease, but
the majority of patients had clinical or serological evidence of an
autoimmune disease (e.g., systemic lupus erythematosus; SLE)
(Beigneux et al., 2017). One patient with SLE and GPIHBP1
autoantibodies became pregnant and delivered a baby girl.
The newborn infant manifested severe chylomicronemia that
resolved after several weeks, presumably coinciding with disap-
pearance of maternal GPIHBP1 autoantibodies (Beigneux et al.,
2017). In another patient with autoimmune thyroid disease and
multiple sclerosis (Kawahara et al., 2017), GPIHBP1 autoanti-
bodies appeared after initiating b-interferon treatment (Eguchi
et al., 2019). (b-interferon can, in some patients, cause wors-
ening of autoimmune diseases; Banchereau and Pascual,
2006; Di Domizio and Cao, 2013; Silva, 2012.) In that patient,
the GPIHBP1 autoantibodies disappeared and the plasma tri-
glycerides normalized after b-interferon therapy was discontin-
ued (Eguchi et al., 2019).
In the patient treated with b-interferon, the GPIHBP1 autoanti-
bodies were largely subclass IgG4 (Eguchi et al., 2019). IgG4
autoantibodies have been reported to cause a variety of autoim-
mune diseases, typically by disrupting protein-protein interac-
tions rather than by fixing complement and inducing tissue injury
(Huijbers et al., 2018; Koneczny, 2018). Whether most patients
with the GPIHBP1 autoantibody syndrome have IgG4 autoanti-
bodies remains to be determined. It will also be important to
determine if immune complexes are present in the plasma of
affected patients. The presence of immune complexes could
have been overlooked by the ELISAs used to detect GPIHBP1
autoantibodies (Beigneux et al., 2017; Miyashita et al., 2018a).
The frequency of the GPIHBP1 autoantibody syndrome
among patients with acquired forms of chylomicronemia is un-
clear, but our experience, while limited, suggests that the syn-
drome may not be particularly rare (Beigneux et al., 2017). In
our experience, if any GPIHBP1 autoantibody is detected, the
58 Cell Metabolism 30, July 2, 2019
Cell Metabolism
Review
patient invariably has severe chylomicronemia. We have no
evidence that GPIHBP1 autoantibodies explain cases of mild-
to-moderate hypertriglyceridemia. In one patient, the chylomi-
cronemia resulting from GPIHBP1 autoantibodies resolved
spontaneously (Hu et al., 2017a). In two cases, the chylomicro-
nemia resolved after initiating immunosuppressive treatment,
and in one of those cases, the normalization of triglyceride levels
was accompanied by the disappearance of GPIHBP1 autoanti-
bodies (Beigneux et al., 2017). More experience is needed to
establish the natural history and treatment strategies for the
GPIHBP1 autoantibody syndrome.
A Crystal Structure for the GPIHBP1-LPL Complex
The structure of LPL remained elusive for decades, forcing in-
vestigators to rely on a homology model (Kobayashi et al.,
2002) derived from the structure of pancreatic lipase, a distantly
related lipase family member (Winkler et al., 1990). The absence
of a structure for LPL was not for lack of trying; several labora-
tories attempted to generate LPL crystals, but none succeeded,
likely because of the propensity of LPL to unfold (Mysling et al.,
2016a). Inspired by the discovery that GPIHBP1 stabilizes LPL
structure and activity, Birrane and coworkers solved the crystal
structure of a GPIHBP1-LPL complex at 2.8-Å resolution (Birrane
et al., 2019; Horton, 2019).
The crystallographic unit contained two LPL-GPIHBP1 com-
plexes, with the two LPL molecules arranged in a head-to-tail
orientation, with the Trp-rich loop in LPL’s carboxy-terminal
domain interacting with the catalytic pocket in the LPL’s
amino-terminal domain. The same conformation was observed
for the complex in solution by SAXS studies (Birrane et al.,
2019). The structure of LPL contained an amino-terminal hydro-
lase domain with six a helices and ten b strands and a carboxy-
terminal flattened b-barrel domain containing twelve b strands
(Birrane et al., 2019). LPL contained two N-linked glycans and
one calcium ion in the amino-terminal catalytic domain.
Portions of the Trp-rich loop and the lid covering the catalytic
pocket were not well defined in the structure by Birrane and co-
workers (Birrane et al., 2019). However, very recently, Arora and
coworkers solved the crystal structure for a GPIHBP1-LPL com-
plex in the presence of a weak competitive inhibitor of LPL (Arora
et al., 2019). That LPL structure was similar to the one reported
earlier (Birrane et al., 2019); however, the presence of the inhib-
itor in LPL’s active site permitted a higher definition of the Trp-
rich loop and the lid domain.
The GPIHBP1 was bound to the carboxy-terminal domain of
LPL and, as expected, contained a classic LU fold with three
loops stabilized by five disulfide bonds (Leth et al., 2019). Elec-
tron densities corresponding to GPIHBP1’s acidic domain could
not be identified in the GPIHBP1-LPL crystal structures, which
was not surprising, given that the acidic domain is disordered
and binds LPL in a transient and dynamic fashion (Kristensen
et al., 2018).
The interface between GPIHBP1 and LPL is large (940 Å2)
and mediated largely by hydrophobic contacts (Birrane et al.,
2019). Consistent with results from mutagenesis and HDX-MS
studies (Beigneux et al., 2011, 2015; Mysling et al., 2016a),
Trp-109 and adjacent residues in the second finger of GPIHBP1
were directly involved in the binding interface. Extensive se-
quences within the carboxy-terminal domain of LPL, including
Cell Metabolism
Review
Figure 5. Three-Dimensional Structure of the Human LPL-GPIHBP1
Complex
The crystal structure of LPL as a surface representation colored by its elec-
trostatic potential (acidic [red], neutral [white], and basic [blue]) and human
GPIHBP1 in a cartoon representation with b strands colored green (modified
from the crystal structure by Birrane and coworkers [Birrane et al., 2019]). (A)
Electrostatic potential map, revealing the binding of GPIHBP1’s LU domain to
the carboxy- terminal domain (CTD) of LPL.
(B) Electrostatic potential map (rotated 90 degrees compared to A), revealing a
large basic patch one surface of LPL (blue), spanning both the amino-terminal
domain (NTD) and the CTD. The amino- and carboxy-terminal regions of
GPIHBP1 are marked by N and C, respectively.
(C) A model of GPIHBP1 in solution, as defined by SAXS analyses, illustrating
the large space (112 Å) occupied by GPIHBP1’s intrinsically disordered acidic
domain (modified from Kristensen et al., 2018). Different ensembles from the
modeling of the disordered acidic domain are illustrated by colored spheres,
while the folded LU domain (located below) is represented by a cartoon rep-
resentation. The acidic domain of GPIHBP1 is absent in the models shown in
(A) and (B) because the acidic domain is disordered and was not visualized in
the crystal structure.
(D) Schematic representation of the GPIHBP1-LPL complex, depicting a po-
tential interaction between GPIHBP1’s intrinsically disordered acidic domain
(the semi-transparent ‘‘red cloud’’) and LPL’s large basic patch (blue).
sequences that had been implicated by mutagenesis and HDX-
MS studies (Hu et al., 2017b; Mysling et al., 2016a; Voss et al.,
2011), participated in the GPIHBP1-LPL binding interface.
With the GPIHBP1-LPL crystal structure in hand, one of our
early questions was whether the binding site for GPIHBP1 on
LPL might resemble the binding site for colipase on pancreatic
lipase (PL). Like GPIHBP1, colipase is a small, multi-fingered
cysteine-knot protein stabilized by five disulfide bonds (van
Tilbeurgh et al., 1992). A comparison of crystal structures re-
vealed that GPIHBP1 and colipase bind to opposite sides of
the carboxy-terminal domains of LPL and PL, respectively (Bir-
rane et al., 2019). Also, colipase-PL interactions were discrete
and mediated by electrostatic interactions (van Tilbeurgh et al.,
1992), whereas GPIHBP1-LPL interactions were extensive and
driven primarily by hydrophobic contacts (Birrane et al., 2019).
In hindsight, the existence of distinct binding sites for GPIHBP1
and colipase was not surprising. While the two proteins have su-
perficial similarities, they are evolutionarily distinct and play very
different functions, with colipase serving to activate triglyceride
hydrolysis and GPIHBP1 serving to transport LPL and anchor it
to capillary endothelial cells.
LPL Contains a Single Large Basic Patch
LPL contains several positively charged heparin-binding motifs,
scattered within the primary sequence, and their functional
importance has been investigated by several groups (Hata
et al., 1993; Hill et al., 1998; Ma et al., 1994; Sendak et al.,
1998). Interestingly, an electrostatic surface potential map,
based on the GPIHBP1-LPL crystal structure, revealed that the
various heparin-binding motifs converge to create a single large
contiguous basic patch (covering 2,400 Å2) (Birrane et al.,
2019) (Figures 5A and 5B). This patch extends into both the
amino- and carboxy-terminal domains of LPL and spans the nar-
row hinge region connecting the two domains. Of note, two
lysines (K472 and K473) located within the last five residues of
LPL were not visualized in the electron density map (Birrane
et al., 2019). Had they been visualized, the size of LPL’s basic
patch would have appeared even larger.
As noted earlier, GPIHBP1’s intrinsically disordered acidic
domain increases the initial encounter rate with LPL and stabi-
lizes LPL’s structure. These effects are mediated by transient in-
teractions (Kristensen et al., 2018; Mysling et al., 2016a, 2016b),
resulting in the formation of a ‘‘fuzzy complex’’ (Borgia et al.,
2018; Ivic et al., 2019). Given the transient nature of the interac-
tions, it was not surprising that the acidic domain could not be
visualized in the crystal structure. However, it is noteworthy
that GPIHBP1 was positioned such that its acidic domain,
>60 Å in length (Kristensen et al., 2018), would be expected to
project toward—and potentially interact with—the entirety of
LPL’s basic patch (Figures 5C and 5D). It seems likely that these
interactions stabilize LPL structure and activity (Birrane et al.,
2019). GPIHBP1’s acidic domain could provide a ‘‘backbone’’
spanning LPL’s amino- and carboxy-terminal domains—one
that provides sufficient flexibility for enzyme function but also
the rigidity required to prevent the unfolding that might otherwise
be initiated by torsion of LPL’s hinge region (or collapse of LPL’s
two domains upon each other).
Structural Insights into LPL Mutations Causing
Chylomicronemia
Because the interface between GPIHBP1’s LU domain and
the carboxy-terminal domain of LPL is large, Birrane and
coworkers predicted that it would be possible to identify muta-
tions that would disrupt GPIHBP1-LPL interactions (Birrane
et al., 2019). They were intrigued by a missense mutation in
LPL, M404R, identified in a Middle Eastern patient with
Cell Metabolism 30, July 2, 2019 59

chylomicronemia. That mutation was reported to abolish LPL
secretion from cells (Pingitore et al., 2016), but the crystal struc-
ture revealed that M404 is located within the GPIHBP1-LPL
binding interface—implying that the mutation might actually
cause disease by interfering with LPL’s ability to bind to
GPIHBP1. Indeed, GPIHBP1-LPL binding assays revealed that
the M404R mutation abolishes LPL binding to GPIHBP1 but
does not affect LPL secretion from cells (Birrane et al., 2019).
The GPIHBP1-LPL crystal structure also yielded insights into a
D201V mutation in LPL, identified in two Lebanese families with
chylomicronemia (Abifadel et al., 2004). The crystal structure re-
vealed that D201 participates in the coordination of LPL’s sole
calcium ion. A valine cannot fulfill that function. Birrane and co-
workers suspected that the inability of LPL-D201V to coordinate
calcium would interfere with LPL folding and abolish LPL
secretion from cells. Indeed, when LPL-D201V was introduced
into CHO cells, the mutant LPL was produced in normal
amounts, but none was secreted from cells, providing a clear
explanation for the chylomicronemia phenotype. A mutation in
D202, another residue involved in coordinating the calcium ion,
also abolished LPL secretion from cells (Birrane et al., 2019).
Insights from the GPIHBP1-LPL crystal structure, along with
straightforward biochemical assays, should make it possible to
understand and classify most LPL mutations associated with
chylomicronemia.
Relevance of the Head-to-Tail LPL Homodimer
Conformation in the Crystal Structure
The fact that the crystal structure revealed two LPL molecules in
a head-to-tail orientation appeared, at least at first glance, to
support long-held assumptions about the structure of LPL. For
decades, the functional unit of active LPL has been assumed
to be a homodimer (Berryman and Bensadoun, 1995; Garfinkel
et al., 1983; Iverius and Ostlund-Lindqvist, 1976; Krapp et al.,
1995; Lookene and Bengtsson-Olivecrona, 1993; Lookene
et al., 1994; Olivecrona and Bengtsson-Olivecrona, 1990; Olive-
crona et al., 1985; Osborne et al., 1985; Peterson et al., 2002;
Vannier and Ailhaud, 1989; Wong et al., 1994; Zhang et al.,
2005), and homodimer assembly has been considered essential
for LPL secretion from cells (Doolittle et al., 2010; Doolittle and
Péterfy, 2010). Several studies concluded that active LPL homo-
dimers have a head-to-tail orientation (Kobayashi et al., 2002;
Wong et al., 1997), with the Trp-rich loop in the carboxy-terminal
domain of one monomer supplying substrates to the catalytic
pocket in the amino-terminal domain of the partner monomer.
However, in the crystal structures for the GPIHBP1-LPL complex
(Birrane et al., 2019; Horton, 2019; Arora et al., 2019), the Trp-
rich loop in the carboxy-terminal domain of one monomer inter-
acted with the catalytic pocket of the partner monomer, raising
doubts about whether that particular conformation would be
compatible with either lipoprotein binding or the hydrolysis of
triglyceride substrates. One possibility is that the reciprocal in-
teractions between the two LPL monomers do not represent
physiologic interactions and instead reflect a stable conforma-
tion that LPL adopts at high protein concentrations—one in
which the catalytic pocket shields the hydrophobic Trp-rich
loop from the aqueous environment. It is important to empha-
size, however, that all of the GPIHBP1-LPL crystal structures
were generated in the absence of triglyceride substrates or lipo-
60 Cell Metabolism 30, July 2, 2019
Cell Metabolism
Review
proteins. The conformation of the GPIHBP1-LPL complex in the
presence of lipoproteins remains to be determined.
In recent studies, Beigneux and coworkers revisited accepted
dogma holding that LPL is only active as a homodimer (Beigneux
et al., 2019). When they subjected freshly secreted human LPL to
density gradient ultracentrifugation, they found that the LPL
mass and activity peaks corresponded to the size of monomers
(55 kDa)—overlapping with or slightly smaller than the 66-kDa
albumin standard and with minimal overlap with a 97-kDa phos-
phorylase b standard. Incubating the LPL with guanidine hydro-
chloride inactivated catalytic activity but did not alter the
apparent size of LPL, as judged by density gradient ultracentrifu-
gation. Also, the GPIHBP1-LPL complex, expected to have a
size of 75 kDa, was slightly larger than the albumin standard
by density gradient ultracentrifugation, with minimal overlap
with the phosphorylase b standard (Beigneux et al., 2019).
When LPL is loaded onto a heparin-Sepharose column and
then eluted with a linear gradient of sodium chloride, there
are two distinct LPL peaks—a low-salt peak with little catalytic
activity and a high-salt peak containing large amounts of cata-
lytic activity. For years, the low-salt peak was presumed to
contain inactive LPL monomers, whereas the high-salt peak
was assumed to contain active homodimers. However, when
Beigneux and coworkers subjected the active LPL in the high-
salt peak to density gradient ultracentrifugation, they found
that it exhibited the size of a monomer (i.e., slightly smaller
than the albumin standard) (Beigneux et al., 2019). Most of the
LPL in the low-salt peak was inactive and in the form of aggre-
gates at the bottom of the density gradient tube. Beigneux and
coworkers also examined the LPL in the low- and high-salt peaks
with a sandwich ELISA in which a single LPL-specific mono-
clonal antibody (either 5D2 or 88B8) was used to both capture
and detect the LPL. This single-antibody sandwich ELISA format
had been touted as being specific for catalytically active LPL
homodimers (Peterson et al., 2002). However, Beigneux and
coworkers found that the catalytically active LPL in the high-
salt peak exhibited little reactivity in single-antibody sandwich
ELISAs, whereas the aggregate-containing LPL species in the
low-salt peak yielded a robust signal (Beigneux et al., 2019).
Importantly, the LPL in both the low- and high-salt fractions
yielded robust signals in sandwich ELISAs employing two
different monoclonal antibodies (one to capture the LPL and
the other to detect the LPL). The two-antibody sandwich ELISAs
are sensitive in detecting LPL monomers as well as aggregated
LPL species. The findings by Beigneux and coworkers implied
that the catalytically active LPL in the high-salt peak, long pre-
sumed to be in the form of homodimers, is largely in the form
of monomers (Beigneux et al., 2019).
Beigneux and coworkers also tested whether CHO cells, when
co-transfected with two LPL constructs harboring different
epitope tags, would secrete catalytically active ‘‘mixed homo-
dimers’’ (i.e., LPL species containing both epitope tags)
(Beigneux et al., 2019). The medium of the co-transfected cells
contained large amounts of both differentially tagged LPL pro-
teins, and both LPL proteins were catalytically active. However,
the amount of mixed LPL species in the medium was very low, as
judged by specific ELISAs, but still detectable. Interestingly,
the mixed species in the medium were primarily located in
the inactive low-salt peak, as judged by heparin-Sepharose
Cell Metabolism
Review
chromatography, not in the catalytically active high-salt peak
(Beigneux et al., 2019). These studies provided further support
for the notion that freshly secreted, catalytically active LPL is pri-
marily in the form of monomers. The concept that LPL can be
active as a monomer (Beigneux et al., 2019) was endorsed by
Arora and coworkers (Arora et al., 2019), based in part on their
suspicion that the head-to-tail LPL homodimers in their crystal
structure were unlikely to be physiologically active in binding li-
poproteins or hydrolyzing triglycerides.
The view that LPL can be active as a monomer is not inconsis-
tent with the existence of LPL homodimers in specific settings.
LPL homodimers can exist, as documented by the crystallog-
raphy and SAXS studies (Arora et al., 2019; Birrane et al.,
2019), but the relevance of those conformations to the physio-
logic function of LPL is definitely open to question. It is important
to emphasize that the LPL homodimers in the crystallization and
SAXS studies were observed when the protein concentrations
were high, whereas the density gradient and immunochemical
studies supporting the existence of active LPL monomers were
performed at far lower concentrations of LPL (Beigneux et al.,
2019). It is possible that LPL homodimers that appear at high
protein concentrations dissociate into active monomers at lower
protein concentrations. However, at this point, the existence of
stable and catalytically active LPL homodimers appears ques-
tionable. Moreover, dogma that has accompanied the notion
that LPL is active only when it is in the form of homodimers
may need additional scrutiny. For example, the notion that
ANGPTL4 functions by converting catalytically active LPL homo-
dimers into inactive LPL monomers (Sukonina et al., 2006) may
need to be revisited. Also, the concept that lipase maturation
factor 1 (LMF-1) (an integral membrane protein of the ER) func-
tions to convert newly synthesized LPL monomers into catalyti-
cally active and secretion-competent homodimers (Doolittle
et al., 2010; Doolittle and Péterfy, 2010) may also require addi-
tional scrutiny.
Concluding Remarks
Studies over the past decade have established that GPIHBP1 is
LPL’s partner—required for transporting LPL into capillaries, li-
poprotein margination, and for preserving LPL structure and ac-
tivity. Impaired GPIHBP1-LPL interactions, whether from genetic
defects or autoantibodies, cause chylomicronemia. Biophysical
studies have demonstrated that GPIHBP1’s acidic domain stabi-
lizes LPL structure and activity, likely by creating a fuzzy complex
with LPL’s large basic patch. For years, the properties of LPL
have been analyzed primarily with purified preparations of LPL,
but future research will need to focus on the properties of the
GPIHBP1-LPL complex. We need a better understanding of lipo-
protein interactions with the GPIHBP1-LPL complex and how
regulators of lipolysis (e.g., apo-CII and apo-AV) influence the
activity of the GPIHBP1-LPL complex. We also need a clear un-
derstanding of how the ‘‘ANGPTL proteins’’ initiate unfolding of
LPL and a more precise molecular understanding of the interac-
tions between GPIHBP1’s acidic domain and LPL’s basic patch.
Finally, commonly occurring LPL polymorphisms are known to
influence plasma triglyceride levels (Almeida et al., 2007; Jensen
et al., 2009; Rip et al., 2006); the impact of those polymorphisms
needs to be investigated in the context of the GPIHBP1-LPL
complex.
ACKNOWLEDGMENTS
Supported by grants (HL090553, HL087228, and HL125335) from the National
Heart, Lung, and Blood Institute; a Transatlantic Network grant (12CVD04)
from the Leducq Foundation; and Novo Nordisk Foundation grants
(NNF17OC0026868 and NNF18OC0033864).
DECLARATION OF INTERESTS
M.M. is an employee of Shire/Takeda and holds stock and stock options in the
company. K.N. holds stock in Immunobiologic Laboratories and serves as a
consultant for Skylight and Sysmex. S.G.Y., A.P.B., L.G.F., and K.N. and their
respective institutions have applied for a patent on an assay for a diagnostic
test for the GPIHBP1 autoantibody syndrome.
REFERENCES
Abifadel, M., Jambart, S., Allard, D., Rabès, J.P., Varret, M., Derré, A., Chou-
ery, E., Salem, N., Junien, C., Aydénian, H., et al. (2004). Identification of the
first Lebanese mutation in the LPL gene and description of a rapid detection
method. Clin. Genet. 65, 158–161.
Acton, S., Rigotti, A., Landschulz, K.T., Xu, S., Hobbs, H.H., and Krieger, M.
(1996). Identification of scavenger receptor SR-BI as a high density lipoprotein
receptor. Science 271, 518–520.
Allan, C.M., Larsson, M., Hu, X., He, C., Jung, R.S., Mapar, A., Voss, C., Miya-
shita, K., Machida, T., Murakami, M., et al. (2016). An LPL-specific monoclonal
antibody, 88B8, that abolishes the binding of LPL to GPIHBP1. J. Lipid Res. 57,
1889–1898.
Allan, C.M., Jung, C.J., Larsson, M., Heizer, P.J., Tu, Y., Sandoval, N.P., Dang,
T.L.P., Jung, R.S., Beigneux, A.P., de Jong, P.J., et al. (2017a). Mutating a
conserved cysteine in GPIHBP1 reduces amounts of GPIHBP1 in capillaries
and abolishes LPL binding. J. Lipid Res. 58, 1453–1461.
Allan, C.M., Larsson, M., Jung, R.S., Ploug, M., Bensadoun, A., Beigneux,
A.P., Fong, L.G., and Young, S.G. (2017b). Mobility of "HSPG-bound" LPL ex-
plains how LPL is able to reach GPIHBP1 on capillaries. J. Lipid Res. 58,
216–225.
Allan, C.M., Heizer, P.J., Tu, Y., Sandoval, N.P., Jung, R.S., Morales, J.E., Sajti,
E., Troutman, T.D., Saunders, T.L., Cusanovich, D.A., et al. (2019). An up-
stream enhancer regulates Gpihbp1 expression in a tissue-specific manner.
J. Lipid Res. 60, 869–879.
Almeida, K.A., Strunz, C.M., Maranhão, R.C., and Mansur, A.P. (2007). The
S447X polymorphism of lipoprotein lipase: effect on the incidence of prema-
ture coronary disease and on plasma lipids. Arq. Bras. Cardiol. 88, 297–303.
Ariza, M.J., Martı́nez-Hernández, P.L., Ibarretxe, D., Rabacchi, C., Rioja, J.,
Grande-Aragón, C., Plana, N., Tarugi, P., Olivecrona, G., Calandra, S., et al.
(2016). Novel mutations in the GPIHBP1 gene identified in 2 patients with
recurrent acute pancreatitis. J. Clin. Lipidol. 10, 92–100.
Arora, R., Nimonkar, A.V., Baird, D., Wang, C., Chiu, C.H., Horton, P.A., Han-
rahan, S., Cubbon, R., Weldon, S., Tschantz, W.R., et al. (2019). Structure of
lipoprotein lipase in complex with GPIHBP1. Proc. Natl. Acad. Sci. U.S.A.
116, 10360–10365.
Banchereau, J., and Pascual, V. (2006). Type I interferon in systemic lupus
erythematosus and other autoimmune diseases. Immunity 25, 383–392.
Beigneux, A.P. (2010). GPIHBP1 and the processing of triglyceride-rich lipo-
proteins. J. Clin. Lipidol. 5, 575–582.
Beigneux, A.P., Davies, B.S., Gin, P., Weinstein, M.M., Farber, E., Qiao, X.,
Peale, F., Bunting, S., Walzem, R.L., Wong, J.S., et al. (2007). Glycosylphos-
phatidylinositol-anchored high density lipoprotein–binding protein 1 plays a
critical role in the lipolytic processing of chylomicrons. Cell Metab. 5, 279–291.
Beigneux, A.P., Franssen, R., Bensadoun, A., Gin, P., Melford, K., Peter, J.,
Walzem, R.L., Weinstein, M.M., Davies, B.S., Kuivenhoven, J.A., et al.
(2009a). Chylomicronemia with a mutant GPIHBP1 (Q115P) that cannot bind
lipoprotein lipase. Arterioscler. Thromb. Vasc. Biol. 29, 956–962.
Beigneux, A.P., Gin, P., Davies, B.S.J., Weinstein, M.M., Bensadoun, A., Fong,
L.G., and Young, S.G. (2009b). Highly conserved cysteines within the Ly6
Cell Metabolism 30, July 2, 2019 61

domain of GPIHBP1 are crucial for the binding of lipoprotein lipase. J. Biol.
Chem. 284, 30240–30247.
Beigneux, A.P., Davies, B.S., Tat, S., Chen, J., Gin, P., Voss, C.V., Weinstein,
M.M., Bensadoun, A., Pullinger, C.R., Fong, L.G., et al. (2011). Assessing the
role of the glycosylphosphatidylinositol-anchored high density lipoprotein-
binding protein 1 (GPIHBP1) three-finger domain in binding lipoprotein lipase.
J. Biol. Chem. 286, 19735–19743.
Beigneux, A.P., Fong, L.G., Bensadoun, A., Davies, B.S., Oberer, M., Gårds-
voll, H., Ploug, M., and Young, S.G. (2015). GPIHBP1 missense mutations
often cause multimerization of GPIHBP1 and thereby prevent lipoprotein
lipase binding. Circ. Res. 116, 624–632.
Beigneux, A.P., Miyashita, K., Ploug, M., Blom, D.J., Ai, M., Linton, M.F., Kho-
vidhunkit, W., Dufour, R., Garg, A., McMahon, M.A., et al. (2017). Autoanti-
bodies against GPIHBP1 as a cause of hypertriglyceridemia. N. Engl. J.
Med. 376, 1647–1658.
Beigneux, A.P., Allan, C.M., Sandoval, N.P., Cho, G.W., Heizer, P.J., Jung,
R.S., Stanhope, K.L., Havel, P.J., Birrane, G., Meiyappan, M., et al. (2019). Li-
poprotein lipase is active as a monomer. Proc. Natl. Acad. Sci. U.S.A. 116,
6319–6328.
Bengtsson, G., and Olivecrona, T. (1979). Apolipoprotein CII enhances hydro-
lysis of monoglycerides by lipoprotein lipase, but the effect is abolished by
fatty acids. FEBS Lett. 106, 345–348.
Berryman, D.E., and Bensadoun, A. (1995). Heparan sulfate proteoglycans are
primarily responsible for the maintenance of enzyme activity, binding, and
degradation of lipoprotein lipase in Chinese hamster ovary cells. J. Biol.
Chem. 270, 24525–24531.
Birrane, G., Beigneux, A.P., Dwyer, B., Strack-Logue, B., Kristensen, K.K.,
Francone, O.L., Fong, L.G., Mertens, H.D.T., Pan, C.Q., Ploug, M., et al.
(2019). Structure of the lipoprotein lipase-GPIHBP1 complex that mediates
plasma triglyceride hydrolysis. Proc. Natl. Acad. Sci. U.S.A. 116, 1723–1732.
Bishop, J.R., Passos-Bueno, M.R., Fong, L., Stanford, K.I., Gonzales, J.C.,
Yeh, E., Young, S.G., Bensadoun, A., Witztum, J.L., Esko, J.D., et al. (2010).
Deletion of the basement membrane heparan sulfate proteoglycan type XVIII
collagen causes hypertriglyceridemia in mice and humans. PLoS One 5,
e13919.
Blanchette-Mackie, E.J., Masuno, H., Dwyer, N.K., Olivecrona, T., and Scow,
R.O. (1989). Lipoprotein lipase in myocytes and capillary endothelium of heart:
immunocytochemical study. Am. J. Physiol. 256, E818–E828.
Borgia, A., Borgia, M.B., Bugge, K., Kissling, V.M., Heidarsson, P.O., Fer-
nandes, C.B., Sottini, A., Soranno, A., Buholzer, K.J., Nettels, D., et al.
(2018). Extreme disorder in an ultrahigh-affinity protein complex. Nature
555, 61–66.
Breckenridge, W.C., Little, J.A., Steiner, G., Chow, A., and Poapst, M. (1978).
Hypertriglyceridemia associated with deficiency of apolipoprotein C-II.
N. Engl. J. Med. 298, 1265–1273.
Charrière, S., Peretti, N., Bernard, S., Di Filippo, M., Sassolas, A., Merlin, M.,
Delay, M., Debard, C., Lefai, E., Lachaux, A., et al. (2011). GPIHBP1 C89F
neomutation and hydrophobic C-terminal domain G175R mutation in two ped-
igrees with severe hyperchylomicronemia. J. Clin. Endocrinol. Metab. 96,
E1675–E1679.
Chi, X., Shetty, S.K., Shows, H.W., Hjelmaas, A.J., Malcolm, E.K., and Davies,
B.S. (2015). Angiopoietin-like 4 modifies the interactions between lipoprotein
lipase and its endothelial cell transporter GPIHBP1. J. Biol. Chem. 290,
11865–11877.
Chyzhyk, V., Kozmic, S., Brown, A.S., Hudgins, L.C., Starc, T.J., Davila, A.D.,
Blevins, T.C., Diffenderfer, M.R., He, L., Geller, A.S., et al. (2019). Extreme hy-
pertriglyceridemia: genetic diversity, pancreatitis, pregnancy, and prevalence.
J. Clin. Lipidol. 13, 89–99.
Coca-Prieto, I., Kroupa, O., Gonzalez-Santos, P., Magne, J., Olivecrona, G.,
Ehrenborg, E., and Valdivielso, P. (2011). Childhood-onset chylomicronaemia
with reduced plasma lipoprotein lipase activity and mass: identification of a
novel GPIHBP1 mutation. J. Intern. Med. 270, 224–228.
Davies, B.S., Beigneux, A.P., Barnes, R.H., Tu, Y., Gin, P., Weinstein, M.M.,
Nobumori, C., Nyrén, R., Goldberg, I., Olivecrona, G., et al. (2010). GPIHBP1
is responsible for the entry of lipoprotein lipase into capillaries. Cell Metab.
12, 42–52.
62 Cell Metabolism 30, July 2, 2019
Cell Metabolism
Review
Davies, B.S., Goulbourne, C.N., Barnes, R.H., 2nd, Turlo, K.A., Gin, P.,
Vaughan, S., Vaux, D.J., Bensadoun, A., Beigneux, A.P., Fong, L.G., et al.
(2012). Assessing mechanisms of GPIHBP1 and lipoprotein lipase movement
across endothelial cells. J. Lipid. Res. 53, 2690–2697.
de Beer, F., Hendriks, W.L., van Vark, L.C., Kamerling, S.W., van Dijk, K.W.,
Hofker, M.H., Smelt, A.H., and Havekes, L.M. (1999). Binding of beta-VLDL
to heparan sulfate proteoglycans requires lipoprotein lipase, whereas ApoE
only modulates binding affinity. Arterioscler. Thromb. Vasc. Biol. 19, 633–637.
Dewey, F.E., Gusarova, V., O’Dushlaine, C., Gottesman, O., Trejos, J., Hunt,
C., Van Hout, C.V., Habegger, L., Buckler, D., Lai, K.M., et al. (2016). Inactivat-
ing variants in ANGPTL4 and risk of coronary artery disease. N. Engl. J. Med.
374, 1123–1133.
Di Domizio, J., and Cao, W. (2013). Fueling autoimmunity: type I interferon in
autoimmune diseases. Expert Rev. Clin. Immunol. 9, 201–210.
Dijk, W., Beigneux, A.P., Larsson, M., Bensadoun, A., Young, S.G., and Ker-
sten, S. (2016). Angiopoietin-like 4 promotes intracellular degradation of lipo-
protein lipase in adipocytes. J. Lipid. Res. 57, 1670–1683.
Doolittle, M.H., and Péterfy, M. (2010). Mechanisms of lipase maturation.
J. Clin. Lipidol. 5, 71–85.
Doolittle, M.H., Ehrhardt, N., and Péterfy, M. (2010). Lipase maturation factor
1: structure and role in lipase folding and assembly. Curr. Opin. Lipidol. 21,
198–203.
Duchesne, L., Octeau, V., Bearon, R.N., Beckett, A., Prior, I.A., Lounis, B., and
Fernig, D.G. (2012). Transport of fibroblast growth factor 2 in the pericellular
matrix is controlled by the spatial distribution of its binding sites in heparan sul-
fate. PLoS Biol. 10, e1001361.
Eguchi, J., Miyashita, K., Fukamachi, I., Nakajima, K., Murakami, M., Kawa-
hara, Y., Yamashita, T., Ohta, Y., Abe, K., Nakatsuka, A., et al. (2019). GPIHBP1
autoantibody syndrome during interferon beta1a treatment. J. Clin. Lipidol.
13, 62–69.
Emmerich, J., Beg, O.U., Peterson, J., Previato, L., Brunzell, J.D., Brewer,
H.B., Jr., and Santamarina-Fojo, S. (1992). Human lipoprotein lipase. Analysis
of the catalytic triad by site-directed mutagenesis of Ser-132, Asp-156, and
His-241. J. Biol. Chem. 267, 4161–4165.
Enerba€ck, S., Semb, H., Bengtsson-Olivecrona, G., Carlsson, P., Hermans-
son, M.L., Olivecrona, T., and Bjursell, G. (1987). Molecular cloning and
sequence analysis of cDNA encoding lipoprotein lipase of guinea pig. Gene
58, 1–12.
Ferguson, M.A.J., Kinoshita, T., and Hart, G.W. (2009). Glycosylphosphatidy-
linositol anchors. In Essentials of Glycobiology, A. Varki, R.D. Cummings, J.D.
Esko, H.H. Freeze, P. Stanley, C.R. Bertozzi, G.W. Hart, and M.E. Etzler, eds.
(Cold Spring Harbor).
Fong, L.G., Young, S.G., Beigneux, A.P., Bensadoun, A., Oberer, M., Jiang, H.,
and Ploug, M. (2016). GPIHBP1 and plasma triglyceride metabolism. Trends
Endocrinol. Metab. 27, 455–469.
Franssen, R., Young, S.G., Peelman, F., Hertecant, J., Sierts, J.A., Schimmel,
A.W.M., Bensadoun, A., Kastelein, J.J.P., Fong, L.G., Dallinga-Thie, G.M.,
et al. (2010). Chylomicronemia with low postheparin lipoprotein lipase levels
in the setting of GPIHBP1 defects. Circ. Cardiovasc. Genet. 3, 169–178.
Garfinkel, A.S., Kempner, E.S., Ben-Zeev, O., Nikazy, J., James, S.J., and
Schotz, M.C. (1983). Lipoprotein lipase: size of the functional unit determined
by radiation inactivation. J. Lipid Res. 24, 775–780.
Gin, P., Yin, L., Davies, B.S., Weinstein, M.M., Ryan, R.O., Bensadoun, A.,
Fong, L.G., Young, S.G., and Beigneux, A.P. (2008). The acidic domain of
GPIHBP1 is important for the binding of lipoprotein lipase and chylomicrons.
J. Biol. Chem. 283, 29554–29562.
Gin, P., Beigneux, A.P., Voss, C., Davies, B.S., Beckstead, J.A., Ryan, R.O.,
Bensadoun, A., Fong, L.G., and Young, S.G. (2011). Binding preferences for
GPIHBP1, a glycosylphosphatidylinositol-anchored protein of capillary endo-
thelial cells. Arterioscler. Thromb. Vasc. Biol. 31, 176–182.
Gin, P., Goulbourne, C.N., Adeyo, O., Beigneux, A.P., Davies, B.S., Tat, S.,
Voss, C.V., Bensadoun, A., Fong, L.G., and Young, S.G. (2012). Chylomicrone-
mia mutations yield new insights into interactions between lipoprotein lipase
and GPIHBP1. Hum. Mol. Genet. 21, 2961–2972.
Cell Metabolism
Review
Gonzaga-Jauregui, C., Mir, S., Penney, S., Jhangiani, S., Midgen, C., Finegold,
M., Muzny, D.M., Wang, M., Bacino, C.A., Gibbs, R.A., et al. (2014). Whole-
exome sequencing reveals GPIHBP1 mutations in infantile colitis with severe
hypertriglyceridemia. J. Pediatr. Gastroenterol. Nutr. 59, 17–21.
Goulbourne, C.N., Gin, P., Tatar, A., Nobumori, C., Hoenger, A., Jiang, H.,
Grovenor, C.R., Adeyo, O., Esko, J.D., Goldberg, I.J., et al. (2014). The
GPIHBP1-LPL complex is responsible for the margination of triglyceride-rich
lipoproteins in capillaries. Cell Metab. 19, 849–860.
Gutgsell, A.R., Ghodge, S.V., Bowers, A.A., and Neher, S.B. (2019). Mapping
the sites of the lipoprotein lipase (LPL)-angiopoietin-like protein 4 (ANGPTL4)
interaction provides mechanistic insight into LPL inhibition. J. Biol. Chem. 294,
2678–2689.
Haller, J.F., Mintah, I.J., Shihanian, L.M., Stevis, P., Buckler, D., Alexa-Braun,
C.A., Kleiner, S., Banfi, S., Cohen, J.C., Hobbs, H.H., et al. (2017). ANGPTL8
requires ANGPTL3 to inhibit lipoprotein lipase and plasma triglyceride clear-
ance. J. Lipid. Res. 58, 1166–1173.
Hata, A., Ridinger, D.N., Sutherland, S., Emi, M., Shuhua, Z., Myers, R.L., Ren,
K., Cheng, T., Inoue, I., and Wilson, D.E. (1993). Binding of lipoprotein lipase to
heparin. Identification of five critical residues in two distinct segments of the
amino-terminal domain. J. Biol. Chem. 268, 8447–8457.
Havel, R.J. (2010). Triglyceride-rich lipoproteins and plasma lipid transport.
Arterioscler. Thromb. Vasc. Biol. 30, 9–19.
Havel, R.J., and Gordon, R.S., Jr. (1960). Idiopathic hyperlipemia: metabolic
studies in an affected family. J. Clin. Invest. 39, 1777–1790.
Havel, R.J., and Kane, J.P. (2001). Introduction: structure and metabolism of
plasma lipoproteins. In The metabolic and molecular bases of inherited dis-
ease, C.R. Scriver, A.L. Beaudet, W.S. Sly, D. Valle, B. Childs, K.W. Kinzler,
and B. Vogelstein, eds. (McGraw-Hill), pp. 2705–2716.
Havel, R.J., Shore, V.G., Shore, B., and Bier, D.M. (1970). Role of specific gly-
copeptides of human serum lipoproteins in the activation of lipoprotein lipase.
Circ. Res. 27, 595–600.
He, C., Weston, T.A., Jung, R.S., Heizer, P., Larsson, M., Hu, X., Allan, C.M.,
Tontonoz, P., Reue, K., Beigneux, A.P., et al. (2018a). NanoSIMS analysis of
intravascular lipolysis and lipid movement across capillaries and into cardio-
myocytes. Cell Metab. 27, 1055–1066.e3.
He, L., Vanlandewijck, M., Ma €e, M.A., Andrae, J., Ando, K., Del Gaudio, F., Na-
har, K., Lebouvier, T., Laviña, B., Gouveia, L., et al. (2018b). Single-cell RNA
sequencing of mouse brain and lung vascular and vessel-associated cell
types. Sci. Data 5.
Helgadottir, A., Gretarsdottir, S., Thorleifsson, G., Hjartarson, E., Sigurdsson,
A., Magnusdottir, A., Jonasdottir, A., Kristjansson, H., Sulem, P., Oddsson, A.,
et al. (2016). Variants with large effects on blood lipids and the role of choles-
terol and triglycerides in coronary disease. Nat. Genet. 48, 634–639.
Henderson, H.E., Hassan, F., Marais, D., and Hayden, M.R. (1996). A new mu-
tation destroying disulphide bridging in the C-terminal domain of lipoprotein
lipase. Biochem. Biophys. Res. Commun. 227, 189–194.
Henderson, H., Leisegang, F., Hassan, F., Hayden, M., and Marais, D. (1998). A
novel Glu421Lys substitution in the lipoprotein lipase gene in pregnancy-
induced hypertriglyceridemic pancreatitis. Clin. Chim. Acta 269, 1–12.
Hill, J.S., Yang, D., Nikazy, J., Curtiss, L.K., Sparrow, J.T., and Wong, H.
(1998). Subdomain chimeras of hepatic lipase and lipoprotein lipase. Localiza-
tion of heparin and cofactor binding. J. Biol. Chem. 273, 30979–30984.
Horton, J.D. (2019). Intravascular triglyceride lipolysis becomes crystal clear.
Proc. Natl. Acad. Sci. U.S.A. 116, 1480–1482.
Hu, X., Dallinga-Thie, G.M., Hovingh, G.K., Chang, S.Y., Sandoval, N.P., Dang,
T.L.P., Fukamachi, I., Miyashita, K., Nakajima, K., Murakami, M., et al. (2017a).
GPIHBP1 autoantibodies in a patient with unexplained chylomicronemia.
J. Clin. Lipidol. 11, 964–971.
Hu, X., Sleeman, M.W., Miyashita, K., Linton, M.F., Allan, C.M., He, C., Lars-
son, M., Tu, Y., Sandoval, N.P., Jung, R.S., et al. (2017b). Monoclonal anti-
bodies that bind to the Ly6 domain of GPIHBP1 abolish the binding of LPL.
J. Lipid Res. 58, 208–215.
Huijbers, M.G., Plomp, J.J., van der Maarel, S.M., and Verschuuren, J.J.
(2018). IgG4-mediated autoimmune diseases: a niche of antibody-mediated
disorders. Ann. N. Y. Acad. Sci. 1413, 92–103.
Ioka, R.X., Kang, M.J., Kamiyama, S., Kim, D.H., Magoori, K., Kamataki, A., Ito,
Y., Takei, Y.A., Sasaki, M., Suzuki, T., et al. (2003). Expression cloning and
characterization of a novel glycosylphosphatidylinositol-anchored high den-
sity lipoprotein-binding protein, GPI-HBP1. J. Biol. Chem. 278, 7344–7349.
Iverius, P.H., and Ostlund-Lindqvist, A.M. (1976). Lipoprotein lipase from
bovine milk. Isolation procedure, chemical characterization, and molecular
weight analysis. J. Biol. Chem. 251, 7791–7795.
Ivic, N., Potocnjak, M., Solis-Mezarino, V., Herzog, F., Bilokapic, S., and Halic,
M. (2019). Fuzzy interactions form and shape the histone transport complex.
Mol. Cell 73, 1191–1203.
Jensen, M.K., Rimm, E.B., Rader, D., Schmidt, E.B., Sørensen, T.I., Vogel, U.,
Overvad, K., and Mukamal, K.J. (2009). S447X variant of the lipoprotein lipase
gene, lipids, and risk of coronary heart disease in 3 prospective cohort studies.
Am. Heart J. 157, 384–390.
Kawahara, Y., Yamashita, T., Ohta, Y., Sato, K., Tsunoda, K., Takemoto, M.,
Hishikawa, N., Eguchi, J., and Abe, K. (2017). Marked hypertriglyceridemia
induced by interferon-beta1a therapy in a clinically isolated syndrome patient.
J. Neurol. Sci. 373, 144–146.
Kersten, S. (2017). Triglyceride metabolism under attack. Cell Metab. 25,
1209–1210.
Kirchgessner, T.G., Svenson, K.L., Lusis, A.J., and Schotz, M.C. (1987). The
sequence of cDNA encoding lipoprotein lipase. A member of a lipase gene
family. J. Biol. Chem. 262, 8463–8466.
Kobayashi, Y., Nakajima, T., and Inoue, I. (2002). Molecular modeling of the
dimeric structure of human lipoprotein lipase and functional studies of the
carboxyl-terminal domain. Eur. J. Biochem. 269, 4701–4710.
Koneczny, I. (2018). A new classification system for IgG4 autoantibodies.
Front. Immunol. 9, 97.
Korn, E.D. (1955). Clearing factor, a heparin-activated lipoprotein lipase. I.
Isolation and characterization of the enzyme from normal rat heart. J. Biol.
Chem. 215, 1–14.
Korn, E.D., and Quigley, T.W., Jr. (1955). Studies on lipoprotein lipase of rat
heart and adipose tissue. Biochim. Biophys. Acta 18, 143–145.
Kovrov, O., Kristensen, K.K., Larsson, E., Ploug, M., and Olivecrona, G. (2019).
On the mechanism of angiopoietin-like protein 8 for control of lipoprotein
lipase activity. J. Lipid Res. 60, 783–793.
Krapp, A., Zhang, H., Ginzinger, D., Liu, M.S., Lindberg, A., Olivecrona, G.,
Hayden, M.R., and Beisiegel, U. (1995). Structural features in lipoprotein lipase
necessary for the mediation of lipoprotein uptake into cells. J. Lipid Res. 36,
2362–2373.
Kristensen, K.K., Midtgaard, S.R., Mysling, S., Kovrov, O., Hansen, L.B., Skar-
Gislinge, N., Beigneux, A.P., Kragelund, B.B., Olivecrona, G., Young, S.G.,
et al. (2018). A disordered acidic domain in GPIHBP1 harboring a sulfated
tyrosine regulates lipoprotein lipase. Proc. Natl. Acad. Sci. U.S.A. 115,
E6020–E6029.
Lafferty, M.J., Bradford, K.C., Erie, D.A., and Neher, S.B. (2013). Angiopoietin-
like protein 4 inhibition of lipoprotein lipase: evidence for reversible complex
formation. J. Biol. Chem. 288, 28524–28534.
LaRosa, J.C., Levy, R.I., Herbert, P., Lux, S.E., and Fredrickson, D.S. (1970). A
specific apoprotein activator for lipoprotein lipase. Biochem. Biophys. Res.
Commun. 41, 57–62.
Leth, J., Leth-Espensen, K., Kristensen, K., Kumari, A., Winther, A., Young, S.,
and Ploug, M. (2019). Evolution and medical significance of LU domain
containing proteins. Int. J. Mol. Sci. 20, E2760, https://doi.org/10.3390/
ijms20112760.
Lookene, A., and Bengtsson-Olivecrona, G. (1993). Chymotryptic cleavage of
lipoprotein lipase. Identification of cleavage sites and functional studies of the
truncated molecule. Eur. J. Biochem. 213, 185–194.
Lookene, A., Skottova, N., and Olivecrona, G. (1994). Interactions of lipopro-
tein lipase with the active-site inhibitor tetrahydrolipstatin (orlistat). Eur. J. Bio-
chem. 222, 395–403.
Lookene, A., Chevreuil, O., Ostergaard, P., and Olivecrona, G. (1996). Interac-
tion of lipoprotein lipase with heparin fragments and with heparan sulfate: stoi-
chiometry, stabilization, and kinetics. Biochemistry 35, 12155–12163.
Cell Metabolism 30, July 2, 2019 63

Lookene, A., Groot, N.B., Kastelein, J.J., Olivecrona, G., and Bruin, T. (1997a).
Mutation of tryptophan residues in lipoprotein lipase. Effects on stability,
immunoreactivity, and catalytic properties. J. Biol. Chem. 272, 766–772.
Lookene, A., Savonen, R., and Olivecrona, G. (1997b). Interaction of lipopro-
teins with heparan sulfate proteoglycans and with lipoprotein lipase. Studies
by surface plasmon resonance technique. Biochemistry 36, 5267–5275.
Ma, Y., Henderson, H.E., Liu, M.S., Zhang, H., Forsythe, I.J., Clarke-Lewis, I.,
Hayden, M.R., and Brunzell, J.D. (1994). Mutagenesis in four candidate hepa-
rin binding regions (residues 279–282, 291–304, 390–393, and 439–448) and
identification of residues affecting heparin binding of human lipoprotein lipase.
J. Lipid Res. 35, 2049–2059.
Mailly, F., Palmen, J., Muller, D.P., Gibbs, T., Lloyd, J., Brunzell, J., Durrington,
P., Mitropoulos, K., Betteridge, J., Watts, G., et al. (1997). Familial lipoprotein
lipase (LPL) deficiency: a catalogue of LPL gene mutations identified in 20 pa-
tients from the UK, Sweden, and Italy. Hum. Mutat. 10, 465–473.
Mergenthaler, P., Lindauer, U., Dienel, G.A., and Meisel, A. (2013). Sugar for
the brain: the role of glucose in physiological and pathological brain function.
Trends Neurosci. 36, 587–597.
Miyashita, K., Fukamachi, I., Machida, T., Nakajima, K., Young, S.G., Mura-
kami, M., Beigneux, A.P., and Nakajima, K. (2018a). An ELISA for quantifying
GPIHBP1 autoantibodies and making a diagnosis of the GPIHBP1 autoanti-
body syndrome. Clin. Chim. Acta 487, 174–178.
Miyashita, K., Fukamachi, I., Nagao, M., Ishida, T., Kobayashi, J., Machida, T.,
Nakajima, K., Murakami, M., Ploug, M., Beigneux, A.P., et al. (2018b). An
enzyme-linked immunosorbent assay for measuring GPIHBP1 levels in human
plasma or serum. J. Clin. Lipidol. 12, 203–210.
Myocardial Infarction Genetics; CARDIoGRAM Exome Consortia Investiga-
tors, Stitziel, N.O., Stirrups, K.E., Masca, N.G., Erdmann, J., Ferrario, P.G.,
König, I.R., Weeke, P.E., Webb, T.R., Auer, P.L., et al. (2016). Coding variation
in ANGPTL4, LPL, and SVEP1 and the risk of coronary disease. N. Engl. J.
Med. 374, 1134–1144.
Mysling, S., Kristensen, K.K., Larsson, M., Beigneux, A.P., Gårdsvoll, H., Fong,
L.G., Bensadouen, A., Jørgensen, T.J., Young, S.G., and Ploug, M. (2016a).
The acidic domain of the endothelial membrane protein GPIHBP1 stabilizes li-
poprotein lipase activity by preventing unfolding of its catalytic domain. eLife 5,
e12095.
Mysling, S., Kristensen, K.K., Larsson, M., Kovrov, O., Bensadouen, A., Jør-
gensen, T.J., Olivecrona, G., Young, S.G., and Ploug, M. (2016b). The angio-
poietin-like protein ANGPTL4 catalyzes unfolding of the hydrolase domain in
lipoprotein lipase and the endothelial membrane protein GPIHBP1 counter-
acts this unfolding. eLife 5, e20958.
Nevo, Y., Ben-Zeev, B., Tabib, A., Straussberg, R., Anikster, Y., Shorer, Z.,
Fattal-Valevski, A., Ta-Shma, A., Aharoni, S., Rabie, M., et al. (2013). CD59
deficiency is associated with chronic hemolysis and childhood relapsing im-
mune-mediated polyneuropathy. Blood 121, 129–135.
Nilsson, S.K., Anderson, F., Ericsson, M., Larsson, M., Makoveichuk, E., Loo-
kene, A., Heeren, J., and Olivecrona, G. (2012). Triacylglycerol-rich lipopro-
teins protect lipoprotein lipase from inactivation by ANGPTL3 and ANGPTL4.
Biochim. Biophys. Acta 1821, 1370–1378.
Olafsen, T., Young, S.G., Davies, B.S., Beigneux, A.P., Kenanova, V.E., Voss,
C., Young, G., Wong, K.P., Barnes, R.H., 2nd, Tu, Y., et al. (2010). Unexpected
expression pattern for glycosylphosphatidylinositol-anchored HDL-binding
protein 1 (GPIHBP1) in mouse tissues revealed by positron emission tomogra-
phy scanning. J. Biol. Chem. 285, 39239–39248.
Olivecrona, T., and Bengtsson-Olivecrona, G. (1990). Lipoprotein lipase and
hepatic lipase. Curr. Opin. Lipidol. 1, 222–230.
Olivecrona, T., and Bengtsson-Olivecrona, G. (1993). Lipoprotein lipase and
hepatic lipase. Curr. Opin. Lipidol. 4, 187–196.
Olivecrona, T., Bengtsson-Olivecrona, G., Osborne, J.C., Jr., and Kempner,
E.S. (1985). Molecular size of bovine lipoprotein lipase as determined by radi-
ation inactivation. J. Biol. Chem. 260, 6888–6891.
Olivecrona, G., Ehrenborg, E., Semb, H., Makoveichuk, E., Lindberg, A., Hay-
den, M.R., Gin, P., Davies, B.S., Weinstein, M.M., Fong, L.G., et al. (2010).
Mutation of conserved cysteines in the Ly6 domain of GPIHBP1 in familial chy-
lomicronemia. J. Lipid Res. 51, 1535–1545.
64 Cell Metabolism 30, July 2, 2019
Cell Metabolism
Review
Osborne, J.C., Jr., Bengtsson-Olivecrona, G., Lee, N.S., and Olivecrona, T.
(1985). Studies on inactivation of lipoprotein lipase: role of the dimer to mono-
mer dissociation. Biochemistry 24, 5606–5611.
Peterson, J., Ayyobi, A.F., Ma, Y., Henderson, H., Reina, M., Deeb, S.S., San-
tamarina-Fojo, S., Hayden, M.R., and Brunzell, J.D. (2002). Structural and
functional consequences of missense mutations in exon 5 of the lipoprotein
lipase gene. J. Lipid Res. 43, 398–406.
Pingitore, P., Lepore, S.M., Pirazzi, C., Mancina, R.M., Motta, B.M., Valenti, L.,
Berge, K.E., Retterstøl, K., Leren, T.P., Wiklund, O., et al. (2016). Identification
and characterization of two novel mutations in the LPL gene causing type I hy-
perlipoproteinemia. J. Clin. Lipidol. 10, 816–823.
Plengpanich, W., Young, S.G., Khovidhunkit, W., Bensadoun, A., Karnman, H.,
Ploug, M., Gårdsvoll, H., Leung, C.S., Adeyo, O., Larsson, M., et al. (2014).
Multimerization of glycosylphosphatidylinositol-anchored high density lipo-
protein-binding protein 1 (GPIHBP1) and familial chylomicronemia from a
serine-to-cysteine Substitution in GPIHBP1 Ly6 domain. J. Biol. Chem. 289,
19491–19499.
Quagliarini, F., Wang, Y., Kozlitina, J., Grishin, N.V., Hyde, R., Boerwinkle, E.,
Valenzuela, D.M., Murphy, A.J., Cohen, J.C., and Hobbs, H.H. (2012). Atypical
angiopoietin-like protein that regulates ANGPTL3. Proc. Natl. Acad. Sci.
U.S.A. 109, 19751–19756.
Rabacchi, C., D’Addato, S., Palmisano, S., Lucchi, T., Bertolini, S., Calandra, S.,
and Tarugi, P. (2016). Clinical and genetic features of 3 patients with familial chy-
lomicronemia due to mutations in GPIHBP1 gene. J. Clin. Lipidol. 10, 915–921.
Rader, D.J. (2006). Molecular regulation of HDL metabolism and function: im-
plications for novel therapies. J. Clin. Invest. 116, 3090–3100.
Rios, J.J., Shastry, S., Jasso, J., Hauser, N., Garg, A., Bensadoun, A., Cohen,
J.C., and Hobbs, H.H. (2012). Deletion of GPIHBP1 causing severe chylomi-
cronemia. J. Inherit. Metab. Dis. 35, 531–540.
Rip, J., Nierman, M.C., Ross, C.J., Jukema, J.W., Hayden, M.R., Kastelein,
J.J., Stroes, E.S., and Kuivenhoven, J.A. (2006). Lipoprotein lipase S447X:
a naturally occurring gain-of-function mutation. Arterioscler. Thromb. Vasc.
Biol. 26, 1236–1245.
Rye, K.A., and Barter, P.J. (2014). Regulation of high-density lipoprotein meta-
bolism. Circ. Res. 114, 143–156.
Sendak, R.A., Melford, K., Kao, A., and Bensadoun, A. (1998). Identification of
the epitope of a monoclonal antibody that inhibits heparin binding of lipopro-
tein lipase: new evidence for a carboxyl-terminal heparin-binding domain.
J. Lipid Res. 39, 633–646.
Shoemaker, B.A., Portman, J.J., and Wolynes, P.G. (2000). Speeding molec-
ular recognition by using the folding funnel: the fly-casting mechanism. Proc.
Natl. Acad. Sci. U.S.A. 97, 8868–8873.
Silva, M.O. (2012). Risk of autoimmune complications associated with inter-
feron therapy. Gastroenterol. Hepatol. (NY) 8, 540–542.
Simionescu, M., and Simionescu, N. (1986). Functions of the endothelial cell
surface. Annu. Rev. Physiol. 48, 279–293.
Sukonina, V., Lookene, A., Olivecrona, T., and Olivecrona, G. (2006). Angio-
poietin-like protein 4 converts lipoprotein lipase to inactive monomers and
modulates lipase activity in adipose tissue. Proc. Natl. Acad. Sci. U.S.A.
103, 17450–17455.
van Tilbeurgh, H., Sarda, L., Verger, R., and Cambillau, C. (1992). Structure of
the pancreatic lipase–procolipase complex. Nature 359, 159–162.
Vanlandewijck, M., He, L., Ma €e, M.A., Andrae, J., Ando, K., Del Gaudio, F., Na-
har, K., Lebouvier, T., Laviña, B., Gouveia, L., et al. (2018). A molecular atlas of
cell types and zonation in the brain vasculature. Nature 554, 475–480.
Vannier, C., and Ailhaud, G. (1989). Biosynthesis of lipoprotein lipase in
cultured mouse adipocytes. II. Processing, subunit assembly, and intracellular
transport. J. Biol. Chem. 264, 13206–13216.
Vijayagopal, P., Radhakrishnamurthy, B., Srinivasan, S.R., and Berenson, G.S.
(1980). Studies of biologic properties of proteoglycans from bovine aorta. Lab.
Invest. 42, 190–196.
Voss, C.V., Davies, B.S., Tat, S., Gin, P., Fong, L.G., Pelletier, C., Mottler, C.D.,
Bensadoun, A., Beigneux, A.P., and Young, S.G. (2011). Mutations in
Cell Metabolism
Review
lipoprotein lipase that block binding to the endothelial cell transporter
GPIHBP1. Proc. Natl. Acad. Sci. U.S.A. 108, 7980–7984.
Wang, H., and Eckel, R.H. (2009). Lipoprotein lipase: from gene to obesity. Am.
J. Physiol. Endocrinol. Metab. 297, E271–E288.
Weinstein, M.M., Tu, Y., Beigneux, A.P., Davies, B.S., Gin, P., Voss, C., Wal-
zem, R.L., Reue, K., Tontonoz, P., Bensadoun, A., et al. (2010a). Cholesterol
intake modulates plasma triglyceride levels in glycosylphosphatidylinositol
HDL-binding protein 1-deficient mice. Arterioscler. Thromb. Vasc. Biol. 30,
2106–2113.
Weinstein, M.M., Yin, L., Tu, Y., Wang, X., Wu, X., Castellani, L.W., Walzem,
R.L., Lusis, A.J., Fong, L.G., Beigneux, A.P., et al. (2010b). Chylomicronemia
elicits atherosclerosis in mice–brief report. Arterioscler. Thromb. Vasc. Biol.
30, 20–23.
Winkler, F.K., D’Arcy, A., and Hunziker, W. (1990). Structure of human pancre-
atic lipase. Nature 343, 771–774.
Wion, K.L., Kirchgessner, T.G., Lusis, A.J., Schotz, M.C., and Lawn, R.M.
(1987). Human lipoprotein lipase complementary DNA sequence. Science
235, 1638–1641.
Wong, H., Davis, R.C., Thuren, T., Goers, J.W., Nikazy, J., Waite, M., and
Schotz, M.C. (1994). Lipoprotein lipase domain function. J. Biol. Chem. 269,
10319–10323.
Wong, H., Yang, D., Hill, J.S., Davis, R.C., Nikazy, J., and Schotz, M.C. (1997).
A molecular biology-based approach to resolve the subunit orientation of lipo-
protein lipase. Proc. Natl. Acad. Sci. U.S.A. 94, 5594–5598.
Yamamoto, H., Onishi, M., Miyamoto, N., Oki, R., Ueda, H., Ishigami, M., Hir-
aoka, H., Matsuzawa, Y., and Kihara, S. (2013). Novel combined GPIHBP1
mutations in a patient with hypertriglyceridemia associated with CAD.
J. Atheroscler. Thromb. 20, 777–784.
Zhang, L., Lookene, A., Wu, G., and Olivecrona, G. (2005). Calcium triggers
folding of lipoprotein lipase into active dimers. J. Biol. Chem. 280,
42580–42591.
Cell Metabolism 30, July 2, 2019 65
Lead
with
purpose
:HSXWVFLHQFHðUVW(YHU\WKLQJZHGRDLPVWRGULYHUHVHDUFKIRUZDUGò
ZKHWKHULWâVZRUNLQJZLWKDXWKRUVHQKDQFLQJWKHSHHUUHYLHZSURFHVV
RUGHYHORSLQJLQQRYDWLRQVLQSXEOLVKLQJ
:HFKDPSLRQWUDQVIRUPDWLYHDQGJURXQGEUHDNLQJLGHDV7KHDXWKRUDQG
WKHLUUHVHDUFKDUHDWRXUFHQWHU2XUHGLWRULDOWHDPLVNQRZOHGJHDEOHDQG
HQJDJHG2XUSURFHVVHVDUHWKRXJKWIXODQGIDLU
:HEHOLHYHWKDWLI\RXOHDGZLWKSXUSRVH\RXFDQLQVSLUHIXWXUHGLUHFWLRQV
:HKRSH\RXâOOMRLQXV/HWâVPDNHVFLHQFHEHWWHU

John Pham is the new Editor-in-Chief of Cell, the fourth in the journal’s 44-year-old history. He sees strong opportunities for Cell
and Cell Press to attract and publish the best science and to help push science forward.
Cell Metabolism

Short Article

Age Mosaicism across Multiple

Scales in Adult Tissues
Rafael Arrojo e Drigo,1 Varda Lev-Ram,2 Swati Tyagi,1 Ranjan Ramachandra,3 Thomas Deerinck,3 Eric Bushong,3
Sebastien Phan,3 Victoria Orphan,4 Claude Lechene,5 Mark H. Ellisman,2,6,7 and Martin W. Hetzer1,7,8,*
1Salk Institute for Biological Studies, Molecular and Cell Biology Laboratory (MCBL), La Jolla, CA, USA
2Department of Pharmacology, University of California, San Diego School of Medicine (UCSD), La Jolla, CA, USA
3National Center for Microscopy and Imaging Research (NCMIR), University of California, San Diego (UCSD), La Jolla, CA, USA
4Division of Geological and Planetary Sciences, California Institute of Technology, Pasadena, CA, USA
5Division of Genetics, Department of Medicine, Brigham and Women’s Hospital, Harvard Medical School, Boston, MA, USA
6Department of Neurosciences, University of California, San Diego School of Medicine (UCSD), La Jolla, CA, USA
7These authors contributed equally
8Lead Contact

*Correspondence: hetzer@salk.edu
https://doi.org/10.1016/j.cmet.2019.05.010

SUMMARY
Most neurons are not replaced during an animal’s
lifetime. This nondividing state is characterized by
extreme longevity and age-dependent decline of
key regulatory proteins. To study the lifespans of
cells and proteins in adult tissues, we combined
isotope labeling of mice with a hybrid imaging
method (MIMS-EM). Using 15N mapping, we show
that liver and pancreas are composed of cells with
vastly different ages, many as old as the animal.
Strikingly, we also found that a subset of fibroblasts
and endothelial cells, both known for their replicative
potential, are characterized by the absence of cell di-
vision during adulthood. In addition, we show that
the primary cilia of beta cells and neurons contains
different structural regions with vastly different life-
spans. Based on these results, we propose that
age mosaicism across multiple scales is a funda-
mental principle of adult tissue, cell, and protein
complex organization.
INTRODUCTION
The lifespan of a terminally differentiated cell is quite variable
among organs: 3 to 4 days for epithelial intestinal cells, to ol-
factory neuronal cells replaced from basal stem cells, to a life
time for the majority of neurons, cardiomyocytes, and all inner
hear hair cells (De Anda et al., 2016; Brann and Firestein,
2014; Foglia and Poss, 2016; Steinhauser et al., 2012; Zhang
et al., 2012). In some cases, somatic stem cells can respond
to tissue damage and proliferate according to tissue-specific
needs, as in the striated muscle, which can somewhat regen-
erate after wound because of activation of its satellite stem
cells (Blau et al., 2015). A similar pattern is also observed at
the proteome level, where proteins have different lifespans,
ranging from hours and days to years (Ori et al., 2015; Toyama
et al., 2013). Like stressed and/or damaged cells, misfolded
and damaged proteins must be degraded and replaced
with new and functional versions (Taylor and Dillin, 2011).
Over the last decades these insights steered biomedical
research toward focusing on cellular and molecular replace-
ment processes.
It remains poorly understood how neurons, cardiomyocytes,
and potentially other long-lived cells (LLCs), maintain functional
integrity and protein homeostasis over the span of several
decades. Because these cells are almost never replaced, they
are essentially as old as the animal itself and must function
properly throughout life, which in humans can be more than a
century (De Anda et al., 2016). Understanding how functionality
is maintained in LLCs is important given that aging is associated
with physiological impairments in these kinds of cells (e.g.,
neurons and cardiomyocytes) (D’Angelo et al., 2009; Mattson
and Magnus, 2006).
Recent advances in whole-animal metabolic-labeling strate-
gies and quantitative mass spectrometry (MS/MS) have enabled
system-wide, high-resolution analyses of global protein turnover

Context and Significance

Some cells in the body, such as skin, are constantly being replaced during one’s lifetime, while most brain cells are not. Cells
with minimal to no turnover contain long-lived proteins whose function declines with age. Researchers at the Salk Institute
and University of California, San Diego, in California used a detailed isotope-tracing method to measure and map the age of
cells and protein in different mouse organs. Unexpectedly, most organs are a mix of cells and proteins of vastly different
ages, regardless of whether they are slow or quick to regenerate. Some cells also contain proteins of different ages. Under-
standing the principles of cellular aging will be important in helping combat the age-associated decline in organ function.

Cell Metabolism 30, 343–351, August 6, 2019 ª 2019 Elsevier Inc. 343
rates. However, imaging approaches for studying extreme cell
longevity are not well developed. We recently combined stable
isotope metabolic pulse-chase labeling of rats using 15N (herein
called 15N-SILAM [stable isotope labeling in mammals]) with
quantitative MS/MS to discover a class of long-lived proteins
(LLPs) in neurons (Toyama et al., 2013). These neuronal LLPs
localize primarily to the nuclear pore complex (NPC) and chro-
matin, and are maintained with no or limited turnover, in striking
contrast to the majority of the proteome, which is renewed within
hours or days (Ori et al., 2015; Schoenheimer, 1942). Only a few
other LLPs have been previously identified, including lens crys-
talline, collagen, and myelin basic protein (Fischer and Morell,
1974; Lynnerup et al., 2008; Verzijlbergen et al., 2010). These
proteins deteriorate with age and, with the exception of myelin,
are unlikely to contribute to cellular aging because they typically
reside in cells with minimal metabolic activity and/or play struc-
tural roles (Lynnerup et al., 2008). In contrast, the neuronal LLPs
that we identified play critical roles in gene regulation and nuclear
trafficking pathways and age-dependent loss of these LLPs im-
pairs nuclear function (D’Angelo et al., 2009; Ibarra et al., 2016;
Toyama et al., 2013).
LLCs face a constant lifelong demand for performance to
maintain organ function and homeostasis and are constantly
exposed to drivers of molecular and cellular damage. In fact,
the accumulation of such damage over time may eventually alter
gene expression and protein homeostasis pathways, leading to
impaired cellular function and disease (D’Angelo et al., 2009;
Ibarra et al., 2016). Understanding these mechanisms requires
the identity and distribution pattern of LLCs in different organs.
However, lifelong cell persistence has not been analyzed in a
systematic and quantitative manner, and therefore fundamental
questions about cell type-specific turnover rates remains poorly
characterized. In addition, while implementation of 15N-SILAM
together with quantitative MS/MS allowed us to identify LLPs
(Toyama et al., 2013), this approach lacks spatial information
(other than the region of tissue that was dissected) and does
not provide any information regarding cell identities. To over-
come this challenge, we made use of 15N-SILAM-labeled
animals initially labeled for MS/MS to develop a new imaging
pipeline for correlated scanning electron microscopy (SEM)
and multi-isotope imaging mass spectroscopy (MIMS) (herein
called MIMS-electron microscopy [EM]). MIMS is a mapping
technique compatible with electron microscopy that can deter-
mine the relative amount of different stable isotopes incorpo-
rated into nucleic acids and proteins in cells and tissues
(Lechene et al., 2006; Steinhauser et al., 2012; Zhang et al.,
2012). Given the high spatial resolution and sensitivity of
EM and MIMS, we used MIMS-EM to visualize and quantify
cell and protein turnover in the brain, liver, and pancreas in young
and old 15N-SILAM mice and rats.
Using MIMS-EM on tissue samples from exceptionally long-
lived mice initially labeled for stable isotope mass spectrometry
experiments (Toyama et al., 2013), we found that neurons and
cardiomyocytes are not the only cell types with exceptional
longevity. Much to our surprise, adult mouse liver and pancreas
contain populations of cells with vastly different lifespans, many
of which are as old as neurons. The pancreas serves a paradigm
for the observed cellular age mosaicism, in that a subset of
insulin-beta cells are as old as cortical neurons, while others
344 Cell Metabolism 30, 343–351, August 6, 2019
have been replaced through cell division during adulthood. In
the liver, contrary to what has been inferred from liver regenera-
tion models, we discovered that most hepatocytes, bile duct
cells, and stellate-like cells (i.e., hepatic stellate cells and Kupffer
cells) are LLCs. Age mosaicism can also be observed among
capillary endothelial cells, which exhibit vastly different lifespans
in different tissue such as the brain, pancreas, and liver. Further-
more, age mosaicism also occurs at the molecular level, where
proteins with vastly different lifespans coexist at the supramolec-
ular level in the basal body and axoneme of the primary cilium of
neurons and beta cells.
RESULTS AND DISCUSSION
Identification of LLCs and Structures in the Central
Nervous System (CNS) with MIMS-EM
Information about the lifespan of different cells is still lacking for
many adult tissues, largely because of technical limitations. To
determine which and whether cells divide or persist in a nondi-
viding state throughout life we combined 15N metabolic labeling
of mice with MIMS-EM (Steinhauser et al., 2012). In brief, we
used animals initially intended for protein MS/MS as in our
previous studies (Toyama et al., 2013). As such, molecular com-
ponents of these animals (i.e., nucleic and amino acids) were
labeled with 15N throughout their embryonic development and
fed an 15N-rich diet until the 21st or 45th day of post-natal devel-
opment (see details in STAR Methods). We refer to these mice as
P21- or P45-15N-SILAM mice, respectively. After the 15N-label-
ing period, 15N-animals were chased with a normal diet with
the natural 99.4% content of 14N for 6, 18, or 26 months of
age. During the chase period, when 15N-labeled cells are
replaced by newly synthesized cells, 15N is replaced by 14N.
Therefore, the rate of cell (and protein turnover) can be quantified
with MIMS by simultaneously imaging 15N and 14N in a target re-
gion and determining the 15N/14N ratio (i.e., old-to-young ratio).
Importantly, our previous work has shown that the relative
amount of nuclear 15N-labeled protein represents less than 2%
of the whole nuclear proteome after 12 months of chase (Toyama
et al., 2013). Therefore, the vast majority of the nuclear 15N is
expected to stem from 15N-DNA and which decreases by 50%
after each cell division, as expected (Steinhauser et al., 2012).
Since our SILAM approach cannot distinguish between 15N-
DNA and rare 15N-proteins in the nucleus, we used MIMS-EM
to quantitatively determine the 15N content of random sections
of neuronal nuclei in different regions of the CNS. This allowed
us to establish a mean baseline and range of the nuclear 15N
signal that is characteristic for nondividing cells (De Anda
et al., 2016; D’Angelo et al., 2009). We targeted neurons in the
mouse layer-2 (L2) in the motor cortex and in the rat cerebellum
from animals labeled with 15N until P45 and chased for 6 months.
MIMS-EM of the mouse L2 cortex showed that most of the 15N
signal was nuclear or overlapping with myelin sheaths, as ex-
pected (Figure S1C) (Toyama et al., 2013). We measured whole
nuclei 15N/14N ratios and found that they were markedly higher
than the natural occurrence (15N/14N = 37 3 104) because of a
significant retention of 15N in all neuronal nuclei mapped (Figures
1A–1I and S1A–S1C). Since we observed species-specific differ-
ences (Figure S1E), we therefore decided to focus the rest of our
analysis on mouse tissues.
Figure 1. MIMS-EM of LLCs and Structures in the CNS
(A and B) SEM (A) and MIMS (B) of two capillaries in the optic nerve head (ONH) of a 6-month chase mouse. An endothelial cell nucleus (yellow arrow) and myelin
sheaths (pink arrows) are indicated. A pericyte nucleus is visible to the left the capillary lumen.
(C and D) SEM (C) and MIMS (D) of a capillary in the ONH. An old endothelial cell nucleus (yellow arrow), a fibroblast (pink arrow) and 15N-rich extracellular matrix
(ECM) (white arrow) are indicated.
(E and F) SEM (E) and MIMS (F) of a capillary in the ONH, with an old fibroblast (top) and a young endothelial cell (nucleus delineated in white and indicated by the
yellow asterisk).
(G and H) SEM (G) and MIMS (H) of an L2 neuron (top) and an endothelial cell.
(I) Close-up of granular cells in the rat cerebellum with an old endothelial cell visible. Full mosaic is shown in Figure S1B.
Scale bars: 5 mm (A, C, E, and G) (SEM). At the bottom of the MIMS images, the heatmap shows the 15N/14N 3 104 and scaled with a hue saturation intensity (HSI).

It is well established that cortical neurons do not divide during
adulthood, thus we decided to use the mean 15N/14N ratio from
random nuclear sections of L2 neurons as baseline level for the
identification of nondividing cells. We observed that the mean
15
N/14N in the L2 neuron population analyzed varied between
5- and 19-fold higher than the natural ratio (Figure S3D). This
variation is likely explained by the random occurrence of dense
heterochromatin patches in the nuclear sections imaged with
MIMS-EM, whereby sections with a higher density of hetero-
chromatin in a given section are expected to yield higher 15N
signals due to a relatively higher concentration of 15N-DNA
in comparison with euchromatin-rich sections. In fact, we
observed a maximum level of 15N labeling in intranuclear regions
that correlate with dense patches of heterochromatin, which
may also contain rare 15N-proteins in addition to 15N-DNA (Fig-
ures 1G, 1H, S1A, S1D, and S1E).
Accordingly, we decided to classify cells as ‘‘old’’ when
their nuclear mean 15N/14N ratio was similar to L2 neurons,
namely 15N/14N ratios at least 5.5-fold over the natural ratio
(i.e., 15N/14N > 208, Figure S1F). Importantly, the relative
age of non-neuronal cells in nervous and somatic organs
was determined by comparing their 15N/14N with that of L2
neurons from P45- or P21-SILAM (e.g., cells from P45-SILAM
animals were benchmarked against P45-SILAM L2 neurons).
Of note, while this classification approach can determine the
relative longevity of ‘‘old’’ cells after 6, 18, or 26 months of
chase, it cannot determine their proliferative potential during
the pulse period or how many times a cell has divided during
the chase period (see below, Limitations of Study). Neverthe-
less, we found that cells in the inner nuclear layer of the retina;
and most endothelial cells in the mouse optic nerve head
(ONH) and all in the cerebellum were LLCs and as old as
neurons, and did not show any sign of turnover (Figures 1A–
1F and S1A–S1D). In fact, we observed retention of 15N in
endothelial cells located in the L2 cortical layer of a
P21-15N-SILAM mouse chased for almost 26 months (Figures

Cell Metabolism 30, 343–351, August 6, 2019 345

1G and 1H). In addition, the nuclear 15N/14N ratio in peri-
vascular fibroblasts and an oligodendrocyte was >20-fold
higher than the natural 15N/14N ratio, similar to neurons (Fig-
ures 1A–1F and 4D). These data suggest that lifelong cell
persistence extends to non-neuronal cells, is more prominent
than previously thought and also confirms a recent report
regarding oligodendrocyte longevity (Tripathi et al., 2017). In
contrast, we found one example of a capillary endothelial
cell in the ONH that had close-to-background 15N levels
(Figures 1E and 1F), which indicates that turnover of endothe-
lial cells in the brain can happen, although it may be a rare
event. Together, these data from these exceptionally long-
lived SILAM mice provides the first evidence that the
neurovascular unit, and, more specifically, endothelial cells
in the brain, are mostly old. However, it also suggests that
some vessels are mosaics, built of cells with vastly different
lifespans (Figures 1A–1F). We conclude that MIMS-EM is a
powerful tool to study cellular lifespan and we decided to
explore the concept of age mosaicism in other tissues
(see below).
Cellular Age Mosaicism in the Liver
Several studies in the context of tissue injury have led to the
well-established concept that the liver has a high regenerative
capacity (Malato et al., 2011). In line with this concept, our pre-
vious 15N-SILAM MS/MS experiments in rats did not identify
15
N-labeled peptides in the liver after only a 4-month chase,
which further suggested that these cells are replaced within
months (Toyama et al., 2013). However, little is known about
the longevity of hepatocytes and other key liver cell types in
healthy adult mice (Magami et al., 2002; Malato et al., 2011).
We applied MIMS-EM to map cells located near bile ducts,
portal or central veins, and sinusoid capillaries (Figures 2A
and 2B), and found that most hepatocytes in P45-15N-SILAM
mice were as old as neurons. Long-lived hepatocytes were
observed in the vicinity of central veins, portal veins, and capil-
lary sinusoids after 6 and 18 months of chase (92% and 95%,
respectively) (Figures 2A, 2B, and S2A–S2D). Similarly, MIMS-
EM revealed that most cholangiocytes were old as neurons
and, therefore, are also LLCs (red arrows in Figures 2A, 2B,
2D, 2F, and S2A–S2D). In contrast, while endothelial cells in
portal and central veins did not divide (Figures 2A–2C, white ar-
rows; Figures S2A–S2D), most sinusoid capillary endothelial
cells contained 15N levels at the 18-month time point that
were lower than at the 6-month time point (Figures 2A, 2B,
and 2E, white arrows; Figures S2A–S2D). A similar phenotype
was observed in stellate-like cells (Figures 2A, 2B, and 2F,
green arrows; Figures S2A–S2D). These results indicate that,
while hepatocytes remain largely quiescent throughout life,
endothelial cells in hepatic sinusoids and most stellate-like cells
undergo a major turnover event between 6 and 18 months of
life. Lastly, the exceptional longevity of liver cells was associ-
ated with old extracellular matrix elements located to the peri-
vascular space of portal and central veins and interstitial space
(Figures 2A–2D and S2B). These data suggest that the aging
liver has a tissue architecture characterized by an intercellular
age mosaicism, where a young sinusoid vasculature would
support the function of relatively older cells, a notion that is
further supported by hepatocyte lineage tracing studies that
346 Cell Metabolism 30, 343–351, August 6, 2019
estimate that hepatocytes could indeed last for at least 1
year (Magami et al., 2002; Malato et al., 2011).
Adult Pancreas is Composed of Cells with Vastly
Different Lifespans
Having identified blood vessels containing endothelial cells with
different ages, we asked whether age mosaicism is a more gen-
eral principle and turned our attention to the pancreas, which
contains exocrine and endocrine cells. Development and matu-
ration of pancreatic endocrine and exocrine cells extends to the
early postnatal period when cells are mostly generated by self-
duplication (Dor et al., 2004; Houbracken and Bouwens, 2017).
Here, we used MIMS-EM to determine the longevity and turnover
of the endocrine and exocrine pancreas in adult animals labeled
until two different time points in early life (post-natal day 21 [P21]
or P45, shown in Figures 3A and 3B) and chased for almost
2 years. First, we analyzed islets from a P21-15N-SILAM mouse
chased with 14N for 26 months and observed that the turnover
rate of alpha, beta, and delta cells is not uniform. As expected,
we observed a large number of alpha and beta cells exhibiting
background 15N (representing 77% and 76%, respectively) (Fig-
ures 3F and S3D), likely reflecting the wave of alpha and beta cell
proliferation that takes place in neonates and is important for
beta cell maturation (Stolovich-Rain et al., 2015). Strikingly,
almost a quarter of alpha and beta cells (representing 23% and
24%, respectively) and all delta cells had 15N levels similar to
neurons (Figure 3F), and therefore suggest that a subpopulation
of islet endocrine cells can become LLCs before or at the wean-
ing stage and may not proliferate for an entire lifetime.
Next, we analyzed islets from a P45-15N-SILAM mouse
chased for 18 months and observed that most beta cells (61%)
had 15N levels approximately 10-fold higher than the natural ra-
tio, similar to L2 neurons after 18 months of chase (Figure S3D),
while 39% of beta cells had lower 15N levels and therefore had
proliferated during adulthood (Figures 3B–3G and S3D). Recent
data indicate that different types of beta cells exist in the islet,
including proliferation-prone and/or immature beta cells located
in a neogenic niche at the islet periphery (Bader et al., 2016; van
der Meulen et al., 2017). We mapped the location of proliferative
beta cells in relation to the islet periphery and found no correla-
tion between beta cell age and intraislet cellular location (Fig-
ure S3E), thus suggesting that beta cell proliferation is random
within the islet ultrastructure. Similar results were observed for
alpha cells, pancreatic endothelial cells, and stellate cells, which
were also found to be mostly LLCs (Figures 3B–3G and S3D–
S3F). In striking contrast, all delta cells and the majority of ductal
cells did not show age mosaicism, and their 15N levels were
similar to neurons (Figures 3A–3G and S3B–S3D). In contrast,
most acinar cells we observed had close to or background 15N
levels and thus have clearly undergone significant turnover
during adulthood (Figures 3A and 3B, white arrow; Figure S3D),
as expected (Houbracken and Bouwens, 2017). Notably, the
observation of long-lived acinar cells suggests that some acinar
cells become long-lived in early adult life (Figure 3G, pink arrow),
which could have a role in pancreatic cancer (Pan et al., 2013;
Westphalen et al., 2016; Wollny et al., 2016).
Together, these data indicate that the islet is a mosaic
composed of cells with vastly different lifespans, where most
alpha and beta cells assume a quiescent and LLC phenotype
Figure 2. LLCs in the Liver
(A and B) SEM and MIMS of cells near bile ducts, central veins or capillary sinusoids from a 15N-SILAM P45 mouse chased for 6 months (A) or 18 months (B). ECM
(yellow arrowheads), hepatocytes (pink arrowheads), cholangiocytes (red arrowheads), stellate-like (green arrowheads), and endothelial cells (white arrowheads)
are indicated. Cv, central vein; Bd, bile duct; c, capillary.
(C–F) Relative turnover of liver hepatocytes (C), cholangiocytes (D), sinusoid endothelial cells (E), and hepatic stellate-like cells (HSCs) (F) after a 6- or 18-month
(mo) chase. The total number of cells analyzed for each cell type is indicated underneath each pie chart. At the bottom of the MIMS images, the heatmap shows
the 15N/14N 3 104 and scaled with a HSI. Scale bar, 10 mm.

between weaning (P21) and the first month (P45) of adult life.
Moreover, it suggests that proliferation of alpha and beta cells
is not uniform throughout adulthood, contrary to what earlier
studies have indicated (Brennand et al., 2007; Dor et al., 2004).
While the specific differences between young and old beta or
alpha cells remains to be understood, our data adds to recent
studies that have begun to investigate the cellular identity,
the heterogeneity and aging of beta cell subpopulations and
its potential role in age-associated type II diabetes onset
(Aguayo-Mazzucato et al., 2017; Almaça et al., 2014; Basu
et al., 2003; Johnston et al., 2016; van der Meulen et al., 2017;
Segerstolpe et al., 2016).
Visualization of LLP Assemblies In Vivo
Our previous SILAM MS/MS experiments revealed that LLCs do
contain a subset of proteins that persist throughout life. Specif-
ically, we found that scaffold NPC proteins are maintained in
the nuclear envelope throughout adulthood in neurons (Savas
et al., 2012; Toyama et al., 2013). However, it was unclear
whether this exceptional longevity extended to other subcellular
components. While studying long-lived beta cells, we observed
the retention of significant 15N signal by a small, ‘‘star-shaped’’
structure within the cytoplasm of a beta cell from a P45-15N-
SILAM mouse (data not shown). We hypothesized that this signal
was associated with the basal body (Bb) of the cell’s primary

Cell Metabolism 30, 343–351, August 6, 2019 347

Figure 3. LLCs in the Pancreas
(A and B) SEM (A) and MIMS (B) of a cross-section of the islets of Langerhans. Old and young acinar cells are indicated by pink and white arrow, respectively.
Yellow dotted box highlights cells shown in (C).
(C) Enlarged view of boxed region in (A) and (B). SEM and MIMS of beta cells (yellow arrows) and an old alpha cell (pink arrow).
(D) An old delta cell (left) next to a younger beta cell (top right).
(E) SEM and MIMS of a young (bottom) and an old (top right) endothelial cell. Old pancreatic stellate cells are seen in the top and lower right corners.

(legend continued on next page)

348 Cell Metabolism 30, 343–351, August 6, 2019

 Basal Basal
Tz body Tz body
A Pm B C

Tz Basal 1.5-2x 1.5-2x 2-3x

body
basal root
microtubules
transition fibers
D 5-10 10-25 25-50 50-100x 0nm 80nm 160nm 240nm

320nm 400nm 480nm 560nm

Figure 4. Longevity of Primary Cilia in LLCs

(A) Cartoon of a primary cilium Bb. Microtubules (black), transition fibers (cyan), basal root (pink), transition zone (Tz), axoneme, and plasma membrane (Pm)
are shown.
(B and C) Two different sections of a beta cell primary cilium from a P21-15N-SILAM mouse chased for 26 months and imaged with MIMS-EM. Top rows: SEM
micrographs with visible transition zone, Bb, transition fibers (cyan arrows), and basal root (pink arrows). Bottom rows: overlay of SEM micrographs and 15N/14N
thresholds: 1.53–23 (green) and 23–33 (cyan) the natural 15N/14N of 37 3 104.
(D) SEM micrograph overlaid with thresholded 15N/14N of an L2 neuron from a P45-15N-SILAM mouse chased for 6 months. The yellow box indicates the Bb.
On the right are serial sections of the neuron Bb and primary cilium. 15N/14N thresholds: 53–103 (green), 103–253 (cyan), 253–503 (yellow), and 503–1003
(magenta) the natural 15N/14N. At the bottom of the MIMS images, the heatmap shows the 15N/14N 3 104 and scaled with a HSI.
Scale bars: 300 nm (B), 1 mm (D), 500 nm (D, inset).

cilium (Figure 4A), which is a signaling organelle important
for beta-cell function (Gerdes et al., 2014). We tested this
hypothesis by imaging different beta-cell BBs from P21- and
P45-15N-SILAM mice chased for 18 and 26 months, respectively.
We applied MIMS-EM to serial cross-sections of different primary
cilia within beta cells and discovered that, indeed, the Bb con-
tained high 15N levels, while the rest of the cilium has been re-
placed by new components (Figures 4B, 4C, S4C, and S4D).
We expanded this analysis to L2 neurons, where we observed
that the 15N signal was not limited to the Bb and was also found
in the axoneme, decreasing as the cilium projected toward the
extracellular space (Figure 4D). These findings provide evidence
of the occurrence of protein age mosaicism within one subcellular
structure, while suggesting that long-lived structures are present
in the primary cilium and that their turnover could be different in
beta cells and neurons (Figures 4B–4D). While the identity of
these long-lived components remains to be discovered, it is note-
worthy that long-lived nucleoporins, including Nup93, are known
to form a permeability barrier in the ciliary gate in the cilium tran-
sition zone (McClure-Begley and Klymkowsky, 2017). These
exciting findings raise the question of whether aging is associated
with impaired primary cilium function due to loss of LLPs, as seen
for NPCs in old neurons (D’Angelo et al., 2009).
Analogous to genetic mosaicism, which refers to the presence
of two or more populations of cells with different genotypes in one
individual, we propose the concept of age mosaicism as a key
feature of adult tissue organization. Our data indicate that non-
uniform turnover of cells and proteins gives rise to tissue- and
cell-specific aging architectures characterized by a mosaic orga-
nization of young and old elements at the cell and protein level.
These data provide insight into an organization pattern in biology
that forms an age mosaic across the mesoscale, where the close

(F) Relative turnover in percentages of cells that are as old (gray) or younger (white) than L2 neurons from 15N-SILAM P21 mouse chased for 26 months.
(G) Same as in (F), but from a 15N-SILAM P45 mouse chased for 18 months. The total number of cells analyzed for each cell type is listed underneath each pie
chart. At the bottom of the MIMS images, the heatmap shows the 15N/14N 3 104 and scaled with an HSI.
Scale bars: 5 mm (A, C, and E) and 2.5 mm (D).

Cell Metabolism 30, 343–351, August 6, 2019 349

relationship of young and old components is found in protein su-
per complexes and organelles (i.e., Bb in the primary cilium), to
cells (e.g., beta cells and hepatocytes) and tissues (e.g., brain,
liver, and pancreas). This study raises the exciting possibility
that the age-dependent decline in LLPs is not limited to neurons,
but may be a widespread phenomenon for all tissues rich in LLCs
and links the loss of LLPs with the onset of age-related disease
such as diabetes (Basu et al., 2003). More broadly, future SILAM
studies will contain the use of different stable isotope-labeled
molecules (i.e., 15N-thymidine for DNA and 18O-Leucine for pro-
teins) to help identify and functionally characterize other exam-
ples of heterogeneity in the aging of cells and proteins in different
organs and in health and disease.
Limitations of Study
Our pulse-chase labeling approach using 15N-enriched food can
provide information about protein identity using MS/MS ana-
lyses (Toyama et al., 2013). However, our experimental set up
using MIMS-EM cannot distinguish between 15N-labeled nucleic
acids and amino acids in the regions we investigated. Although
most of the 15N signal shown in brain, liver, and pancreatic cells
is most likely due to 15N-DNA, we cannot rule out a potential
small contribution of 15N-peptides from intranuclear LLPs
(Toyama et al., 2013). However, regardless of the nature of the
15
N signal our results suggest that these cells are long-lived.
Second, in its current stage, MIMS-EM is not a high-throughput
imaging technique as it is time and labor intensive, requiring cells
and tissues to be cut in thin sections (50–250 nm in thickness)
and long acquisition times (e.g., 0.5–48 h for a 30 3 30-mm
raster), thus making volumetric imaging of any number of cells
very challenging. As a result, this study was limited by the acqui-
sition of random nuclear sections, which can add variability to
the 15N measurements by the random distribution of heterochro-
matin-rich domains in the sections analyzed. Nevertheless, we
have imaged >700 cells in total with MIMS-EM to provide an
important insight into the location and identity of LLCs outside
the brain. MIMS-EM is a very sensitive and quantitative tech-
nique with multiplexed detection capabilities, and therefore
future applications of MIMS-EM will take advantage of a
SILAM approach that labels nucleic acids, amino acids, and fatty
acids with different stable isotopes (15N, 18O, and 13C, for
example). Together with improved tools for cell/tissue mapping
and data acquisition pipelines, MIMS-EM can be used to inves-
tigate a large number of cells and intracellular components from
cells in virtually any SILAM tissue, including humans (Stein-
hauser et al., 2012).
STAR+METHODS
Detailed methods are provided in the online version of this paper
and include the following:
d KEY RESOURCES TABLE
d CONTACT FOR REAGENT AND RESOURCE SHARING
d EXPERIMENTAL MODEL AND SUBJECT DETAILS
B Animal Studies
d METHOD DETAILS
B Stable Isotope Metabolic Labeling of Mam-
mals (SILAM)
350 Cell Metabolism 30, 343–351, August 6, 2019
B Processing of Samples for Electron Microscopy
B X-Ray Microscopy (XRM)
B Correlative Electron Microscopy and Multi-isotope
Mass Spectroscopy (MIMS)
B MIMS-EM Image Analysis and Display
d QUANTIFICATION AND STATISTICAL ANALYSIS
15 14
B Quantification of N/ N Ratio in Tissues and Cell Age
Classification
B Cell Type Classification in Electron Micrographs
d DATA AND SOFTWARE AVAILABILITY
d ADDITIONAL RESOURCES
SUPPLEMENTAL INFORMATION
Supplemental Information can be found online at https://doi.org/10.1016/j.
cmet.2019.05.010.
ACKNOWLEDGMENTS
This work was supported by grants to M.W.H. from the NIH Transformative
Research Award (R01 NS096786), the Keck Foundation, and the NOMIS Foun-
dation. It was also supported by grants to M.H.E. and V.L.-R. from the NIH
NINDS (NIH RO1 NS027177-30), which supported the creation of the 15N/14N
mice; and the NIH NIGMS (5P41 GM103412-29), which supports the advanced
technologies and instrument development activities of the NCMIR, most heavily
used, and whose new methods developments were partly driven by this work.
V.O. and M.H.E. are also supported to develop and carry out correlated light
and electron microscopy with MIMS under an award from the US Department
of Energy, Office of Science, Office of Biological and Environmental Research
(DE-SC0016469). R.A.eD. is supported by an American Diabetes Association
postdoctoral fellowship (1-18-PMF-007) and benefitted from assistance from
Daniela Rhodes and the Nanyang Institute of Structural Biology (NISB) (at the
Nanyang Technological University, Singapore) from a career development sti-
pend. The authors are also thankful to Yunbin Guan, Ph.D., from the Caltech
Microanalysis Center in the Division of Geological and Planetary Sciences, Cal-
ifornia Institute of Technology for technical support and assistance with MIMS
image acquisition; to Dr. Ting-Di Wu and Dr. Jean-Luc Guerquin-Kern from the
INSERM, Université Paris-Sud, Paris; and the Cell and Tissue Imaging Facility of
the Institut Curie (PICT), a member of the France BioImaging National Infrastruc-
ture (ANR-10-INBS-04), for the use of Curie NanoSIMS instrument; Greg
McMahon from the National Center of Excellence in Mass Spectrometry Imag-
ing, National Physical Laboratory (NPL), UK; and to David O’Keefe (Salk Insti-
tute) for critical inputs on manuscript editing and revision.
AUTHOR CONTRIBUTIONS
R.A.eD. collected and analyzed the data and wrote the article. V.L.-E., S.T.,
R.R., T.D., E.B., S.P., and C.L. collected the data. V.O., M.H.E., and M.W.H.
designed the study and wrote the article.
DECLARATION OF INTERESTS
The authors declare no competing interests.
Received: June 14, 2018
Revised: October 18, 2018
Accepted: May 11, 2019
Published: June 6, 2019
REFERENCES
Adams, S.R., Mackey, M.R., Ramachandra, R., Palida Lemieux, S.F.,
Steinbach, P., Bushong, E.A., Butko, M.T., Giepmans, B.N.G., Ellisman,
M.H., and Tsien, R.Y. (2016). Multicolor electron microscopy for simultaneous
visualization of multiple molecular species. Cell Chem. Biol. 23, 1417–1427.
Aguayo-Mazzucato, C., van Haaren, M., Mruk, M., Lee, T.B., Crawford, C.,
Hollister-Lock, J., Sullivan, B.A., Johnson, J.W., Ebrahimi, A., Dreyfuss,
J.M., et al. (2017). Beta cell aging markers have heterogeneous distribution
and are induced by insulin resistance. Cell Metab. 25, 898–910.
Almaça, J., Molina, J., Arrojo e Drigo, R., Abdulreda, M.H., Jeon, W.B., Berggren,
P.O., Caicedo, A., and Nam, H.G. (2014). Young capillary vessels rejuvenate
aged pancreatic islets. Proc. Natl. Acad. Sci. USA 111, 17612–17617.
Bader, E., Migliorini, A., Gegg, M., Moruzzi, N., Gerdes, J., Roscioni, S.S.,
Bakhti, M., Brandl, E., Irmler, M., Beckers, J., et al. (2016). Identification of pro-
liferative and mature b-cells in the islets of Langerhans. Nature 535, 430–434.
Basu, R., Breda, E., Oberg, A.L., Powell, C.C., Dalla Man, C., Basu, A., Vittone,
J.L., Klee, G.G., Arora, P., Jensen, M.D., et al. (2003). Mechanisms of the age-
associated deterioration in glucose tolerance: contribution of alterations in in-
sulin secretion, action, and clearance. Diabetes 52, 1738–1748.
Blau, H.M., Cosgrove, B.D., and Ho, A.T.V. (2015). The central role of muscle
stem cells in regenerative failure with aging. Nat. Med. 21, 854–862.
Brann, J.H., and Firestein, S.J. (2014). A lifetime of neurogenesis in the olfac-
tory system. Front. Neurosci. 13, 112.
Brennand, K., Huangfu, D., and Melton, D. (2007). All b cells contribute equally
to islet growth and maintenance. PLOS Biol. Published May 29, 2007. https://
doi.org/10.1371/journal.pbio.0050163.
D’Angelo, M.A., Raices, M., Panowski, S.H., and Hetzer, M.W. (2009). Age-
dependent deterioration of nuclear pore complexes causes a loss of nuclear
integrity in postmitotic cells. Cell 136, 284–295.
De Anda, F.C., Madabhushi, R., Rei, D., Meng, J., Gra €ff, J., Durak, O., Meletis,
K., Richter, M., Schwanke, B., Mungenast, A., et al. (2016). Cortical neurons
gradually attain a post-mitotic state. Cell Res. 26, 1033–1047.
Dor, Y., Brown, J., Martinez, O.I., and Melton, D.A. (2004). Adult pancreatic
beta-cells are formed by self-duplication rather than stem-cell differentiation.
Nature 429, 41–46.
Fischer, C.A., and Morell, P. (1974). Turnover of proteins in myelin and myelin-
like material of mouse brain. Brain Res. 74, 51–65.
Foglia, M.J., and Poss, K.D. (2016). Building and re-building the heart by car-
diomyocyte proliferation. Development 143, 729–740.
Gerdes, J.M., Christou-Savina, S., Xiong, Y., Moede, T., Moruzzi, N., Karlsson-
Edlund, P., Leibiger, B., Leibiger, I.B., Östenson, C.G., Beales, P.L., et al.
(2014). Ciliary dysfunction impairs beta-cell insulin secretion and promotes
development of type 2 diabetes in rodents. Nat. Commun. 5, 5308.
Goldstein, J.I., Lyman, C.E., Newbury, D.E., Lifshin, E., Echlin, P., Sawyer, L.,
Joy, D.C., and Micheal, J.R. (2003). Scanning electron microscopy and micro-
analysis. Scanning Electron Microsc. X-Ray Microanal. 34, 690.
Herrmann, A.M., Ritz, K., Nunan, N., Clode, P.L., Pett-Ridge, J., Kilburn, M.R.,
Murphy, D.V., O’Donnell, A.G., and Stockdale, E.A. (2007). Nano-scale sec-
ondary ion mass spectrometry - a new analytical tool in biogeochemistry
and soil ecology: a review article. Soil Biol. Biochem. 39, 1835–1850.
Houbracken, I., and Bouwens, L. (2017). Acinar cells in the neonatal pancreas
grow by self-duplication and not by neogenesis from duct cells. Sci. Rep. 7.
Ibarra, A., Benner, C., Tyagi, S., Cool, J., and Hetzer, M.W. (2016).
Nucleoporin-mediated regulation of cell identity genes. Genes Dev. 30,
2253–2258.
Johnston, N.R., Mitchell, R.K., Haythorne, E., Pessoa, M.P., Semplici, F.,
Ferrer, J., Piemonti, L., Marchetti, P., Bugliani, M., Bosco, D., et al. (2016).
Beta cell hubs dictate pancreatic islet responses to glucose. Cell Metab. 24,
389–401.
Lechene, C., Hillion, F., McMahon, G., Benson, D., Kleinfeld, A.M., Kampf,
J.P., Distel, D., Luyten, Y., Bonventre, J., Hentschel, D., et al. (2006). High-res-
olution quantitative imaging of mammalian and bacterial cells using stable
isotope mass spectrometry. J. Biol. 5, 20.
Lynnerup, N., Kjeldsen, H., Heegaard, S., Jacobsen, C., and Heinemeier, J.
(2008). Radiocarbon dating of the human eye lens crystallines reveal proteins
without carbon turnover throughout life. PLoS One 3, e1529.
Magami, Y., Azuma, T., Inokuchi, H., Kokuno, S., Moriyasu, F., Kawai, K., and
Hattori, T. (2002). Cell proliferation and renewal of normal hepatocytes and bile
duct cells in adult mouse liver. Liver 22, 419–425.
€rmann, N., Ng, R., Wang, B., Zape, J., Kay, M.A.,
Malato, Y., Naqvi, S., Schu
Grimm, D., and Willenbring, H. (2011). Fate tracing of mature hepatocytes in
mouse liver homeostasis and regeneration. J. Clin. Invest. 121, 4850–4860.
Mattson, M.P., and Magnus, T. (2006). Ageing and neuronal vulnerability. Nat.
Rev. Neurosci. 7, 278–294.
McClure-Begley, T.D., and Klymkowsky, M.W. (2017). Nuclear roles for cilia-
associated proteins. Cilia 6, 8.
Ori, A., Toyama, B.H., Harris, M.S., Bock, T., Iskar, M., Bork, P., Ingolia, N.T.,
Hetzer, M.W., and Beck, M. (2015). Integrated transcriptome and proteome
analyses reveal organ-specific proteome deterioration in old rats. Cell Syst.
1, 224–237.
Pan, F.C., Bankaitis, E.D., Boyer, D., Xu, X., Van de Casteele, M., Magnuson,
M.A., Heimberg, H., and Wright, C.V.E. (2013). Spatiotemporal patterns of mul-
tipotentiality in Ptf1a-expressing cells during pancreas organogenesis and
injury-induced facultative restoration. Development 140, 751–764.
Savas, J.N., Toyama, B.H., Xu, T., Yates, J.R., and Hetzer, M.W. (2012).
Extremely long-lived nuclear pore proteins in the rat brain. Science 335, 942.
Schoenheimer, R. (1942). The dynamic state of body constituents. Cancer
Res. 2, 810.
Segerstolpe, Å., Palasantza, A., Eliasson, P., Andersson, E.M., Andréasson,
A.C., Sun, X., Picelli, S., Sabirsh, A., Clausen, M., Bjursell, M.K., et al. (2016).
Single-cell transcriptome profiling of human pancreatic islets in health and
Type 2 diabetes. Cell Metab. 24, 593–607.
Sosinsky, G.E., Deerinck, T.J., Greco, R., Buitenhuys, C.H., Bartol, T.M., and
Ellisman, M.H. (2005). Development of a model for microphysiological simula-
tions: small nodes of Ranvier from peripheral nerves of mice reconstructed by
electron tomography. Neuroinformatics 3, 133–162.
Steinhauser, M.L., Bailey, A.P., Senyo, S.E., Guillermier, C., Perlstein, T.S.,
Gould, A.P., Lee, R.T., and Lechene, C.P. (2012). Multi-isotope imaging
mass spectrometry quantifies stem cell division and metabolism. Nature
481, 516–519.
Stolovich-Rain, M., Enk, J., Vikesa, J., Nielsen, F.C., Saada, A., Glaser, B., and
Dor, Y. (2015). Weaning triggers a maturation step of pancreatic b cells. Dev.
Cell 32, 535–545.
Taylor, R.C., and Dillin, A. (2011). Aging as an event of proteostasis collapse.
Cold Spring Harb. Perspect. Biol. 3, 1–17.
Thévenaz, P., Ruttimann, U.E., and Unser, M. (1998). A pyramid approach to
subpixel registration based on intensity. IEEE Trans. Image Process 7, 27–41.
Toyama, B.H., Savas, J.N., Park, S.K., Harris, M.S., Ingolia, N.T., Yates, J.R.,
and Hetzer, M.W. (2013). Identification of long-lived proteins reveals excep-
tional stability of essential cellular structures. Cell 154, 971–982.
Tripathi, R.B., Jackiewicz, M., McKenzie, I.A., Kougioumtzidou, E., Grist, M.,
and Richardson, W.D. (2017). Remarkable stability of myelinating oligodendro-
cytes in mice. Cell Rep. 21, 316–323.
van der Meulen, T., Mawla, A.M., DiGruccio, M.R., Adams, M.W., Nies, V.,
Dólleman, S., Liu, S., Ackermann, A.M., Cáceres, E., Hunter, A.E., et al.
(2017). Virgin beta cells persist throughout life at a neogenic niche within
pancreatic islets. Cell Metab. 25, 911–926.
Verzijlbergen, K.F., Menendez-Benito, V., van Welsem, T., van Deventer, S.J.,
Lindstrom, D.L., Ovaa, H., Neefjes, J., Gottschling, D.E., and van Leeuwen, F.
(2010). Recombination-induced tag exchange to track old and new proteins.
Proc. Natl. Acad. Sci. USA 107, 64–68.
Westphalen, C.B., Takemoto, Y., Tanaka, T., Macchini, M., Jiang, Z., Renz,
B.W., Chen, X., Ormanns, S., Nagar, K., Tailor, Y., et al. (2016). Dclk1 defines
quiescent pancreatic progenitors that promote injury-induced regeneration
and tumorigenesis. Cell Stem Cell 18, 441–455.
€ne, D., Ziebell, F.,
Wollny, D., Zhao, S., Everlien, I., Lun, X., Brunken, J., Bru
Tabansky, I., Weichert, W., Marciniak-Czochra, A., et al. (2016). Single-cell
analysis uncovers clonal acinar cell heterogeneity in the adult pancreas.
Dev. Cell 39, 289–301.
Zhang, D.S., Piazza, V., Perrin, B.J., Rzadzinska, A.K., Poczatek, J.C., Wang,
M., Prosser, H.M., Ervasti, J.M., Corey, D.P., and Lechene, C.P. (2012). Multi-
isotope imaging mass spectrometry reveals slow protein turnover in hair-cell
stereocilia. Nature 481, 520–524.
Cell Metabolism 30, 343–351, August 6, 2019 351
Where big ideas multiply
Cell Reports Physical Science—a new broad-scope, open access journal
from Cell Press—publishes cutting-edge research across the full spectrum
of the physical sciences, including chemistry, physics, materials science,
energy science, engineering, and related interdisciplinary work.

Submit your paper today.

www.cell.com/cell-reports-physical-science
 ll

Resource
Patch-Seq Links Single-Cell Transcriptomes
to Human Islet Dysfunction in Diabetes
Joan Camunas-Soler,1,2,8 Xiao-Qing Dai,3,4,8 Yan Hang,5 Austin Bautista,3,4 James Lyon,4 Kunimasa Suzuki,4
Seung K. Kim,5,6,* Stephen R. Quake,1,2,6,7,9,10,* and Patrick E. MacDonald3,4,9,*
1Department of Bioengineering, Stanford University, Stanford, CA 94305, USA
2Chan Zuckerberg Biohub, San Francisco, CA 94518, USA
3Department of Pharmacology, University of Alberta, Edmonton, AB T6G 2E1, Canada
4Alberta Diabetes Institute, University of Alberta, Edmonton, AB T6G 2E1, Canada
5Department of Developmental Biology, Stanford University School of Medicine, Stanford, CA 94305, USA
6Stanford Diabetes Research Center, Stanford University, Stanford, CA 94305, USA
7Department of Applied Physics, Stanford University, Stanford, CA 94305, USA
8These authors contributed equally
9Senior author
10Lead Contact

*Correspondence: seungkim@stanford.edu (S.K.K.), steve@quake-lab.org (S.R.Q.), pmacdonald@ualberta.ca (P.E.M.)

https://doi.org/10.1016/j.cmet.2020.04.005

SUMMARY

Impaired function of pancreatic islet cells is a major cause of metabolic dysregulation and disease in hu-
mans. Despite this, it remains challenging to directly link physiological dysfunction in islet cells to precise
changes in gene expression. Here we show that single-cell RNA sequencing combined with electrophys-
iological measurements of exocytosis and channel activity (patch-seq) can be used to link endocrine
physiology and transcriptomes at the single-cell level. We collected 1,369 patch-seq cells from the
pancreata of 34 human donors with and without diabetes. An analysis of function and gene
expression networks identified a gene set associated with functional heterogeneity in b cells that can
be used to predict electrophysiology. We also report transcriptional programs underlying dysfunction
in type 2 diabetes and extend this approach to cryopreserved cells from donors with type 1 diabetes,
generating a valuable resource for understanding islet cell heterogeneity in health and disease.

INTRODUCTION derstand the connection between dynamic physiological pro-

Cells use exocytosis to secrete a wide variety of molecules,
including proteins, hormones, and neurotransmitters. From islet
cells in the pancreas to follicle cells in the reproductive system,
endocrine cells secrete hormones that regulate metabolism,
growth, and development (Burgoyne and Morgan, 2003;
Williams, 2016). Exocytosis can be monitored at the single-cell
level by using patch-clamp electrophysiology to measure
changes in membrane capacitance as vesicles fuse with the cell
membrane and release their content. It is of great interest to un-
cesses such as exocytosis with the generally slower changes
in the transcriptomes of cells, which reflect the cell’s gene
expression and genome regulatory processes. Single-cell RNA
sequencing (scRNA-seq) allows us to profile the molecular
identity of individual cells and detect changes in gene expression
that arise in pathological settings. This technique has pro-
vided increasingly accurate cell-type classifications in diverse
organs based on their transcriptional fingerprint (La Manno
et al., 2016; Tabula Muris Consortium et al., 2018; Villani et al.,
2017). However, it remains challenging to directly attribute

Context and Significance

Understanding the molecular mechanisms behind the dysfunction of pancreatic islet cells during diabetes could lead to bet-
ter therapies. Recently, key cell types in the islets have been found to exist as different subpopulations; for example, not all b
cells (the source of insulin) are the same in a given pancreas. But how these subpopulations may be functionally different has
been unclear. Here Camunas-Soler, Dai et al. address this issue by combining scRNA-seq (as a measure of molecular
phenotype) with single-cell electrophysiological measurements (as a measure of functional phenotype) in islet cells from
individuals with or without diabetes. The authors report induction of genes suggestive of compensation mechanisms,
and identify transcriptomic patterns associated with dysfunction, in b cells during T1D and T2D.

Cell Metabolism 31, 1017–1031, May 5, 2020 ª 2020 Published by Elsevier Inc. 1017
 ll
physiological properties in a cell to its measured transcriptomic
phenotype, and to establish unambiguous links with cellular
dysfunction in disease (Stuart and Satija, 2019; Wang and Kaest-
ner, 2019).
The islets of Langerhans control circulating glucose levels in
the body through regulated exocytosis of the two principal glucor-
egulatory hormones, insulin and glucagon. Several single-cell ge-
nomics studies revealed significant variability across islet cell
types, as well as the existence of transcriptionally distinct cellular
subpopulations (Baron et al., 2016; Enge et al., 2017; Muraro
et al., 2016; Segerstolpe et al., 2016). This heterogeneity has
been confirmed through identification of surface markers and
mass spectrometry signatures (Bader et al., 2016; van der Meulen
et al., 2017; Wang et al., 2016a). Prior seminal physiologic studies
established that islet cells, and especially insulin-producing b cells,
have heterogeneous function (Pipeleers, 1992; Salomon and
Meda, 1986; Stefan et al., 1987), while impaired b cell function
and reduced exocytosis are hallmark features of early-stage type
2 diabetes (T2D) (Alejandro et al., 2015; Ashcroft and Rorsman,
2012; Kahn et al., 2006). However, the ability to link islet cell molec-
ular heterogeneity directly to such physiologic properties remains
limited (Bakhti et al., 2019; Wang and Kaestner, 2019). Moreover,
the connection between genes up- or downregulated in T2D (Se-
gerstolpe et al., 2016) to functional consequences in islets remains
unclear, and major gaps persist in our mechanistic understanding
of T2D ‘‘risk’’ candidate genes identified by genome-wide associ-
ation studies (Mahajan et al., 2018; Prasad and Groop, 2016;
Tritschler et al., 2017).
To address such limitations, we combine here the use of
whole-cell patch-clamp measurements and scRNA-seq
(together referred to as patch-seq; Cadwell et al., 2016; Földy
et al., 2016; Fuzik et al., 2016) in dispersed islet cells. This
approach enabled us to simultaneously measure the cellular
function and transcriptomes of human endocrine cells, allowing
us to link gene expression and quantitative physiologic measure-
ments of vesicle exocytosis and ion-channel activity. We
collected 1,021 patch-seq cells from 28 donors with no diabetes
(ND) or with T2D, most with short disease duration, and
dissected function-to-gene expression relationships across islet
cell types. We discovered a subset of 484 genes that predicts
multiple b cell electrophysiological phenotypes, and we vali-
dated putative novel regulators of b cell physiology through ge-
netic loss-of-function experiments. We also identified pathways
associated with reduced exocytosis in b cells from donors with
T2D, and found that these are distinct from those observed in
nondiabetic donors. In b cells from individuals with T2D we
also identified an upregulation of genes linked to a high exocy-
tosis phenotype, compared to b cells from nondiabetic donors,
and propose a genetic mechanism underlying physiological
dysfunction of b cells during T2D. For example, we show that
the transcription factor ETV1 likely plays a major role in b cell
dysfunction in humans with T2D. Finally, we demonstrate that
our approach can also be applied to cryopreserved human islet
samples. Using 348 cells from rare cryopreserved islets of do-
nors with T1D, we obtained profiles from a, g, d, duct, and surviv-
ing b cells. Our work maps islet cell function and its transcrip-
tome at the single-cell level, providing a valuable resource for
exploring islet physiological and genetic dysfunction in health
and disease.
1018 Cell Metabolism 31, 1017–1031, May 5, 2020
Resource
RESULTS
Patch-Seq in Endocrine Cells to Simultaneously Assess
Physiology and the Transcriptome
To achieve patch-clamp followed by scRNA-seq (patch-seq) in
individual human islet cells, we initially isolated cells from 28 do-
nors, including 18 with no diabetes (ND), 7 with T2D, and 3 addi-
tional donors that were excluded from further analysis due to the
reasons specified in Table S1. We established patch-seq as a
two-step process: (1) we performed electrophysiological mea-
surements using whole-cell patch-clamp, and (2) within 5 min
from ‘‘break-in’’ we collected cellular content using a larger sec-
ondary pipette filled with lysis buffer (Figure 1A; STAR Methods).
This allowed intracellular access for whole-cell recording and
exocytosis measurements and was followed by recovery of
full-length transcriptomes using Smart-seq2 that were
sequenced to an average depth of 1–2 million reads (Figures
S1A–S1C; STAR Methods). A total of 1,021 cells (80%) passed
quality control for both electrophysiology and sequencing and
were classified into major cell types based on the expression
of key marker genes in a tSNE projection (Figure 1B; STAR
Methods). We obtained representatives of all major islet cell
types (a, b, d, and g cells) and non-islet types such as acinar cells
(Figures 1C and 1D). For each cell we measured parameters rep-
resenting cell size, exocytosis, Na+ channel currents, and Ca2+
channel currents (Figures 1E and 1F). Thus, we obtained a broad
survey of electrophysiological activity of all major islet cell types
in both ND and T2D settings (Figures S1D–S1K). In addition to
expected cell-type differences in size measured by whole-cell
capacitance (Figures 1C and 1F), these data demonstrate sub-
stantial variation in exocytosis and channel activity in different
pancreatic cell types and disease states, including altered
exocytotic function in both a and b cells in T2D (Figures
S1D–S1K).
To rigorously assess the robustness of patch-seq, we also
collected an additional 3,518 cells by fluorescence-activated
cell sorting (FACS) for scRNA-seq without patch clamping
from 14 of these donors (8 ND, 6 T2D; Figure S1A; STAR
Methods). The transcriptomes of cells after patch-seq or
FACS purification from the same donors led to comparable
quality metrics (Figure S2A). While we observed a difference
in the number of genes with detectable transcripts, the values
from cells after patch-seq or FACS were within the range of
previously published datasets (Figure S2B; Segerstolpe et al.,
2016). Analysis of genes required to maintain fundamental
cellular functions (hereafter, ‘‘housekeeping’’), islet identity,
and immediate early genes (IEGs) showed that most differ-
ences are driven by (1) varying sensitivity to genes expressed
at low levels (drop-out of genes with low expression), and (2)
varying expression of stress-response genes likely reflecting
steps like islet shipping and dispersion (Figure S2C). Gene
expression in patch-clamped cells showed a high number of
islet-specific transcripts and low IEG activation (Figure S2C)
like in prior scRNA-seq studies of human islets (Segerstolpe
et al., 2016). A pseudo-bulk average analysis of our patch-
seq and FACS dataset confirmed these results (Figures S2D
and S2E). We further investigated the effects of cell culture
by collecting a subset of patch-seq cells on the same day of
dispersion. We found an initial increase in IEG transcripts that
 ll
Resource

A B C 10
Cell Size (pF)
0
Patch-clamp Cell collection β-cells γ-cells 16

GCG
for scRNAseq α-cells δ-cells 8
acinar 0
16

INS
log2 (CPM)
8

tSNE2
0
Measurements: Lysis 16

SST
Cell Size buffer 8
Exocytosis 0
Ca2+ Currents 16

PPY
Na+ Currents
8
secreted 0
endocrine cell

PRSS1
granules 16
tSNE1 8
0
cells
D E
α-cells, 689
0 mV 1s 30 mV
-70 mV -20 mV 50 ms
1.0 -70 mV 500 ms

Na+ Current (I/Imax)

-120 mV
ND acinar, 18 b Membrane 0.8

T2D d 500 ms
γ-cells, 21
capacitance 0.6 h
other a
δ-cells, 24 0.4
g
0.2
e
c 0.0
-150 -100 -50 0
f
Current Voltage (mV)

β-cells, 268

F
Cell size Exocytosis Na+ currents Ca2+ currents
e
Relative Frequency (%)

Relative Frequency (%)

Relative Frequency (%)

Relative Frequency (%)

Relative Frequency (%)

50 30
Relative Frequency (%)

Relative Frequency (%)

30 30 25 20 f 25
a d g h
α-cells 25 40
20 20
β-cells 15
20 20 20
15 30 15
15 10
10 20 10
10 10 10
5 10
5 5 5

0 0 0 0 0 0 0
0 2 4 6 8 10 12 0 10 20 30 40 -20 0 20 40 60 80 10 -80 -60 -40 -20 0 10 20 30 40 0 5 10 15 -2 0 2 4 6
Capacitance(pF) Early Exocytosis (fF/pF) Total Exocytosis (fF/pF) Na+ Current Half- Peak Na+ Current (pA/pF) Early Ca2+ Current (pA/pF) Late Ca2+ Current (pA/pF)
Inactivation (mV)

Figure 1. Patch-Seq Measurements in Endocrine Cells

(A) Schematic of patch-seq.
(B) tSNE projection of patch-seq cells clustered by expression of over-dispersed genes.
(C) Cell size (as membrane capacitance) for each cell and expression of key marker genes. Color code as in (B). Dashed line indicates average value per cell type.
(D) Patch-seq cells collected in this study.
(E) In each cell we measured the following: a, early exocytosis; b, late exocytosis; c, Ca2+ integral; d, total exocytosis; e, Na+ current half-inactivation; f, peak Na+
current; g, early Ca2+ current; h, late Ca2+ current (not shown: cell size, exocytosis normalized to Ca2+, reversal potential, and Na+ and Ca2+ conductance).
(F) Distribution of selected parameters demonstrating functional heterogeneity of a (red) and b (black) cells. Inset letters (a–h) correspond to parameters in (E).
Distribution of exocytotic responses in a and d at 1 mM glucose (a cells) or 5–10 mM glucose (b cells).
See also Figures S1 and S2 and Table S1.

decreased after 1 day in culture, while transcripts encoding is representative of the total secretory capacity of a b cell and a
islet-specific genes recovered (Figures S2F and S2G). This key indicator of b cell function and dysfunction in T2D (Fer-
recovery was also observed for cells obtained from donors daoussi et al., 2015; Gembal et al., 1992). In this way, we found
with T2D (Figure S2H). Finally, patch-seq itself did not genes positively or negatively associated with b cell exocytotic
appear to impact gene expression since cells collected with capacity (Figure 2B). Among top correlates, we found several
or without patch-clamp overlapped in a tSNE representation genes linked to pathways thought to regulate insulin secretion,
(Figure S2I). including b cell transcription factors (MAFA and ETV1), mole-
cules associated with insulin granules (SLC30A8, VAMP2,
Patch-Seq Identifies Novel Genetic Regulators of SCG2, and INS), metabolic enzymes (PDK4, PDHA1, and
Human b Cell Physiology GYG1), and ion channels (ABCC9, KCNH2, KCNMB2, and
We applied the patch-seq approach to find genes associated NALCN) including the L-type Ca2+ channel encoded by
with electrical and exocytotic function in b cells. In these cells, CACNA1C (Rorsman and Braun, 2013; Lu et al., 2002; Ait-Lounis
we found that electrophysiological properties cluster by ‘‘func- et al., 2010; Zhang et al., 2005; Chimienti et al., 2004). Gene set
tional group’’ (exocytosis, Ca2+, and Na+ currents) and are enrichment analysis (GSEA) using correlation scores confirmed
uncorrelated with other parameters such as donor age, sex, or many of these pathways and revealed additional enriched
BMI (Figures 2A and S3A). We first correlated the transcriptome categories, such as neuronal regulators, transcription fac-
of b cells to total exocytosis at stimulatory glucose levels, which tors, and regulators of cell polarity or stress (Figures 2C and

Cell Metabolism 31, 1017–1031, May 5, 2020 1019

 ll
Resource

A B
Total Exocytosis
1.0 Cell Size
spearman r

0.8
Exocytosis VAMP2 4.0
0.6
Ca2+ channels

Total Exocytosis
0.4 MAFA 3.2
Na+ channels

(fF/pF, log)
0.2 SCG2 2.4
FAM159B 1.6
CACNA1C

Correlated
Cell size 0.8
INS
GPRASP1 0.0
0.71 0.58 0.52 0.38 Exocytosis Norm Ca2+
RGS9
Exocytosis

OGDHL 1.5
0.71 0.85 0.76 0.33 Early Exocytosis

Gene Expression
KCNH2
1.0
0.58 0.85 0.98 0.35 Total Exocytosis SLC16A9

(z-score)
0.5
SLC30A8
KCNMB2 0.0
0.52 0.76 0.98 0.34 Late Exocytosis
−0.5
-0.91 0.33 Peak Na+ Current −1.0
Na+

Anticorrelated
−1.5
-0.91 Na+ Conductance

0.33 0.35 0.34 -0.62 -0.59 -0.68 Late Ca2+ Conductance 36

RBP4
32
0.38 -0.62 0.73 0.72 Ca2+ Charge Entry PDHA1
Ca2+

BMI
GYG1 28
0.33 -0.59 0.73 0.77 Early Ca2+ current (pA/pF) 24
DLD
20
-0.68 0.72 0.77 Late Ca2+ current (pA/pF)
ID2
ME1
BMI
Donor
Sex

Other ion channel

glucose and related:
ATP1B1
C D KCNH2
KCNMB2
GYG1
glucose-6-phosphate glycogen ABCC9
NAGK NALCN
pyruvate Nucleus:
MAFA
ME1 ETV1
CELL ADHESION MOLECULES PC PDHA1
PDK4 NKX6.1
DLD CACNA1C
REGULATION OF BETA CELL DEVELOPMENT malate TCA
SLC25A6
ATP CACNA2D2
ADP
ION CHANNEL TRANSPORT OGDHL

NOTCH SIGNALING PATHWAY Ca2+

REGULATION OF GENE EXPRESSION IN BETA CELLS Granule

proteins:
ION TRANSPORT BY P TYPE ATPASES GPCR signalling: INS
TYPE II DIABETES MELLITUS GPRASP1 SCG2 Vesicular Membrane fusion:
RGS9 membrane: AP3B1
ALDOSTERONE REGULATED SODIUM REABSORPTION SLC30A8
SSTR5-AS1 SNAP25
NEURONAL SYSTEM PRKACB SYN1 VAMP2
PROTEASOME PRKAR1A NAPA
STX3
GLYCOLYSIS GLUCONEOGENESIS AP2A1
PATHOGENIC ESCHERICHIA COLI INFECTION
n.s.
SIGNALING BY WNT
****
MITOCHONDRIAL PROTEIN IMPORT E
DESTABILIZATION OF MRNA BY AUF1 HNRNP 100
Total Exocytosis (fF/pF)

APOPTOSIS
IL1 SIGNALING 75
RESPONSE ELEVATED PLATELET CYTOSOLIC CA2+
-2 -1 0 1 2 50

Enrichment (NES)
25

0
l

9b

S9

1
L
tro

YG
H

15

G
PA
D
on

G
G

m
C

TS

si

si
O

Fa
si

si

si

Figure 2. Correlation of b Cell Exocytosis to Single-Cell Gene Expression and Pathway Analysis
(A) Spearman correlation of electrophysiological parameters shows clustering of each functional group and low cross-correlation across clusters. All parameters
are normalized to cell size.
(B) Heatmap of top genes correlated and anticorrelated to total exocytosis in b cells from ND at 5–10 mM glucose. Cells are sorted by exocytotic response from
highest (left, dark green) to lowest (right, yellow). Gene expression is shown as a Z score after smoothing (n = 20 cells). Metadata for each cell are shown at bottom
(BMI, donor, sex).
(C) Top enriched pathways in genes correlated (green) and anticorrelated (red) to total exocytosis (KEGG and Reactome databases, false discovery rate
[FDR] < 0.1).
(D) Summarized map of cellular location and pathways with genes correlated (green) and anticorrelated (red) to exocytosis.
(E) Exocytosis in b cells measured at 10 mM glucose following target gene knockdown compared to control cells from same donors (348 cells, n S 3 donors per
knockdown experiment).
**p < 0.01, ****p < 0.0001 (Mann-Whitney U test and Benjamini-Hochberg [BH] correction). See also Figure S3 and Tables S2, S3, and S4.

S3B). Islet transcription factors with weaker but significant asso- permitting visualization and analysis of gene expression signifi-
ciation to exocytosis included PAX6, FOXO1, and NKX6-1 cantly correlated with exocytotic function in b cells.
(Figure S3B). This analysis nominated multiple genes—without previously
The majority of genes whose expression correlated known roles in b cells—as candidate regulators of physiological
significantly with b cell exocytosis are islet specific and included function, including OGDHL, FAM159B, TSPAN1, RGS9, and
candidate T2D risk genes like YWHAG (Fernández-Tajes et al., GYG1. Prior studies have linked OGDHL to metabolism, while
2019). Top gene correlates to exocytosis also included regula- FAM159B and TSPAN1 encode membrane proteins and RGS9
tors of oxidation and detoxification that could mitigate T2D- specifies a regulator of G-protein signaling (Bunik et al., 2008;
associated stress (Otter and Lammert, 2016), such as Danielsson et al., 2014; Uhlén et al., 2015). To test our predic-
glutathione peroxidase and transferases (GPX3, GSTK1, and tions, we performed small interfering RNA (siRNA) knockdown
GSTA4; Table S2). We integrated these results into a map that followed by patch-clamp for this set of genes in islets from an
combined correlations of transcript levels with four exocytosis additional 8 ND donors (Table S3; Figure 2E). Knockdown of
measures (early, total, late, and normalized to Ca2+; Figure 2D), each gene, confirmed by qPCR (Table S4), was followed by

1020 Cell Metabolism 31, 1017–1031, May 5, 2020

 ll
Resource

A Exocytosis Ca2+ Na+ % expressed

positively
correlated
negatively
correlated
B
ABCC9
ACIN1
AP3B1 Exocytosis Ca2+ Na+ % expressed
Exocytosis BMP5 RBP4 1

Norm. ephys (log) Norm. expression

ATP1B1 ADK
BIRC2 ADSL
BMP5
C21orf91
ANP32B
ATP5B
ρBMP5=0.28(8)
C9orf142
CACNA1C
CHCHD7 ρRBP4=-0.26(8)
COPS3
CAMK2B COQ6 ρSYN1=0.17(8)
CCDC57 CTSB
CFC1 CYC1 ρMAFA=0.24(8) 0
CHGA DHPS
CLSTN1 DLD
CMC1 DNAJC8 Peak Na+ current SYN1 MAFA
COTL1
DMKN
DNTTIP2
EIF4EBP1 1
FAM159B
F3 GET4
GSTO1
ρBMP5=0.14(9)
GALK2 GYG1 ρRBP4=-0.44(7)

tsne 2
GPRASP1 H1F0
GPRIN1 HGS ρSYN1=0.28(8)
GPX3 ID2
GRIA4 ILF3-AS1 ρMAFA=0.17(8)
GSTK1 IMP3 0
IFI6 JTB 100 tsne 1
KCNH2 LMBRD2
KCNMB2 MAGEH1
LRCH4 80
MRFAP1L1

% cells expressed
MAFA MTHFD2
MDK MTND4P12 60
MRAP2 NDUFA9
MT-ND1
NAXD
NOL7
PCGF1 40
C
NEO1 PFN1
NPTX2 PNPO
OGDHL 20
POLR2G
PAXBP1 PSMB7
PDK4 RBP4 0
PIGX RSF1
PLEKHF2 RUVBL2
RAI1 SARS
SCG2 SCAND1
SEPT3 SEC13
SIN3A SIGMAR1
SLC16A9 SLC25A6
SLC30A8 SLC3A2
SNX27 SLC9A3R1
STAG2 SNF8
STAT3 SNRNP40
SYN1 SNRPB
TPPP SRXN1
TRIP12 STARD3NL
TSPAN2 STK25
VWDE TBCC
WDFY2 TCEB2
WDR59 UGDH
La Exo Size
Ex Ea Exo osis

Ca 2osis xocy is
Ea ha rm C is
rly rge a 2+
y
C nt

Na + Na + tan t
Co Cu e
uc nt
ce

La Exo Size
Ex Ea Exoc osis

Ca 2osis xocy is
Ea ha rm C is
rly rge a 2+

C ry
C nt

Na + Na + tan t
Co Cu e
uc nt
ce
uc n
E ytos
tos

La La a 2+ Entr

uc n
c

E ytos
tos

c
e
Pe ond urre

nd rre
tan

t
e
Pe ond urre

nd rre
tan
Ca 2 En
Ca 2 Ca 2 urr
t

Ca 2 Ca 2 urr
t
cy
tal Cell

cy
tal Cell
c

C
+
C o

+
C o
n

n
te
oc rly

te
oc rly
C

C
ak

ak
te te
To

te te
To

+
yt

yt

La La

D E
Total Exocytosis Late Ca2+ Conductance Peak Na+ Current Cell Size
= 0.59, P = 2e-10 100 30
= 0.54, P = 8e-09 = 0.51, P = 9e-08 0.8
= 0.46, P = 0.02 = 0.41, P = 0.02 = 0.64, P = 0.0001 Na+ Total
* *
Predicted k-NN

0.6
0 6
Predicted k-NN

Predicted k-NN

80 Conductance Exocytosis Cell Size Exocytosis

20 20 Ca2+ genes Na+ genes
0.4
4 genes genes
60
Peak Na +
0
0.2 Early
10 40 10 Current * 0 Exocytosis
20
0 0
0 Late Ca2+ * * Late
Conductance Exocytosis
0 20 40 0 50 100 0 20 40 Early Ca 2+
Measured patch-clamp Measured patch-clamp Measured patch-clamp Predictive Set genes
Ca2+ Current Charge Entry
random genes
(95% CI)

Figure 3. Gene Networks Associated with b Cell Functional Heterogeneity

(A) Genes with significant correlation to electrophysiology (Z score > 2 for n > 5 parameters; STAR Methods). Significant positive and negative correlations are
indicated in red and blue, respectively.
(B) tSNE projection of b cells from ND using genes in (A). Spearman rank correlation between each parameter and gene is shown as mean and error in last digit
(STAR Methods).
(C) Network of genes connecting different functional groups (‘‘PS genes’’). Genes (small dots) connecting different functional groups (large dots) are selected if
they show significant correlations (Z score > 2) to at least one functional parameter in each group. Edge color indicates positive/negative correlations (red/blue),
and gene color identifies clusters connected to the same functional groups.
(D) Predictions of the k-NN model using ‘‘PS genes’’ for training (black) and validation (red) sets. Spearman correlation and p for each parameter are indicated
as inset.
(E) Performance of the model using the ‘‘PS genes’’ (dark gray), measured as Spearman correlation between experimental and predicted values. Comparison to
model using 484 random genes of equivalent expression in b cells (10,000 permutations, 95% CI interval in light gray). Asterisk indicates parameters with p < 0.05
in validation set. Smaller spider plots show performance of the ‘‘PS gene set’’ (gray) versus top correlated genes to exocytosis (pink), Ca2+ (green), Na+ (orange),
and cell size (blue).
See also Figure S4 and Table S5.

electrophysiological characterization at 10 mM glucose and in- stimulated at these high-glucose conditions. Together, these
sulin immunostaining to confirm cell type. In 4/4 cases of genes data show that patch-seq analysis robustly identified previously
positively correlated to exocytosis (OGDHL, FAM159B, unknown regulators of human b cell physiology.
TSPAN1, and RGS9), we observed significant reduction of the
exocytotic responses after knockdown (Figure 2E). By contrast, Prediction of b Cell Electrophysiology from Gene
knockdown of GYG1, whose transcript levels are anticorrelated Expression
with exocytosis, did not lead to significant changes of exocy- To identify and investigate additional genes linked by patch-seq
tosis, possibly reflecting that b cell activity is already maximally to b cell excitability, we broadened our analysis to include

Cell Metabolism 31, 1017–1031, May 5, 2020 1021

 ll
additional electrophysiological parameters (Ca2+ currents and
Na+ currents) and selected transcripts showing consistent posi-
tive or negative correlations to multiple parameters as likely reg-
ulators of b cell physiology in nondiabetic states (Figure 3A). This
analysis identified several key genes, including genes previously
associated with b cell transcriptional heterogeneity (Baron et al.,
2016; Dorrell et al., 2016; Rui et al., 2017; Segerstolpe et al.,
2016). For instance, transcripts encoding retinol binding protein
4 (RBP4) correlated negatively with cell size, Na+ currents, and
total exocytosis, while transcripts encoding the b cell surface
protein FAM159B correlated positively to exocytosis, Ca2+ entry,
and Na+ currents. A tSNE projection using these highly corre-
lated genes shows a gradient of functional measures across b
cells that overlaps with patterns of gene expression (Figure 3B).
To understand further how genetic pathways are connected to
each major group of physiological function (exocytosis, Ca2+
currents, and Na+ currents), we repeated GSEA on their aver-
aged correlation scores (Figure S3C). Results for overall exocy-
tosis are consistent with those reported above (Figure 2C); by
contrast, GSEA further identified that Na+ and Ca2+ currents
are linked to pathways related to increased excitability and circa-
dian rhythms, respectively (Perelis et al., 2015). Several of the
observed pathways identified by GSEA on averaged correlation
scores also overlap with those recently implicated in islet
dysfunction in T2D using genomic data (Fernández-Tajes
et al., 2019).
We then asked whether genes that correlate to multiple groups
of physiological function (‘‘functional groups’’) could be used to
develop predictive algorithms that link transcriptional signatures
to b cell function. To do so, we generated a network of genes with
significant correlation to more than one functional group (e.g.,
Ca2+ and exocytosis) and selected those with highest median
expression (Figure 3C; Table S5). This list of highly connected
genes, which we termed our predictive set (PS), contains genes
(1) previously linked to b cell excitability, (2) with heterogenous
expression in b cell subpopulations and in T2D (FFAR4,
FXYD2, and ID4), and (3) with unknown function in b cells (Table
S5). We then attempted to predict the measured electrophysi-
ology of each cell from the average values of its nearest neigh-
bors (NN) in PS gene expression (n = 484 genes, k-NN with 5
neighbors) (STAR Methods). We obtained significant correla-
tions between the experimentally measured patch-clamp pa-
rameters and the predictions from the k-NN model (Figures 3D
and S4A, black). We compared the performance of PS genes
to randomly selected genes of equivalent expression (n =
10,000 permutations) and found that the PS genes performed
significantly better for all parameters (Figure 3E). We also tested
our model by comparing its predictive power to that obtained us-
ing the genes most correlated to each functional group. We
found that PS genes showed consistent performance across
all measured parameters, whereas group-specific gene sets
only performed well for their subset of parameters (Figures 3E
and S4B). Finally, we validated the predictive power of PS genes
with a ‘‘test’’ set of data including gene expression and functional
parameters of cells withheld from the analysis that generated the
PS gene set. This analysis demonstrated appropriate recovery of
significant correlations for 5 of the measured functional parame-
ters (Figures 3D, S4A, red, and 3E). Thus, patch-seq and network
analysis combined with machine learning generated unprece-
1022 Cell Metabolism 31, 1017–1031, May 5, 2020
Resource
dented algorithms that reliably link gene expression to b cell
function.
Markers of b Cell Heterogeneity Correlate to b Cell
Excitability
Within the identified ‘‘PS gene set’’ we found that transcripts en-
coding RBP4 are significantly correlated with b cell functional het-
erogeneity. Prior studies have noted heterogeneous expression of
RBP4 in b cell subsets, but did not establish direct links to func-
tional heterogeneity (Baron et al., 2016; Dorrell et al., 2016; Rui
et al., 2017; Segerstolpe et al., 2016). Other studies show that
RBP4 is an adipokine with roles in homeostatic regulation of meta-
bolism (Broch et al., 2007; Tritschler et al., 2017). We observe that
RBP4 transcript levels have significant correlation with multiple b
cell functional phenotypes; for example, it is the ‘‘PS gene’’ with
strongest anti-correlation to b cell Na+ channel activity. RBP4+ b
cells have decreased exocytosis and Na+ currents despite having
normal Ca2+ current activity (Figure 4A). Consistent with this, we
observed that RBP4+ b cells also have significantly reduced
expression of key regulators of stimulus-secretion coupling, like
KCNJ8, ABCC9, and SCN3A (Figure 4B). In rodent b cells,
Scn3a encodes the principal physiologically relevant Na+ channel
(Zhang et al., 2003). Together, these results provide evidence that
RBP4 expression is a marker of b cells with reduced function and
show that previously identified markers of b cell transcriptomic
heterogeneity identify subpopulations with heterogeneous func-
tion (Dorrell et al., 2016; Johnston et al., 2016).
In T2D, b Cells Induce Genes Linked to Increased and
Decreased Exocytosis
Islets from the 7 donors with T2D (Table S1) showed reduced in-
sulin content and secretion (Figure 5A), and impaired b cell
exocytosis (Figures 5B and S5A). All b cells were obtained from
donors with recent T2D onset (<9 years), and who were not
receiving insulin treatment (as we did not record any b cells for
R241, the only donor on insulin treatment; STAR Methods). To
identify genetic drivers of b cell dysfunction in T2D, we first as-
sessed transcripts that we previously identified to be significantly
correlated to exocytosis in b cells from ND (Figure 2B; Table S2).
Unexpectedly, we found that genes that positively correlate with
exocytosis in b cells from ND are upregulated in T2D, while genes
that negatively correlate with exocytosis in b cells from ND had
reduced expression in T2D (Figure 5C). We found that this tran-
scriptomic shift is also observed in b cells from obese ND individ-
uals (Figure S5B). These data suggest that b cells attempt to alter
transcript levels to increase functionality, perhaps in response to
increased insulin demand associated with T2D and obesity. But
in T2D this ultimately fails to maintain an increase of b cell func-
tion, since their exocytotic response is significantly impaired
compared to ND controls (Figure 5B).
To find genes that could underlie this effect, we performed
correlation analysis in cells from donors with T2D and integrated
the ND and T2D results in a correlation map of exocytosis with
the overall changes in gene expression (Figure 5D; Table S2).
Next, we performed a pathway analysis for each of four gene
subsets (ND/T2D and correlated/anticorrelated to exocytosis)
(Figures 5E and S5C). Our data show that the pathways associ-
ated with reduced exocytosis in T2D are distinct from those
associated with low exocytosis in ND. While we found that low
 ll
Resource

A
Na+ currents Ca2+ currents Exocytosis
*** * * * *
17.5

Late Ca2+ Conductance (pS/pF)

60 60
3.0

Ca2+ Charge Entry (pC/pF)

Early Ca2+ Current (pA/pF)

1.50 8 120
Peak Na+ Current (pA/pF)

Na+ Conductance (pS/pF)

Late Ca2+ Current (pA/pF)

1000 40

Early Exocytosis (fF/pF)

Total Exocytosis (fF/pF)
15.0

Late Exocytosis (fF/pF)

50 50
1.25 2.5 100
800 6 12.5
40 40 30
1.00 2.0 80 10.0
30 600 1.5 30
0.75 4 60 20 7.5
20 400 0.50 1.0 40 20 5.0
2 10
10 200 0.25 0.5 20 10 2.5
0 0.00 0 0.0 0 0 0 0.0
0
RBP4+
B Na+ channels ATP-sensitive K+ channel Ca2+ channels RBP4-
log2(CPM+1)

** ** ***
10
5
0
100
75
% cells

50

25

0
KCNJ11
SCN1B

SCN2A

SCN3A

SCNM1

SCN3B

SCN9A

KCNJ8

ABCC8

ABCC9

CACNA1A

CACNA1C

CACNA1D

CACNA1H

CACNA2D1

CACNA2D2

CACNB2
Figure 4. Functional and Transcriptomic Differences in b Cell Subpopulations
(A) Electrophysiological function in RBP4 subpopulations of b cells. Significantly lower Na+ currents and exocytosis is observed for RBP4+ cells.
(B) Gene expression values (top) and % b cells with detectable expression (bottom) of Na+, ATP-sensitive K+ channels, and Ca2+ channels.
***p < 0.001, **p < 0.01, *p < 0.05 (Mann-Whitney U test with BH correction).

exocytosis in b cells from ND is linked to metabolic pathways and STAT3. These results indicate that optimal insulin secretion
glucose metabolism, the dysfunction in T2D is related to immune may require balanced expression of ETV1 and its downstream
response, cell-cycle pathways, and altered transcription factor targets, and that an imbalance may lead to dysregulated secre-
expression. For instance, we observe induction of pathways tion during early T2D (Figure 5G). To test the differential role of
like NFkB signaling, cell-cycle checkpoints, and auto-degrada- ETV1 in b cell physiology, we performed siRNA knockdown of
tion of the ubiquitin ligase COP1. Genes in T2D that show ETV1 followed by patch-clamp in an additional set of human is-
increased expression associated with reduced b cell exocytosis lets. We found that knockdown of ETV1 rescued exocytosis in b
included known regulators of immune and inflammatory path- cells from donors with T2D, whereas it did not alter exocytosis in
ways like NFKBIA, IL6R, and IRAK1, and the transcription factors b cells from ND (Figures 5H and S5G), further supporting a signif-
ETV1 and STAT3 (Figures 5D–5F and S5D). In particular, we icant role for ETV1 in b cell dysfunction in T2D.
found ETV1 to be significantly enriched in b cells from donors
with T2D, and to show ‘‘reversed’’ correlations to exocytosis in Transcriptional and Physiological Heterogeneity
T2D and ND. Overall, we found that in T2D, increased ETV1 Observed in a Cells
mRNA levels were associated with reduced b cell exocytosis Cell size and Na+ channel properties are often used to identify a
(Figure 5D, yellow box). These results were consistent across and b cells in rodents (Briant et al., 2017; Zhang et al., 2014);
several measured exocytosis parameters (Figures S5E and however, this is not generally applicable to humans, making it
S5F). Hence, our results suggest that dysfunctional cells in difficult to distinguish islet cell types at the time of patch-clamp-
T2D express higher levels of ETV1 than functional ones, whereas ing. For example, a cells (definitively identified post hoc from
the opposite happens in ND. scRNA-seq) were frequently mis-classified as possible b cells
The ETV transcription factors impair insulin secretion in hyper- during our initial patch-clamp measures of capacitance that indi-
glycemic mouse models and are negatively regulated through cated cell size (>4 pF; Figure 1F). To improve cell-type identifica-
COP1-dependent targeted degradation (Figure 5G) (Suriben tion prior to scRNA-seq, we implemented a machine learning
et al., 2015). We tested genes previously reported to show algorithm based in ‘‘random forests’’ (Breiman, 2001) to classify
ETV-dependent expression in mouse islets (Suriben et al., cell types from their electrophysiological profile. We trained and
2015) and found several human orthologs with correlating validated the model in cells from ND donors using 10-fold cross-
expression in single b cells, including MAFA, SLC30A8, and validation, obtaining an AUC of 0.95 in the validation dataset

Cell Metabolism 31, 1017–1031, May 5, 2020 1023

 ll
Resource

A *
B ** ** n.s. C
4000 * 2.5 ND 40 40 10
***

Early Exocytosis (fF/pF)

Total Exocytosis (fF/pF)
35 35

Late Exocytosis (fF/pF)

Insulin Content (ng/ml)

T2D
2.0 8
3000 Insulin Secretion 30 30
(% content) genes anticorrelated
25 25 6
1.5 to exocytosis in ND
2000 20 20
genes correlated
1.0 15 15 4 to exocytosis in ND

1000 10 10
0.5 2 −0.4 −0.2 0.0 0.2
5 5 median logFC (T2D vs ND)
0 0.0 0 0 0
ND T2D 1 mM 10 mM ND T2D ND T2D ND T2D
Glucose

Increased β-cell exocytosis in ND

D E
Pathways
log2FC (T2D vs ND)

1.5 Relative diff. in GPD2

% cells expressing
gene (normalized):
ND high exocytosis genes (top level) i ii iii iv
0.0
0.0 5 SNARE interactions in vesicular transport
0.4 DDX46 CTSF PDK4 RGS9
Regulation of beta-cell development
0.25

-log10 (FDR )
−1.5 0.5 4 Type II diabetes mellitus
FXYD6 Cell adhesion molecules (CAMs)
STXBP5-AS1 MRAP2
3
Glycosaminoglycan biosynthesis
SCG5
2 and heparan sulfate
0.2 1 Pyruvate metabolism

Decreased β-cell exocytosis in T2D

RBP4 TSPAN2
Correlation to Exocytosis (T2D)

CACNA1C
TCF7L2
TSPAN1 0
MAFA

KCNH2
Vesicle-mediated
SLC30A8
transport Metabolism of carbohydrates
0.0 Developmental Biology
Metabolism Glycolysis
PDHA1 Disease
TNS2 Neuronal System
ENO1 GUCY1A3
ID2 Signaling molecules
−0.2 and interaction Autodegradation of the E3
ADAM10 ITGB1
VAMP2
Transcription / ubiquitin ligase COP1
CTNNAL1
IL6R TXNDC11
DNA replication Cell Cycle Checkpoints
Immune System
Cell Cycle Activation of NF-κβ in B Cells
−0.4 STAT3
Transport of small
T2D low IRAK1
ETV1
exocytosis genes AP1S2
IFNAR1 genes with molecules ND T2D ND T2D
reversed correlation Hemostasis
PRELID1 NFKBIA high low
exocytosis exocytosis
−0.6
−0.4 −0.3 −0.2 −0.1 0.0 0.1 0.2 0.3 0.4
Correlation to Exocytosis (ND)
H 100
Total Exocytosis (fF/pF)

F ETV1 G Increased
75
insulin demand
**
1
optimal
range
***
10
log2 FC (T2D-ND)

Function

dysfunction dysfunction 50
log2(CPM+1)

ETV1 STAT3
5 0 ***
ETV1 / 25
STAT3 / COP1 inflammation /
0 −1 immune genes β-cell
dysregulation 0
STAT3

ND T2D
RBP4
ETV1

IL6R

Expression
ctrl siETV1 ctrl siETV1

ND T2D

Figure 5. Functional and Transcriptomic Changes in b Cells Early in T2D

(A) Insulin content and secretion (as % content) for donors included in this study. *p < 0.05, ***p < 0.001 (Student’s t test; two-way ANOVA and Tukey post test).
(B) Measured exocytosis for b cells in ND and T2D. **p < 0.01 (Mann-Whitney U test and BH correction).
(C) Enrichment analyis in T2D of genes found to be correlated (red) or anticorrelated (blue) to exocytosis in cells from ND. Data show median log fold change in
gene expression between T2D and ND for each subset of genes. Error is SEM. Central bar shows range of log2 fold change obtained by sampling random genes of
equivalent expression (10,000 iterations, 400 genes, 5%–95% range).
(D) Gene correlation map of exocytosis. Scatterplot shows correlation to exocytosis in ND (x axis) and T2D (y axis) for each gene. Genes with significant cor-
relations (Z score > 2) are colored according to their fold enrichment in T2D cells (red, upregulated in T2D; blue, downregulated in T2D), and size is proportional to
relative change in % expression. Regions of interest are highlighted with dotted boxes. Genes with non-significant correlations are shown in gray.
(E) Enriched pathways for genes correlated to exocytosis in ND (1) and T2D (2), and genes anticorrelated to exocytosis in ND (3) and T2D (4). Enrichment is shown
as log10(FDR). Left bar indicates top category of each pathway.
(F) Distribution of ETV1 expression (left) and enrichment of a subset of genes that change between ND and T2D.
(G) Model showing the hypothesized role of ETV1, STAT3, and immune pathways in b cell dysfunction in early T2D. Based on Suriben et al. (2015).
(H) Exocytosis in b cells measured at 5 mM glucose following ETV1 knockdown compared to control cells from same donors (n = 3 donors both for ND and T2D,
***p < 0.001 Kruskal-Wallis test and Dunn’s post test).
See also Figure S5.

(Figure 6A). This model correctly identified 85% (92%) of a cells in cells from T2D donors (AUC = 0.93) (Figure 6A). Compared
(b cells), misidentifying 15% of a cells and 8% of b cells (Fig- with a capacitance-based cell size cut-off at the time of patch-
ure 6B). Features from every functional group contributed to clamping, our model significantly improved the a priori identifica-
the classifier (Figure 6C), and the model also performed robustly tion of a cells and reduced their mis-classification as b cells

1024 Cell Metabolism 31, 1017–1031, May 5, 2020

 ll
Resource

A B D

0.8

0.6

0.4

0.2
1.0 Original prediction at

α-cells
0.85 0.15 the time of patch-clamp
True Positive Rate

True
0.8

α-cells
β-cells
0.08 0.92
0.6
α-cells β-cells
ND α-cells, validation Predicted
0.4 ND β-cells, validation
Predicted by model
AUC = 0.95 Feature importance
C

α-cells
0.2 T2D α-cells 0 0.2 0.4
T2D β-cells Cell Size
AUC = 0.93 Peak Na+ Current
Early Exocytosis
0.0 Total Exocytosis
0.0 0.2 0.4 0.6 0.8 1.0 Na+ Conductance correctly predicted
False Positive Rate Late Exocytosis misidentified as β-cells
Early Ca2+ Current
Late Ca2+ Current
remaining unclassified
Ca2+ Charge Entry

E
Capacitance (Cell Size) DDIT3 PPP1R15A SQSTM1 XBP1
ρPPP1R15A=-0.44(3)
ρLOXL4=0.40(3)
1

Norm. expression
ρFEV=0.37(4)
ρSCG2=0.57(3)
ρFFAR1=0.44(4)

Peak Na+ Current ARX LOXL4 FEV MAFB

ρPPP1R15A=-0.24(4) 0
ρLOXL4=0.32(4)
ρFEV=0.26(4)
1

Norm. ephys (log)

ρSCG2=0.33(4)
ρFFAR1=0.33(4)
Na+ Conductance SCG2 FAP FFAR1 GPR119
ρPPP1R15A=-0.22(4)
ρLOXL4=0.32(4)
ρFEV=0.26(4) 0
tSNE2

ρSCG2=0.30(4)
ρFFAR1=0.23(4)
tSNE1

Figure 6. Transcriptomic and Electrophysiological Heterogeneity in a Cells

(A) ROC curve of cell-type prediction using random forests for the validation dataset in ND (red, blue) and in T2D (green, purple).
(B) Confusion matrix showing accuracy of predictions in ND validation dataset.
(C) Contribution of each feature to the random forest model.
(D) Comparison of a cell identification from predictions at the time of patch-clamping (simple cell size cut-off) versus the model.
(E) tSNE plots showing heterogeneity in gene expression of a cells using over-dispersed genes and normalized electrophysiological measurements.
See also Figure S6.

(Figure 6D). Analysis of a cell transcriptomes revealed significant
transcriptional heterogeneity in islets from multiple human do-
nors, including those with T2D. In particular, a subset of cells
showed enrichment of regulators of a cell maturation (LOXL4,
MAFB, and ARX), regulators of reduced ER stress (DDIT3,
XBP1, and PPP1R15A), transcription factors governing endo-
crine fate (FEV and ISL1), receptors involved in glucose homeo-
stasis (FFAR1 and GPAR119), and secretory pathways (SCG2)
(Figures 6E, S6A, and S6B). This transcriptional heterogeneity
correlated to electrophysiological features including Na+ cur-
rents and cell size (Figure 6E); further work will be required to un-
derstand the relationship of this heterogeneity to a cell dysfunc-
tion in T2D. Thus, in addition to improving prospective
identification of patch-clamped human a cells, our studies pro-
vide evidence for molecular heterogeneity associated with func-
tional heterogeneity in a cells.
Application of Patch-Seq to Rare Cryopreserved
Samples: Islet Cells in T1D
Tissue banking programs can provide unique access to cells from
subjects with specific disorders. Hence, we investigated whether
the patch-seq approach could be extended to rare cryopre-
served islet samples (Manning Fox et al., 2015). We found that
single-cell transcriptomes from cryo-banked islets had compara-
ble quality metrics to those of fresh tissue (Figure S7A), allowing
us to investigate cells from three donors with T1D and three

Cell Metabolism 31, 1017–1031, May 5, 2020 1025

 ll
Resource

A B

α-cells

β-cells

δ-cells
γ-cells
acinar

ductal
PSCs
β-cells γ-cells acinar T1D N total
α-cells δ-cells ductal ND PSCs 16
acinar
PSCs 133 215
1.0 12

0.8 α-cells

Fraction of cells

log2 (CPM+1)
8
0.6
tSNE2

0.4
β-cells 4
0.2
ductal
0.0
ND 1

T1D
γ-cells
tSNE1
δ-cells 0

CXCL8
MMP2
SPARC
DCN
NNMT
MSC
ANXA1
COL1A2
PLIN2
IL24
PRSS2
CTRB2
CTRB1
SPINK1
REG1A
BCAT1
PRSS1
MGST1
AKR1C3
CPA2
GCG
TTR
LOXL4
VIM
GC
RGS4
IGFBP2
HIGD1A
RSAD2
GPX3
INS
IAPP
MEG3
RBP4
HADH
NPTX2
ERO1B
IGF2
PDX1
GSN
TINAGL1
ANXA3
TFPI2
KRT19
CAV2
KRT7
BICC1
PMEPA1
TACSTD2
CRIM1
PPY
GPC5-AS1
ETV1
AQP3
GCNT3
CARD11
FXYD2
PXK

SST
THSD7A
SLITRK6
HHEX
RBP4
LEPR
FAM84B
EDN3
CD9
MS4A8
SLC38A1
HAP1
C β-cells

T1D D
ND 100% β-cells α-cells ND
% cells

75% 20 20
log2 (CPM+1)

log2 (CPM+1)
T1D
CHGA
CHGB

NKX2-2

NKX6-1
PDX1

MAFB
MAFA
GCG

ETV1
INS

50% 15 15
25% 10 10
5 5
0 0
α-cells
mean expression

0.8
0.7
F P4

-D 4

S
A

M K1

M 3
C 1
G 7
G 2
3
EM 163

FK B
LD IA

AD 6

PS A1
H H

3
S1 A

N A
LA R

1
C
4
PX

TM PX

G 00A

D
M
FA

N 76

G 45

P
(norm.)

T1D

B
C

D
0.6
B

IN

PD
H FA

AC
TM EM

D
U
R

SP
ND 0.5
0.4
0.3
CHGA
CHGB

NKX6-1
NKX2-2
MAFB
GCG

ARX
INS

FEV

ETV1

0.2
F ** *
3.5
5

Early Ca2+ Current (pA/pF)

Late Ca2+ Current (pA/pF)

T1D β-cells T1D α-cells 3.0
E 4
ADAPTIVE IMMUNE SYSTEM IMMUNE SYSTEM 2.5
TYPE I DIABETES MELLITUS ADAPTIVE IMMUNE SYSTEM
IMMUNE SYSTEM INTERFERON GAMMA SIGNALING 2.0
GRAFT VERSUS HOST DISEASE ER PHAGOSOME PATHWAY 3
ALLOGRAFT REJECTION ANTIGEN PROCESSING CROSS PRESENTATION 1.5
ENDOSOMAL VACUOLAR PATHWAY ENDOSOMAL VACUOLAR PATHWAY 2
AUTOIMMUNE THYROID DISEASE CLASS I MHC ANTIGEN PROCESSING PRESENTATION 1.0
NK CELL MEDIATED CYTOTOXICITY TCR SIGNALING
INTERFERON GAMMA SIGNALING APOPTOSIS 1 0.5
ER PHAGOSOME PATHWAY INTERFERON SIGNALING
0 1 2 3 4 0 1 2 3 0 0.0
-log10(FDR) -log10(FDR) ND T1D ND T1D

Figure 7. Patch-Seq in Cells from Cryopreserved T1D Islets

(A) Left: tSNE projection of patch-seq cells clustered by expression of over-dispersed genes. Right: cell types and total number of cells obtained for ND and T1D.
(B) Marker genes for each cell type.
(C) Expression of key identity genes on a and b cells from donors with T1D and ND matched controls.
(D) Representative genes obtained in a differential expression analysis between T1D and ND for b and a cells.
(E) Pathways enriched in upregulated genes in T1D a and b cells.
(F) Distribution of Ca2+ parameters showing statistically significant differences between a cells of donors with T1D and ND.
**p < 0.01, *p < 0.05 (Mann-Whitney U test with BH correction). See also Figure S7 and Tables S6 and S7.

controls matched for BMI, age, sex, and storage time (Table S6). script enrichment of genes related to immune activation and allo-
We performed patch-seq in 348 cells from these samples (Fig- graft rejection (HLA-DMA and FAS) (Figures 7D and 7E). In a cells
ure 7A; Table S6) and used a logistical regression model to iden- from donors with T1D, we found increased NKX6.1 and
tify marker genes for each population (Figures 7B and S7B). Sam- decreased NKX2.2 mRNA (Figures 7C and S7E) along with
ples from donors with T1D contained an unanticipated variety of decreased Ca2+ channel activity (Figures 7F and S7D), in general
cell types (Figure 7B), including an enrichment for pancreatic agreement with recent reports using bulk RNA-seq and immuno-
polypeptide-secreting g cells and ductal cells (Figure S7C). histochemistry (Brissova et al., 2018; Chakravarthy et al., 2017),
We obtained 11 b cells from two donors with T1D, consistent although we did not detect a consistent decrease in Ca2+ channel
with prior observations of b cell survival years after T1D onset gene expression as previously reported (Figure S7E). Our anal-
(Keenan et al., 2010; Morgan and Richardson, 2018) (Figure S7C). ysis of a cells in T1D also showed transcript enrichment of mucin
These b cells had similar electrophysiological profiles as cryopre- (MUC) and other genes typically associated with ductal cells, and
served b cells from ND controls (Figure S7D) and appeared to FEV1, a reported endocrine progenitor cell marker in mice (Fig-
maintain equivalent expression of hallmark b cell genes (INS, ures 7C and 7D; Table S7) (Byrnes et al., 2018; Liu et al., 2019).
PDX1, and MAFB) (Figure 7C). Differential expression analysis Together, our data demonstrate that patch-seq can be used in
showed decreased expression of RBP4 and FFAR4 and tran- rare cryo-stored tissue types, and supports the view that the

1026 Cell Metabolism 31, 1017–1031, May 5, 2020


Resource
transcriptomic and functional signatures of surviving b cells may
be preserved in T1D, while a cells lose characteristic functional
and transcriptomic phenotypes, consistent with the observation
of impaired glucagon regulation in T1D.
DISCUSSION
Success in identifying the genetic basis of islet function and
dysfunction in diabetes has been limited by an inability to con-
nect islet cell physiologic function with transcriptome regulation.
Elucidating mechanisms underlying diabetes also suffers from
limited human cell and tissue availability that hinders single-
cell-based investigations, particularly when considering T1D.
Multi-modal single-cell technologies will help address these is-
sues by increasing the depth of available data and by making it
possible to directly link transcriptomes to additional readouts
in individual cells (Macaulay et al., 2017; Stuart and Satija,
2019). One integrated approach previously used in studies of ro-
dent brain slices is patch-seq (Cadwell et al., 2016; Földy et al.,
2016; Fuzik et al., 2016). Like neurons, endocrine cell secretion is
coupled to dynamic electrophysiological mechanisms that
culminate in regulated exocytosis of secretory vesicles. Methods
that allow simultaneous measurements of transcriptomes and
secretory capacity hold promise for understanding the mecha-
nisms by which gene expression enables physiological function
in these cells.
Here we describe the development of patch-seq for endocrine
cell types and report paired physiologic-transcriptomic data for
more than 1,300 human islet cells. This depth of data allowed
us to link transcriptional and physiological heterogeneity at
single-cell resolution, and to identify genes and pathways regu-
lating exocytosis of human islet b and a cells. With a patch-seq-
derived minimal PS of genes and machine-based learning tools,
we developed algorithms that correctly predicted hallmark phys-
iological functions like exocytosis from gene expression, and
accurately classified live isolated islet cell types. For studies of
islets from T1D and T2D donors, we used patch-seq to identify
molecular mechanisms underlying b and a cell dysfunction in
these diseases. Thus, our study provides a powerful resource
and toolset for simultaneous multiplex phenotyping of human
islet cells, in health or disease.
Patch-seq analysis identified key genes and pathways regu-
lating b cell exocytosis, including both known and previously un-
characterized regulators and mechanisms controlling insulin
secretion such as cell adhesion molecules and neuroactive
ligand receptors. Adhesion molecules have previously demon-
strated roles in b cell attachment and motility (Dahl et al., 1996;
Kaido et al., 2004), and are thought to govern b cell polarity
and function through control of directed insulin exocytosis into
the vascular axis (Geron et al., 2015; Parnaud et al., 2015; Gan
et al., 2018). Among neuronal regulators, we identified pairs of
factors such as neurexin 1 (NRXN1) and neuroligin 1 (NLGN1),
which are known to bind in a heteromeric complex (Suckow
et al., 2008). The identification of binding partners in our gene-
function correlation analysis confirms the robustness of this
approach to identify unanticipated candidate regulators of islet
cell function. Similarly, examination of tissue specificity reveals
that several of the encoded proteins correlating to b cell exocy-
tosis are also enriched in other excitable cell types (GPRASP1,
ll
RGS9, and MRAP2 in retinal tissue; GPRIN1 in brain; and
KCNH2 in cardiomyocytes) (Uhlén et al., 2015), indicating shared
mechanisms of signal transduction and showing that the analyt-
ical tools developed here could be applicable to patch-seq
studies in other excitable cells and endocrine tissues.
Regulation of islet exocytosis by metabolism is complex (New-
gard, 2017). We observed that multiple expected metabolic
pathways positively correlated with b cell functional measure-
ments (e.g., glucose flux and metabolism, circadian rhythm),
but also observed unexpected negative correlations of b cell
exocytosis to pathways like glycolysis and pyruvate metabolism.
We integrated several of the observed metabolic correlations to
b cell exocytosis in a functional map (Figure 2D), providing sup-
port for the hypothesis that glucose is preferentially used for TCA
anaplerosis through pyruvate carboxylase to facilitate insulin
secretion (Alves et al., 2015). These findings are also sustained
by measures of CPT1A and SLC16A9 expression, which
mediate the rate-limiting step of fatty acid oxidation and carnitine
efflux, respectively (Aichler et al., 2017; Newgard, 2017). Overall,
the observation of glucose homeostasis pathways across our
analysis also resonates with the importance of metabolites in
shifting transcriptional states of b cells, recently dubbed ‘‘meta-
bolic memory’’ (Rosen et al., 2018; Sharma and Rando, 2017).
Another useful outcome here is our development of a network
analysis tool that links a PS of 484 genes with applied machine-
learning techniques. We used this to predict b cell function solely
from RNA transcript abundance. Genetic regulators of b cell ac-
tivity, including many uncharacterized in islet biology, were iden-
tified in the predictive set, and included antioxidant molecules
(GPX3 and GSTK1), surface proteins (TSPAN1 and TSPAN2),
and b cell enriched molecules (NPTX2 and FAM159B). To vali-
date predictions from patch-seq correlations and this network
analysis, we used genetic loss-of-function approaches, which
provided evidence that previously uncharacterized genes like
OGDHL, RGS9, TSPAN1, and FAM159B are positive regulators
of b cell exocytosis. Prior studies have shown that FAM159B
may be regulated by the b cell transcription factor SIX3 (Arda
et al., 2016), and is a marker of b cell subsets (Dorrell et al.,
2016). Like FAM159B, other genes linked by our analysis to dif-
ferential b cell function were also found to be expressed in b cell
sub-populations (ID2, ID4, and RBP4) (Baron et al., 2016; Dorrell
et al., 2016; Rui et al., 2017; Segerstolpe et al., 2016). For RBP4,
we provide evidence that ND RBP4- b cells, compared to ND
RBP4+ b cells, have relatively superior electrophysiological func-
tion (also see discussion of T1D below). Overall, our data show
that b cells exhibit significant transcriptomic and electrophysio-
logical heterogeneity, supporting the idea that b cells comprise a
spectrum of cellular states. Further studies are needed to
address whether this heterogeneity corresponds to dynamic or
static states and the existence of mechanisms allowing cells to
switch between them, or possibly to a hierarchical physiological
organization of b cells (Johnston et al., 2016). Similarly, further
application of the PS gene signature to FACS datasets in disease
settings (T2D) would enable high-throughput functional pheno-
typing of b cells.
Understanding the basis of islet adaptation or dysfunction in
diabetes is another important result described here. Our work
shows that the pathways associated with reduced b cell exocy-
tosis in T2D significantly differ from those correlated to low
Cell Metabolism 31, 1017–1031, May 5, 2020 1027
 ll
exocytosis in cells from ND. Furthermore, the observation that b
cells from donors with T2D or obesity show increased expres-
sion of genes that are correlated to a high exocytosis phenotype
in ND suggests molecular mechanisms for functional compensa-
tion by b cells in early-stage T2D. However, our analysis also re-
veals as-yet unexplained complexity in b cell responses; for
example, ETV1 and STAT3 mRNA are positively correlated
with ND b cell functional phenotypes, and also with b cell
dysfunction in T2D. To address this, we performed genetic
loss-of-function experiments that showed recovery of exocytotic
capacity after knockdown of ETV1 in b cells from donors with
T2D. Recent studies suggest that b cell ETV1 and STAT3 protein
degradation is regulated by the ubiquitin ligase complex (Dalla-
valle et al., 2016; Suriben et al., 2015). Our work suggests that
ubiquitination and proteosomal degradation pathways in b cells,
including a pathway stimulating COP1 auto-degradation, are
associated with dysfunction in T2D. Thus, in addition to tran-
script-based mechanisms, our results suggest that post-transla-
tional mechanisms involving factors like COP1, ETV1, and
STAT3 may govern b cell dysfunction or responses in T2D
(O’Shea and Plenge, 2012; Saarima €ki-Vire et al., 2017; Suriben
et al., 2015).
Patch-seq also advances our understanding and ability to
study human a cells. Cell size and Na+ channel properties are
key identifiers used to distinguish rodent a and b cells (Briant
et al., 2017; Zhang et al., 2014), but human a cell heterogeneity
results in overlap of these electrophysiological properties. Using
patch-seq data, we developed additional models for improved
cell-type identification of live human islet a and b cells. We
also show that specific phenotypes like Na+ current activities
and cell size can vary significantly in a cells, and that this
functional heterogeneity corresponds well with expression of
transcripts specifying islet cell lineage (ARX, MAFB, and FEV)
or governing cellular stress (DDIT3 and PPP1R15A). Prior studies
have reported the impact of the ER stress response in b cell sub-
populations (Baron et al., 2016; Muraro et al., 2016) and in a cells
(Akiyama et al., 2013; Burcelin et al., 2008; Korsunsky et al.,
2019), suggesting that varying levels of ER stress could drive
altered or dysfunctional states in both islet cell types. We noted
a cell subpopulations expressing markers indicating either low or
high ER stress, both in ND and in T2D settings. Thus, patch-seq
merges electrophysiological and transcriptomic data to docu-
ment, and suggest a basis for, heterogeneous function and tran-
scriptome regulation in human a cells.
Phenotyping live islet cells from donors with T1D and other,
rarer forms of diabetes is a significant challenge that stems
from the paucity of available tissue and low islet cell recovery;
this has retarded understanding of islet and diabetes biology
(Brissova et al., 2018; Wang et al., 2016b). Likewise, procuring
and studying appropriate control tissue for studies of T1D islets,
matched in key properties like donor age, is an enduring chal-
lenge. Electrophysiological studies of islets from only one donor
with T1D have been reported, suggesting that surviving b cells
have normal function and a cells are dysfunctional (Walker
et al., 2011). Similarly, to date only islets from one donor with
T1D have been studied using scRNA-seq (Wang et al., 2016b).
Thus, our extension of patch-seq to studies of cryo-stored islets
from ND and T1D donors represents a significant innovation.
Here we show that library quality from cryo-preserved islets is
1028 Cell Metabolism 31, 1017–1031, May 5, 2020
Resource
comparable to that of control fresh islet cells. Moreover, our
detection of a cell dysfunction and normal function of surviving
b cells using patch-seq with T1D islets correlates well with find-
ings from a recent study using different methods (Brissova et al.,
2018). It is important to note that b cells from donors with T1D
studied here were from islets obtained >11 years (and up to 31
years) following diagnosis. Thus, further studies are needed to
investigate potential b cell dysfunction that may occur at earlier
stages of disease progression. Patch-seq revealed greater a
cell heterogeneity in T1D compared to ND islets, including evi-
dence of impaired maintenance of a cell fate, and we speculate
that this might contribute to the well-recognized impairment of
glucagon secretion observed in T1D (Unger and Cherrington,
2012). Moreover, transcriptomic profiling here reveals that func-
tional preservation in b cells from donors with T1D corresponds
to undetectable expression of RBP4, consistent with work by
others (Brissova et al., 2018; Rui et al., 2017). By contrast, we
show that ductal cells in T1D, which some suggest represent a
potential source of new b cells (Corritore et al., 2016), lack
detectable endocrine physiological phenotypes. Whether from
pancreatic cell reprogramming or other sources (Chakravarthy
et al., 2017; Thorel et al., 2010), rigorous evaluation of ‘‘replace-
ment’’ b cells with patch-seq should emerge as a new bench-
mark to assess functional and transcriptional resemblance to
bona fide b cells.
In conclusion, the approaches presented here enable a precise
molecular characterization of endocrine cell physiology and
represent significant advances in the development of multimodal
scRNA-seq technologies by increasing cell throughput and ex-
tending analyses of simultaneously measured transcriptomic
data and physiological features. Our results help clarify observed
inherent islet cellular heterogeneity and provide valuable tools for
understanding the molecular mechanisms underlying normal islet
function and dysfunction in diabetes. This work also sets a frame-
work that could be applied to other cell types where regulated
exocytosis and secretion are hallmark features of cellular identity.
Limitations of Study
Our patch-seq characterization is performed after islet isolation
and dispersion into single cells. Although we performed controls
to assess the effect of single-cell dispersion and patch-seq,
translation of our findings to in vivo conditions will require further
studies. Similarly, by single-cell dissociation, contextual infor-
mation from neighboring cells and tissue structure is lost.
Patch-seq directly in live pancreas slices could bridge our results
to islet structure, although this still poses significant technical
challenges in a ribonuclease-rich tissue such as the pancreas.
Throughout this work, we used voltage-clamp recordings of
ion channels and measurements of exocytosis as a proxy for islet
cell function. Although these are established markers of islet cell
excitability and function, further experiments recording action
potential firing, Ca2+ signaling, and hormone secretion assays
would strengthen gene candidates identified here and their
mechanistic role in b cell physiology. The PS gene model is
derived using patch-seq cells from nondiabetic donors. Its
refinement through larger datasets that include more diabetic
donors should improve the generality of the model and applica-
bility for b cell benchmarking in biomedical settings. For T1D
studies we used cryopreserved islets. An advantage to this is

Resource
that it allowed us to perform all experiments using matched
controls to minimize confounding effects; however, it would be
of interest to confirm these results in independent studies using
fresh tissue.
STAR+METHODS
Detailed methods are provided in the online version of this paper
and include the following:
d KEY RESOURCES TABLE
d RESOURCE AVAILABILITY
B Lead Contact and Materials Availability
B Data and Code Availability
d EXPERIMENTAL MODEL AND SUBJECT DETAILS
B Islet Isolation, Cryopreservation, and Insulin Secretion
d METHOD DETAILS
B Electrophysiological Phenotyping
B siRNA Transfection
B Islet Dispersion and FACS
B Single-cell RNA-Seq
d QUANTIFICATION AND STATISTICAL ANALYSIS
B Processing and Quality Control of Single-cell RNA-
seq Data
B Clustering and Cell Type Determination
B Correlation between Electrophysiology and Gene
Expression
B Electrophysiological Predictions Using PS Gene Set
B Cell Type Identification from Electrophysiological Data
B Statistical Analysis
SUPPLEMENTAL INFORMATION
Supplemental Information can be found online at https://doi.org/10.1016/j.
cmet.2020.04.005.
ACKNOWLEDGMENTS
This work was supported by funding from the Chan Zuckerberg Biohub and
the California Institute for Regenerative Medicine to S.R.Q., a Foundation
Grant from the Canadian Institutes of Health Research (CIHR: 148451) to
P.E.M., grants from the U.S. National Institutes of Health (1U01DK10830001,
1R01DK107507, 1R01DK108817, and P30 DK116074 to S.K.K. and
1U01DK120447-01 to S.K.K., S.R.Q., and P.E.M.), and JDRF (2-SRA-2019-
698-S-B) to S.K.K. and P.E.M. Support from the Human Islet Research Core
in the Stanford Diabetes Research Center is gratefully acknowledged. Human
islet isolation at the Alberta Diabetes Institute IsletCore was initially subsidized
by the Alberta Diabetes Foundation. We thank Dr. S.H. Kim (Kim lab, Stanford)
and Dylan Hendersson (Chan Zuckerberg Biohub) for help in sample process-
ing, and Dr. Norma Neff (Chan Zuckerberg Biohub) and Jennifer Okamoto
(Chan Zuckerberg Biohub) for sequencing expertise. We thank Dr. Linford
Briant (University of Oxford) for helpful discussions on islet cell-type modeling,
and Dr. Jocelyn Manning Fox (University of Alberta) for critical reading of the
manuscript. We thank the organ procurement organizations across Canada,
particularly the Human Organ Procurement and Exchange (HOPE) program
in Edmonton and the Trillium Gift of Life Network (TGLN) in Ontario, for their
work in obtaining human pancreas for research. We thank Drs. Rita Bottino (Al-
legheny Health Network) and Alvin C. Powers (Vanderbilt University) for kindly
providing islets used in validation studies. We also thank Dr. Manning Fox and
Mrs. Nancy Smith (University of Alberta) for their contributions to human islet
isolations. Finally, we are indebted to organ donors and their families for their
generous support of scientific research.
ll
AUTHOR CONTRIBUTIONS
J.C.-S. and X.-Q.D. developed the pancreas patch-seq pipeline; X.-Q.D. per-
formed and analyzed patch-clamp data; J.C.-S. performed scRNA-seq and
analyzed patch-seq data; X.-Q.D. and K.S. performed knockdown experi-
ments; J.C.-S. and Y.H. performed FACS; J.L. and A.B. performed islet isola-
tions; A.B. performed insulin measurements; J.L. performed cryopreservation;
J.C.-S., X.-Q.D., Y.H., S.K.K., S.R.Q., and P.E.M. discussed data and inter-
preted results; S.K.K., S.R.Q., and P.E.M. supervised the project; and all au-
thors contributed to writing the manuscript.
DECLARATION OF INTERESTS
The authors disclose no conflicts of interest.
Received: November 12, 2019
Revised: January 23, 2020
Accepted: April 2, 2020
Published: April 16, 2020
REFERENCES
Aichler, M., Borgmann, D., Krumsiek, J., Buck, A., MacDonald, P.E., Manning
Fox, J.E., Lyon, J., Light, P.E., Keipert, S., Jastroch, M., et al. (2017). N-acyl
taurines and acylcarnitines cause an imbalance in insulin synthesis and secre-
tion provoking b cell dysfunction in type 2 diabetes. Cell Metab. 25, 1334–
1347.e4.
Ait-Lounis, A., Bonal, C., Seguı́n-Estévez, Q., Schmid, C.D., Bucher, P.,
Herrera, P.L., Durand, B., Meda, P., and Reith, W. (2010). The transcription fac-
tor Rfx3 regulates beta-cell differentiation, function, and glucokinase expres-
sion. Diabetes 59, 1674–1685.
Akiyama, M., Liew, C.W., Lu, S., Hu, J., Martinez, R., Hambro, B., Kennedy,
R.T., and Kulkarni, R.N. (2013). X-box binding protein 1 is essential for insulin
regulation of pancreatic a-cell function. Diabetes 62, 2439–2449.
Alejandro, E.U., Gregg, B., Blandino-Rosano, M., Cras-Méneur, C., and
Bernal-Mizrachi, E. (2015). Natural history of b-cell adaptation and failure in
type 2 diabetes. Mol. Aspects Med. 42, 19–41.
Alves, T.C., Pongratz, R.L., Zhao, X., Yarborough, O., Sereda, S., Shirihai, O.,
Cline, G.W., Mason, G., and Kibbey, R.G. (2015). Integrated, step-wise, mass-
isotopomeric flux analysis of the TCA cycle. Cell Metab. 22, 936–947.
Anders, S., Pyl, P.T., and Huber, W. (2015). HTSeq–a Python framework to
work with high-throughput sequencing data. Bioinformatics 31, 166–169.
Arda, H.E., Li, L., Tsai, J., Torre, E.A., Rosli, Y., Peiris, H., Spitale, R.C., Dai, C.,
Gu, X., Qu, K., et al. (2016). Age-dependent pancreatic gene regulation reveals
mechanisms governing human b cell function. Cell Metab. 23, 909–920.
Ashcroft, F.M., and Rorsman, P. (2012). Diabetes mellitus and the b cell: the
last ten years. Cell 148, 1160–1171.
Bader, E., Migliorini, A., Gegg, M., Moruzzi, N., Gerdes, J., Roscioni, S.S.,
Bakhti, M., Brandl, E., Irmler, M., Beckers, J., et al. (2016). Identification of pro-
liferative and mature b-cells in the islets of Langerhans. Nature 535, 430–434.
Bakhti, M., Böttcher, A., and Lickert, H. (2019). Modelling the endocrine
pancreas in health and disease. Nat. Rev. Endocrinol. 15, 155–171.
Baron, M., Veres, A., Wolock, S.L., Faust, A.L., Gaujoux, R., Vetere, A., Ryu,
J.H., Wagner, B.K., Shen-Orr, S.S., Klein, A.M., et al. (2016). A single-cell tran-
scriptomic map of the human and mouse pancreas reveals inter- and intra-cell
population structure. Cell Syst. 3, 346–360.e4.
Breiman, L. (2001). Random Forests. Mach. Learn. 45, 5–32.
Briant, L.J.B., Zhang, Q., Vergari, E., Kellard, J.A., Rodriguez, B., Ashcroft,
F.M., and Rorsman, P. (2017). Functional identification of islet cell types by
electrophysiological fingerprinting. J. R. Soc. Interface 14, 20160999.
Brissova, M., Haliyur, R., Saunders, D., Shrestha, S., Dai, C., Blodgett, D.M.,
Bottino, R., Campbell-Thompson, M., Aramandla, R., Poffenberger, G., et al.
(2018). a cell function and gene expression are compromised in type 1 dia-
betes. Cell Rep. 22, 2667–2676.
Broch, M., Vendrell, J., Ricart, W., Richart, C., and Fernández-Real, J.-M.
(2007). Circulating retinol-binding protein-4, insulin sensitivity, insulin
Cell Metabolism 31, 1017–1031, May 5, 2020 1029
 ll
secretion, and insulin disposition index in obese and nonobese subjects.
Diabetes Care 30, 1802–1806.
Bunik, V., Kaehne, T., Degtyarev, D., Shcherbakova, T., and Reiser, G. (2008).
Novel isoenzyme of 2-oxoglutarate dehydrogenase is identified in brain, but
not in heart. FEBS J. 275, 4990–5006.
Burcelin, R., Knauf, C., and Cani, P.D. (2008). Pancreatic a-cell dysfunction in
diabetes. Diabetes Metab. 34 (Suppl 2 ), S49–S55.
Burgoyne, R.D., and Morgan, A. (2003). Secretory granule exocytosis. Physiol.
Rev. 83, 581–632.
Byrnes, L.E., Wong, D.M., Subramaniam, M., Meyer, N.P., Gilchrist, C.L.,
Knox, S.M., Tward, A.D., Ye, C.J., and Sneddon, J.B. (2018). Lineage dy-
namics of murine pancreatic development at single-cell resolution. Nat.
Commun. 9, 3922.
Cadwell, C.R., Palasantza, A., Jiang, X., Berens, P., Deng, Q., Yilmaz, M.,
Reimer, J., Shen, S., Bethge, M., Tolias, K.F., et al. (2016).
Electrophysiological, transcriptomic and morphologic profiling of single neu-
rons using Patch-seq. Nat. Biotechnol. 34, 199–203.
Chakravarthy, H., Gu, X., Enge, M., Dai, X., Wang, Y., Damond, N., Downie, C.,
Liu, K., Wang, J., Xing, Y., et al. (2017). Converting adult pancreatic islet a cells
into b cells by targeting both Dnmt1 and Arx. Cell Metab. 25, 622–634.
Chimienti, F., Devergnas, S., Favier, A., and Seve, M. (2004). Identification and
cloning of a beta-cell-specific zinc transporter, ZnT-8, localized into insulin
secretory granules. Diabetes 53, 2330–2337.
Corritore, E., Lee, Y.-S., Sokal, E.M., and Lysy, P.A. (2016). b-cell replacement
sources for type 1 diabetes: a focus on pancreatic ductal cells. Ther. Adv.
Endocrinol. Metab. 7, 182–199.
Dahl, U., Sjødin, A., and Semb, H. (1996). Cadherins regulate aggregation of
pancreatic beta-cells in vivo. Development 122, 2895–2902.
Dallavalle, C., Albino, D., Civenni, G., Merulla, J., Ostano, P., Mello-Grand, M.,
Rossi, S., Losa, M., D’Ambrosio, G., Sessa, F., et al. (2016). MicroRNA-424 im-
pairs ubiquitination to activate STAT3 and promote prostate tumor progres-
sion. J. Clin. Invest. 126, 4585–4602.
Danielsson, A., Pontén, F., Fagerberg, L., Hallström, B.M., Schwenk, J.M.,
Uhlén, M., Korsgren, O., and Lindskog, C. (2014). The human pancreas prote-
ome defined by transcriptomics and antibody-based profiling. PLoS ONE 9,
e115421.
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, 15–21.
Dorrell, C., Schug, J., Canaday, P.S., Russ, H.A., Tarlow, B.D., Grompe, M.T.,
Horton, T., Hebrok, M., Streeter, P.R., Kaestner, K.H., and Grompe, M. (2016).
Human islets contain four distinct subtypes of b cells. Nat. Commun. 7, 11756.
Enge, M., Arda, H.E., Mignardi, M., Beausang, J., Bottino, R., Kim, S.K., and
Quake, S.R. (2017). Single-cell analysis of human pancreas reveals transcrip-
tional signatures of aging and somatic mutation patterns. Cell 171, 321–
330.e14.
Ferdaoussi, M., Dai, X., Jensen, M.V., Wang, R., Peterson, B.S., Huang, C.,
Ilkayeva, O., Smith, N., Miller, N., Hajmrle, C., et al. (2015). Isocitrate-to-
SENP1 signaling amplifies insulin secretion and rescues dysfunctional b cells.
J. Clin. Invest. 125, 3847–3860.
Fernández-Tajes, J., Gaulton, K.J., van de Bunt, M., Torres, J., Thurner, M.,
Mahajan, A., Gloyn, A.L., Lage, K., and McCarthy, M.I. (2019). Developing a
network view of type 2 diabetes risk pathways through integration of genetic,
genomic and functional data. Genome Med. 11, 19.
€dhof, T.C.
Földy, C., Darmanis, S., Aoto, J., Malenka, R.C., Quake, S.R., and Su
(2016). Single-cell RNAseq reveals cell adhesion molecule profiles in electro-
physiologically defined neurons. Proc. Natl. Acad. Sci. USA 113,
E5222–E5231.
Fuzik, J., Zeisel, A., Máté, Z., Calvigioni, D., Yanagawa, Y., Szabó, G.,
Linnarsson, S., and Harkany, T. (2016). Integration of electrophysiological re-
cordings with single-cell RNA-seq data identifies neuronal subtypes. Nat.
Biotechnol. 34, 175–183.
1030 Cell Metabolism 31, 1017–1031, May 5, 2020
Resource
Gan, W.J., Do, O.H., Cottle, L., Ma, W., Kosobrodova, E., Cooper-White, J.,
Bilek, M., and Thorn, P. (2018). Local integrin activation in pancreatic b cells
targets insulin secretion to the vasculature. Cell Rep. 24, 2819–2826.e3.
Gembal, M., Gilon, P., and Henquin, J.C. (1992). Evidence that glucose can
control insulin release independently from its action on ATP-sensitive K+ chan-
nels in mouse B cells. J. Clin. Invest. 89, 1288–1295.
Geron, E., Boura-Halfon, S., Schejter, E.D., and Shilo, B.-Z. (2015). The edges
of pancreatic islet b cells constitute adhesive and signaling microdomains. Cell
Rep. 10, 317–325.
Johnston, N.R., Mitchell, R.K., Haythorne, E., Pessoa, M.P., Semplici, F.,
Ferrer, J., Piemonti, L., Marchetti, P., Bugliani, M., Bosco, D., et al. (2016).
Beta cell hubs dictate pancreatic islet responses to glucose. Cell Metab. 24,
389–401.
Kahn, S.E., Hull, R.L., and Utzschneider, K.M. (2006). Mechanisms linking
obesity to insulin resistance and type 2 diabetes. Nature 444, 840–846.
Kaido, T., Perez, B., Yebra, M., Hill, J., Cirulli, V., Hayek, A., and Montgomery,
A.M. (2004). Alphav-integrin utilization in human beta-cell adhesion,
spreading, and motility. J. Biol. Chem. 279, 17731–17737.
Keenan, H.A., Sun, J.K., Levine, J., Doria, A., Aiello, L.P., Eisenbarth, G.,
Bonner-Weir, S., and King, G.L. (2010). Residual insulin production and
pancreatic ß-cell turnover after 50 years of diabetes: Joslin Medalist Study.
Diabetes 59, 2846–2853.
Korsunsky, I., Millard, N., Fan, J., Slowikowski, K., Zhang, F., Wei, K.,
Baglaenko, Y., Brenner, M., Loh, P.R., and Raychaudhuri, S. (2019). Fast, sen-
sitive and accurate integration of single-cell data with Harmony. Nat. Methods
16, 1289–1296.
La Manno, G., Gyllborg, D., Codeluppi, S., Nishimura, K., Salto, C., Zeisel, A.,
Borm, L.E., Stott, S.R.W., Toledo, E.M., Villaescusa, J.C., et al. (2016).
Molecular diversity of midbrain development in mouse, human, and stem cells.
Cell 167, 566–580.e19.
Liu, R., Mignardi, M., Jones, R., Enge, M., Kim, S.K., Quake, S.R., and Zou, J.
(2019). Modeling spatial correlation of transcripts with application to devel-
oping pancreas. Sci. Rep. 9, 5592.
Lu, D., Mulder, H., Zhao, P., Burgess, S.C., Jensen, M.V., Kamzolova, S.,
Newgard, C.B., and Sherry, A.D. (2002). 13C NMR isotopomer analysis reveals
a connection between pyruvate cycling and glucose-stimulated insulin secre-
tion (GSIS). Proc. Natl. Acad. Sci. USA 99, 2708–2713.
Lyon, J., Spigelman, A., MacDonald, P., and Manning, J. (2019). ADI IsletCore
protocols for the isolation, assessment and cryopreservation of human
pancreatic islets of Langerhans for research purposes. Published online
February 12, 2019. https://doi.org/10.17504/protocols.io.x3mfqk6.
Macaulay, I.C., Ponting, C.P., and Voet, T. (2017). Single-cell multiomics: mul-
tiple measurements from single cells. Trends Genet. 33, 155–168.
Mahajan, A., Taliun, D., Thurner, M., Robertson, N.R., Torres, J.M., Rayner,
N.W., Payne, A.J., Steinthorsdottir, V., Scott, R.A., Grarup, N., et al. (2018).
Fine-mapping type 2 diabetes loci to single-variant resolution using high-den-
sity imputation and islet-specific epigenome maps. Nat. Genet. 50,
1505–1513.
Manning Fox, J.E., Lyon, J., Dai, X.Q., Wright, R.C., Hayward, J., van de Bunt,
M., Kin, T., Shapiro, A.M.J., McCarthy, M.I., Gloyn, A.L., et al. (2015). Human
islet function following 20 years of cryogenic biobanking. Diabetologia 58,
1503–1512.
Morgan, N.G., and Richardson, S.J. (2018). Fifty years of pancreatic islet pa-
thology in human type 1 diabetes: insights gained and progress made.
Diabetologia 61, 2499–2506.
Muraro, M.J., Dharmadhikari, G., Gru€n, D., Groen, N., Dielen, T., Jansen, E.,
van Gurp, L., Engelse, M.A., Carlotti, F., de Koning, E.J.P., and van
Oudenaarden, A. (2016). A single-cell transcriptome atlas of the human
pancreas. Cell Syst. 3, 385–394.e3.
Newgard, C.B. (2017). Metabolomics and metabolic diseases: where do we
stand? Cell Metab. 25, 43–56.
O’Shea, J.J., and Plenge, R. (2012). JAK and STAT signaling molecules in
immunoregulation and immune-mediated disease. Immunity 36, 542–550.

Resource
Otter, S., and Lammert, E. (2016). Exciting times for pancreatic islets: gluta-
mate signaling in endocrine cells. Trends Endocrinol. Metab. 27, 177–188.
Parnaud, G., Lavallard, V., Bedat, B., Matthey-Doret, D., Morel, P., Berney, T.,
and Bosco, D. (2015). Cadherin engagement improves insulin secretion of sin-
gle human b-cells. Diabetes 64, 887–896.
Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O.,
Blondel, M., Prettenhofer, P., Weiss, R., Dubourg, V., et al. (2011). Scikit-learn:
machine learning in Python. J. Mach. Learn. Res. 12, 2825–2830.
Perelis, M., Marcheva, B., Ramsey, K.M., Schipma, M.J., Hutchison, A.L.,
Taguchi, A., Peek, C.B., Hong, H., Huang, W., Omura, C., et al. (2015).
Pancreatic b cell enhancers regulate rhythmic transcription of genes control-
ling insulin secretion. Science 350, aac4250.
Picelli, S., Faridani, O.R., Björklund, Å.K., Winberg, G., Sagasser, S., and
Sandberg, R. (2014). Full-length RNA-seq from single cells using Smart-
seq2. Nat. Protoc. 9, 171–181.
Pipeleers, D.G. (1992). Heterogeneity in pancreatic beta-cell population.
Diabetes 41, 777–781.
Prasad, R.B., and Groop, L. (2016). Single-cell sequencing of human pancre-
atic islets-new kids on the block. Cell Metab. 24, 523–524.
Rorsman, P., and Braun, M. (2013). Regulation of insulin secretion in human
pancreatic islets. Annu. Rev. Physiol. 75, 155–179.
Rosen, E.D., Kaestner, K.H., Natarajan, R., Patti, M.-E., Sallari, R., Sander, M.,
and Susztak, K. (2018). Epigenetics and epigenomics: implications for dia-
betes and obesity. Diabetes 67, 1923–1931.
Rui, J., Deng, S., Arazi, A., Perdigoto, A.L., Liu, Z., and Herold, K.C. (2017). b
cells that resist immunological attack develop during progression of autoim-
mune diabetes in NOD mice. Cell Metab. 25, 727–738.
Saarima €ki-Vire, J., Balboa, D., Russell, M.A., Saarikettu, J., Kinnunen, M.,
Keskitalo, S., Malhi, A., Valensisi, C., Andrus, C., Eurola, S., et al. (2017). An
activating STAT3 mutation causes neonatal diabetes through premature in-
duction of pancreatic differentiation. Cell Rep. 19, 281–294.
Salomon, D., and Meda, P. (1986). Heterogeneity and contact-dependent
regulation of hormone secretion by individual B cells. Exp. Cell Res. 162,
507–520.
Segerstolpe, Å., Palasantza, A., Eliasson, P., Andersson, E.-M., Andréasson,
A.-C., Sun, X., Picelli, S., Sabirsh, A., Clausen, M., Bjursell, M.K., et al.
(2016). Single-cell transcriptome profiling of human pancreatic islets in health
and type 2 diabetes. Cell Metab. 24, 593–607.
Sharma, U., and Rando, O.J. (2017). Metabolic inputs into the epigenome. Cell
Metab. 25, 544–558.
Stefan, Y., Meda, P., Neufeld, M., and Orci, L. (1987). Stimulation of insulin
secretion reveals heterogeneity of pancreatic B cells in vivo. J. Clin. Invest.
80, 175–183.
Stuart, T., and Satija, R. (2019). Integrative single-cell analysis. Nat. Rev.
Genet. 20, 257–272.
Suckow, A.T., Comoletti, D., Waldrop, M.A., Mosedale, M., Egodage, S.,
Taylor, P., and Chessler, S.D. (2008). Expression of neurexin, neuroligin, and
their cytoplasmic binding partners in the pancreatic b-cells and the involve-
ment of neuroligin in insulin secretion. Endocrinology 149, 6006–6017.
Suriben, R., Kaihara, K.A., Paolino, M., Reichelt, M., Kummerfeld, S.K.,
Modrusan, Z., Dugger, D.L., Newton, K., Sagolla, M., Webster, J.D., et al.
(2015). b-cell insulin secretion requires the ubiquitin ligase COP1. Cell 163,
1457–1467.
ll
Tabula Muris Consortium; Overall coordination; Logistical coordination; Organ
collection and processing; Library preparation and sequencing;
Computational data analysis; Cell type annotation; Writing group;
Supplemental text writing group; Principal investigators (2018). Single-cell
transcriptomics of 20 mouse organs creates a Tabula Muris. Nature 562,
367–372.
Thorel, F., Népote, V., Avril, I., Kohno, K., Desgraz, R., Chera, S., and Herrera,
P.L. (2010). Conversion of adult pancreatic a-cells to b-cells after extreme
b-cell loss. Nature 464, 1149–1154.
Tritschler, S., Theis, F.J., Lickert, H., and Böttcher, A. (2017). Systematic sin-
gle-cell analysis provides new insights into heterogeneity and plasticity of the
pancreas. Mol. Metab. 6, 974–990.
Uhlén, M., Fagerberg, L., Hallström, B.M., Lindskog, C., Oksvold, P.,
Mardinoglu, A., Sivertsson, Å., Kampf, C., Sjöstedt, E., Asplund, A., et al.
(2015). Proteomics. Tissue-based map of the human proteome. Science
347, 1260419.
Unger, R.H., and Cherrington, A.D. (2012). Glucagonocentric restructuring of
diabetes: a pathophysiologic and therapeutic makeover. J. Clin. Invest.
122, 4–12.
van der Meulen, T., Mawla, A.M., DiGruccio, M.R., Adams, M.W., Nies, V.,
Dólleman, S., Liu, S., Ackermann, A.M., Cáceres, E., Hunter, A.E., et al.
(2017). Virgin beta cells persist throughout life at a neogenic niche within
pancreatic islets. Cell Metab. 25, 911–926.e6.
Villani, A.-C., Satija, R., Reynolds, G., Sarkizova, S., Shekhar, K., Fletcher, J.,
Griesbeck, M., Butler, A., Zheng, S., Lazo, S., et al. (2017). Single-cell RNA-seq
reveals new types of human blood dendritic cells, monocytes, and progeni-
tors. Science 356, 1–10.
Walker, J.N., Johnson, P.R.V., Shigeto, M., Hughes, S.J., Clark, A., and
Rorsman, P. (2011). Glucose-responsive beta cells in islets isolated from a pa-
tient with long-standing type 1 diabetes mellitus. Diabetologia 54, 200–202.
Wang, Y.J., and Kaestner, K.H. (2019). Single-cell RNA-seq of the pancreatic
islets–a promise not yet fulfilled? Cell Metab. 29, 539–544.
Wang, Y.J., Golson, M.L., Schug, J., Traum, D., Liu, C., Vivek, K., Dorrell, C.,
Naji, A., Powers, A.C., Chang, K.-M., et al. (2016a). Single-cell mass cytometry
analysis of the human endocrine pancreas. Cell Metab. 24, 616–626.
Wang, Y.J., Schug, J., Won, K.-J., Liu, C., Naji, A., Avrahami, D., Golson, M.L.,
and Kaestner, K.H. (2016b). Single-cell transcriptomics of the human endo-
crine pancreas. Diabetes 65, 3028–3038.
Williams, R. (2016). Williams Textbook of Endocrinology (Elsevier).
Wolf, F.A., Angerer, P., and Theis, F.J. (2018). SCANPY: large-scale single-cell
gene expression data analysis. Genome Biol. 19, 15.
Zhang, Q., Chibalina, M.V., Bengtsson, M., Groschner, L.N., Ramracheya, R.,
Rorsman, N.J.G., Leiss, V., Nassar, M.A., Welling, A., Gribble, F.M., et al.
(2014). Na+ current properties in islet a- and b-cells reflect cell-specific
Scn3a and Scn9a expression. J. Physiol. 592, 4677–4696.
Zhang, M., Goforth, P., Bertram, R., Sherman, A., and Satin, L. (2003). The
Ca2+ dynamics of isolated mouse b-cells and islets: implications for mathe-
matical models. Biophys. J. 84, 2852–2870.
Zhang, C., Moriguchi, T., Kajihara, M., Esaki, R., Harada, A., Shimohata, H.,
Oishi, H., Hamada, M., Morito, N., Hasegawa, K., et al. (2005). MafA is a key
regulator of glucose-stimulated insulin secretion. Mol. Cell. Biol. 25,
4969–4976.
Cell Metabolism 31, 1017–1031, May 5, 2020 1031
 The science of data

We believe: Data are boundless. Big ideas deserve a big audience.

Insights fuel action.
Patterns, a new open access data science journal from Cell Press, promotes all
types of research outputs and facilitates sharing and collaboration to solve key
scientific problems and aid in the development of solutions for practice, policies,
and management.
The research we publish is both theoretical and practical. We’re committed to
innovations that make research actionable for humans and machines alike.

Share your data story. Submit your paper to cell.com/patterns

 ll

Article
Niche-Specific Reprogramming of Epigenetic
Landscapes Drives Myeloid Cell Diversity
in Nonalcoholic Steatohepatitis
Jason S. Seidman,1,10 Ty D. Troutman,1,2,10,* Mashito Sakai,1,10 Anita Gola,3 Nathanael J. Spann,1 Hunter Bennett,1
Cassi M. Bruni,1 Zhengyu Ouyang,1 Rick Z. Li,1 Xiaoli Sun,2 BaoChau T. Vu,1 Martina P. Pasillas,1 Kaori M. Ego,1
David Gosselin,4 Verena M. Link,1,5 Ling-Wa Chong,6 Ronald M. Evans,6,7 Bonne M. Thompson,8 Jeffrey G. McDonald,8
Mojgan Hosseini,9 Joseph L. Witztum,2 Ronald N. Germain,3 and Christopher K. Glass1,2,11,*
1Department of Cellular and Molecular Medicine, University of California, San Diego, La Jolla, CA, USA
2Department of Medicine, University of California, San Diego, La Jolla, CA, USA
3Lymphocyte Biology Section, Laboratory of Immune System Biology, National Institute of Allergy and Infectious Diseases, National Institutes

of Health, Bethesda, MD 201892, USA

4Department of Molecular Medicine, Université Laval, Quebec City, QC, Canada
5Faculty of Biology, Division of Evolutionary Biology, Ludwig-Maximilian University of Munich, Munich, Germany
6Gene Expression Laboratory, The Salk Institute for Biological Studies, La Jolla, CA, USA
7Howard Hughes Medical Institute, The Salk Institute for Biological Studies, La Jolla, CA, USA
8Center for Human Nutrition, UT Southwestern Medical Center, Dallas, TX, USA
9Department of Pathology, University of California, San Diego, La Jolla, CA, USA
10These authors contributed equally
11Lead Contact

*Correspondence: ttroutman@health.ucsd.edu (T.D.T.), ckg@ucsd.edu (C.K.G.)

https://doi.org/10.1016/j.immuni.2020.04.001

SUMMARY

Tissue-resident and recruited macrophages contribute to both host defense and pathology. Multiple macro-
phage phenotypes are represented in diseased tissues, but we lack deep understanding of mechanisms con-
trolling diversification. Here, we investigate origins and epigenetic trajectories of hepatic macrophages dur-
ing diet-induced non-alcoholic steatohepatitis (NASH). The NASH diet induced significant changes in Kupffer
cell enhancers and gene expression, resulting in partial loss of Kupffer cell identity, induction of Trem2 and
Cd9 expression, and cell death. Kupffer cell loss was compensated by gain of adjacent monocyte-derived
macrophages that exhibited convergent epigenomes, transcriptomes, and functions. NASH-induced
changes in Kupffer cell enhancers were driven by AP-1 and EGR that reprogrammed LXR functions required
for Kupffer cell identity and survival to instead drive a scar-associated macrophage phenotype. These find-
ings reveal mechanisms by which disease-associated environmental signals instruct resident and recruited
macrophages to acquire distinct gene expression programs and corresponding functions.

INTRODUCTION Resident macrophages and infiltrating monocyte-derived cells

Tissue-resident macrophages exhibit different transcriptomes in
response to environmental signals, enabling tissue-specific
functions (Gautier et al., 2012; Gosselin et al., 2014; Lavin
et al., 2014; Mass et al., 2016). Most tissues acquire resident
macrophage populations early in development from yolk sac or
fetal liver erythromyeloid progenitor (EMP) cells (Ginhoux et al.,
2010; Gomez Perdiguero et al., 2015; McGrath et al., 2015;
Schulz et al., 2012). In the brain and liver, EMP-derived microglia
and Kupffer cells (KCs) continue as exclusive, self-renewing
populations. In other tissues, EMP-derived macrophages are
partially or completely replaced over time by hematopoietic
stem cell (HSC)-derived cells (Bain et al., 2014; De Schepper
et al., 2018; Shaw et al., 2018; Yona et al., 2013).
contribute to a diverse array of human diseases (Holtman et al.,
2017; Kiss et al., 2018; Pirzgalska and Domingos, 2018). Nonal-
coholic steatohepatitis (NASH) is a form of nonalcoholic fatty
liver disease (NAFLD) and develops through combinatorial ac-
tions of steatosis and growing inflammation in response to
cellular stresses of the perturbed fatty liver environment (Haas
et al., 2016; Hardy et al., 2016; Machado and Diehl, 2016; Musso
et al., 2016; Rinella and Sanyal, 2016; Younossi et al., 2016). Dur-
ing NASH or acute liver injury, new macrophages are derived
from circulating monocytes (Karlmark et al., 2009; Seki et al.,
2009a, 2009b; Tacke and Zimmermann, 2014; Zigmond et al.,
2014). These recruited monocyte-derived cells can have both
detrimental and supportive roles, contributing to increases in pa-
thology during fibrosis onset, but hastening recovery when the

Immunity 52, 1057–1074, June 16, 2020 ª 2020 Elsevier Inc. 1057
 ll
damage-evoking agent is removed (Ju and Tacke, 2016; Karl-
mark et al., 2009; Mitchell et al., 2009; Seki et al., 2009a,
2009b; Zigmond et al., 2014). These observations raise ques-
tions as to the mechanisms underlying the phenotypic diversity
of disease-associated macrophages.
Here, we define the identities, developmental origins, and mi-
cro-anatomic locations of the major myeloid populations in
response to a NASH-inducing diet. By sorting these cell popula-
tions and performing deep transcriptomic and epigenomic anal-
ysis, we provide evidence for combinatorial effects of diet and
anatomic location on regulatory pathways and transcription fac-
tors (TFs) that explain the emergence of disease-associated
macrophage phenotypes.
RESULTS
Single-Cell RNA-Seq Defines Myeloid Diversity During
Dietary NASH
To investigate immune cell heterogeneity during NASH, we per-
formed single-cell RNA sequencing (scRNA-seq) on liver non-
parenchymal cells from healthy mice fed a control diet or
NASH diet. As noted in previous studies (Hall et al., 2010; Pous-
sin et al., 2011), C57BL/6J mice exhibited rapid weight gain
when fed the NASH diet (Figure S1A) and developed steatosis,
inflammation, and a modest degree of fibrosis (Figure S1B).
Non-parenchymal cells from control mice and mice fed the
NASH diet for 30 weeks were purified using fluorescence-acti-
vated cell sorting (FACS) (Figure S1C). Over 6,000 scRNA-seq li-
braries that passed quality thresholds were created from these
isolated cells.
The NASH diet induced qualitative transcriptional differences
in cell clustering between cells from control livers and NASH
livers (Figure 1A left, Figure 1B; Table S1). Further, three of 17
clusters were primarily derived from cells from control diet
mice while eight clusters were primarily derived from NASH
diet mice (Figures 1A and S1D; Table S1). Interrogation of the
most abundant and most differentially expressed (DE) tran-
scripts in each cluster allowed a cell identity to be readily as-
signed to most clusters. Further studies were focused on the
five most abundant macrophage clusters. (Figure S1E;
Table S1).
The three most abundant clusters (KC-H, healthy KC;
KC-N, NASH KC; and KN-RM, named for being a recruited
macrophage occupying the KC niche, as described later)
had high expression of Adgre1, encoding the tissue macro-
phage marker F4/80, and high expression of putative KC
lineage-determining transcription factors (LDTFs) Mafb
and Maf, and KC-specific genes, as well as low expression
of Itgam, encoding CD11b (Figures 1C and S1E; Lavin et al.,
2014). KC-H consisted almost entirely of cells from controls,
while KC-N and KN-RM consisted of cells from the NASH
diet animals (Figures 1A and 1B; Table S1). The other two
major macrophage clusters were predominantly found in
NASH livers and had characteristic gene expression of
previously described monocyte-derived liver macrophages
(Figures 1C and S1E; Krenkel et al., 2018). Cluster 4 (hereafter
referred to as Ly6Chi-RM) expressed higher amounts of
transcripts typical of Ly6Chi monocytes such as Ly6c2,
Chil3, F13a1, and Fn1, while cluster 9 (hereafter referred to
1058 Immunity 52, 1057–1074, June 16, 2020
Article
as Ly6Clo-RM) highly expressed Cd209a, Cd7, and Itgax
(Figure S1E).
Using flow cytometry, we observed both Tim4+ and Tim4
cells in the CD11bloF4/80hi KC gate during NASH in similar
proportions to KC-N and KN-RM and collected them for bulk
RNA-seq and epigenomic profiling (Figure 1D). We also
collected CD11bhiF4/80lo-Ly6Chi-RM and Ly6Clo-RM for RNA-
seq and assay for transposase-accessible chromatin (ATAC-
seq) (Figure 1E; Buenrostro et al., 2013). RNA-seq analysis of
these populations yielded concordant expression profiles to
scRNA-seq when comparing Seurat marker genes, indicating
that we isolated the corresponding macrophage populations
defined by scRNA-seq (Figure 1F).
We next compared these mouse macrophages to human
scRNA-seq populations of non-parenchymal liver cells from in-
dividuals with either advanced cirrhosis or without evident liver
disease (Ramachandran et al., 2019). Using orthogonal genes
to cluster the combined set of human and mouse cells under
control and disease conditions resulted in eight clusters (0–7,
Figures S1F and S1G). Human livers exhibited a greater diver-
sity of myeloid cells and, in contrast to the mouse model, all cell
populations were present to some extent in individuals with or
without cirrhosis. Importantly, mouse KC-H cells overlapped
with human KCs (clusters 0 and 1). A key finding of the human
studies was the increased presence of a ‘‘scar-associated
macrophage’’ (SAM) population in cirrhotic livers (Ramachan-
dran et al., 2019). These cells were found to reside in close
proximity to areas of fibrosis and were characterized by
increased expression of TREM2 and CD9. Importantly, the
cluster containing SAMs also contained substantial fractions
of the NASH-associated KC-N and KN-RM cells (cluster 3),
but not KC-H cells, consistent with the former cell type’s
increased expression of Trem2 and Cd9 (Figure 1F). Thus,
despite relatively sparse fibrosis, a SAM phenotype emerges
in this model.
Ontogeny and Environment Contribute to Diversity of
Myeloid Cells During NASH
We hypothesized that, similar to experimental KC ablation (Bon-
nardel et al., 2019; Sakai et al., 2019; Scott et al., 2016), Tim4
KC-like macrophages that arise during NASH (KN-RM) may be
ontogenically distinct from the embryonically derived Tim4+
KCs (KC-N). To test this hypothesis, we performed lineage-
tracing experiments using Cx3cr1CreERT2; Rosa26tdTomato/+
mice (Figure S2A; Theurl et al., 2016; Yona et al., 2013). This
model labels tissue macrophages if tamoxifen is administered
perinatally, while only certain macrophages such as microglia,
but not KCs, are labeled if mice are pulsed during adulthood
(Dick et al., 2019; Theurl et al., 2016; Yona et al., 2013). In
contrast, during adulthood, tamoxifen labels peripheral mono-
cytes and thus monocyte-derived macrophages.
Tamoxifen administration on days 1 and 2 post-parturition re-
sulted in ~50% labeling of embryonic KCs (Figures S2B–S2D).
After 20 weeks on the NASH diet, nearly all tdTomato+ cells
were Tim4+ KCs (KC-N) (Figures S2B–S2D), indicating that
Tim4 KC-like macrophages (KN-RM) are not long-lived daugh-
ters of embryonic KCs. Conversely, when adult mice fed the
NASH diet for 20 weeks were given tamoxifen 4 or 1 week prior
to sacrifice, ~90% of circulating monocytes were TdTomato+
 ll
Article

A F

D E

Figure 1. Transcriptional Diversity of Hepatic Macrophages during NASH

(A) tSNE projections of identified graph-based cell clusters from scRNA-seq data derived from hepatic CD45+CD146 cells from control mice or mice fed a NASH
diet for 30 weeks. (Left) cells are colored based on dietary condition. (Right) cells are colored based on cell identity of the five major myeloid cells identified. Data
represent two independent donor mice per group.
(B) Macrophage proportions from (A) for control mice (left) or mice with NASH (right).
(C) Gene expression of normalized scRNA-seq data of genes supporting cluster identities. Boxes denote quartiles, and whiskers denote data range excluding
outliers.
(D) Terminal FACS gates for purification of KCs from control mice (left), or CD11bloF4/80hiTim4+ and Tim4 cells from mice with NASH (right).
(E) Terminal FACS gates for purification of Ly6Chi and Ly6Clo CD11bhiF4/80loLy6G CX3CR1+ recruited hepatic macrophages (RM). ‘‘PMN’’ stands for poly-
morphonuclear leukocyte.
(F) Comparison of myeloid clusters defined by scRNA-seq (left) and bulk RNA-seq (right) for the corresponding sorted populations. Left heatmap (100 cells per
column) depicts normalized and scaled expression values for marker genes identified using a Wilcoxon rank sum test through Seurat. Right heatmap shows z-
normalized row expression of each gene for RNA-seq from bulk purified cell populations from independent biological duplicates.
Please see also Figure S1.

2 days after the final tamoxifen injection (data not shown), and could be due to low expression of Cx3cr1 in Tim4+ KC-N cells
more than 10% of Tim4 KC-like macrophages (KN-RM) were or to the gradual upregulation of Tim4 by recruited cells (Scott
TdTomato+ (Figures S2C and S2D). We also observed a small et al., 2016). Overall, these results indicate that embryonic KCs
degree of labeling of Tim4+ KCs (Figures S2C and S2D), which remain in the liver during development of NASH and are Tim4+,

Immunity 52, 1057–1074, June 16, 2020 1059

 ll
Article

A B
NASH diet
E-cadherin tdTomato+ F4/80+

Ly6Chi&lo-RM KN-RM Ly6Chi&lo-RM Mgl2+

17% KN-RM (Ly6Clo-RM)
KC-N
PV

tdTomato

Mgl2
MFI
6.94%

cv
75.3%
PV Mgl2–
KC-N
(Ly6Chi-RM)
PV

F4/80 Tim4 F4/80

C D F
E-cadherin Ly6Chi&lo-RM KN-RM KC-N 1000 ***
*** E-cadherin Mgl2+ Mgl2–

neighbor distance (μm)

750

KC−N nearest
PV

500
CV PV
***
250

CV 0

M M M
−R hi −R lo −R
cv KN y6C y6C
L L
E G
PV PV ***
Distance to nearest

Distance to nearest
125 *** *** ***
large vessel (μm)

800

CV or PV (μm)
**
100 *** ***
600
75
400
50
PV
200
25
PV
0 0
CV PV CV PV
−N RM M
KC KN− hi&lo −R Ly6Chi−RM Ly6Clo−RM
6 C
H Ly
F4/80 TUNEL Tim4 F4/80 TUNEL Tim4
KC-N

KN-RM
NASH diet

KN-RM

KN-RM

KC-N

I J K
TUNEL nearest neighbor
Total TUNEL+ cell count

***
100 *** 100
per area 1,000 μm2

Percent KN-RM

***
1000
distance (μm)

75
1 100
50

10 25
0.01

1 0
KC−N KN−RM Ly6Chi&lo−RM 0 10 20 30
−H RM RM −N RM RM Weeks on AMLN NASH diet
KC KN− hi&lo − KC KN− hi&lo −
6C 6C
Ly Ly
Control NASH

(legend on next page)

1060 Immunity 52, 1057–1074, June 16, 2020
 ll
Article

while most recruited CD11bloF4/80hi macrophages are Tim4 .
These results also indicate that recruited KN-RM cells can
survive at least 4 weeks in the NASH liver while Ly6Chi&lo-RMs
are shorter lived.
To determine the localization of myeloid cells in the liver during
NASH, immunofluorescence (IF) microscopy was performed on
livers from Cx3cr1CreERT2; Rosa26tdTomato/+ mice pulsed with
tamoxifen 7 days prior to tissue collection (Figures 2A, 2B,
S2A, S2E, and S3A). As expected, tdTomato+ monocyte-derived
cells were abundant in NASH livers but nearly absent in control
livers (Figure S2E). Using multi-parameter imaging and histo-cy-
tometry (Gerner et al., 2012), we distinguished KC-H or KC-N,
KN-RM, and Ly6Chi&lo-RM in the control and NASH livers based
on Tim4, F4/80, and tdTomato expression. As Ly6C histological
staining could not be performed, Mgl2 was used as a surrogate
marker for Ly6Chi&lo-RM, with the Ly6Chi fraction being Mgl2
and Ly6Clo being represented by Mgl2+ (Figures 2A–2C, see Fig-
ure 3G for Ly6Clo-specific expression of Mgl2). Following in situ
confirmation of these myeloid cell subsets, we assessed their
spatial distribution using the positional data preserved in histo-
cytometry and nearest-neighbor distance analyses. The two
subsets (KN-RM and Ly6Chi&lo-RM) that were increased during
NASH displayed significant differences in their spatial distribu-
tions. By comparing nearest-neighbor distances from KC-N,
KN-RM cells were found to be distributed significantly closer
to KC-N cells than Ly6Chi&lo-RM cells (Figure 2D). In addition,
the small number KN-RM cells observed in control livers were
also in close proximity to KC-H cells (Figure S3B and data
not shown).
KCs reside within the hepatic sinusoids, but the sub-anatom-
ical organization of other liver macrophages during NASH is
less certain (Ju and Tacke, 2016). High-magnification imaging
using Collagen IV to visualize the endothelial basement mem-
brane demonstrated that KC-N and KN-RM cells resided within
hepatic sinusoids in this model (Figure S3C). In contrast, Ly6-
Chi&lo-RM cells were predominantly not found within the liver
sinusoids, but were highly enriched around both portal and cen-
tral vein vasculature identifiable by the large vessel diameter
(15 mm or larger) (Figures 2D and 2E). Further, Ly6Chi-RM
(Mgl2 ) cells were bi-modally distributed around portal and
central veins, while Ly6Clo-RM (Mgl2+) cells were more uniformly
distributed in closer proximity to central veins (Figures 2F and
2G). Together, these findings further supported Ly6Chi&lo-RM
cells as being positionally separated from KC-N and KN-RM
cells based on measurements that were independent of
anatomical landmarks. These results provide evidence that
KC-N and KN-RM cells reside in a similar niche within the liver
sinusoids, which is spatially distinct from the anatomic positions
occupied by Ly6Chi&lo-RMs.
Previous work found that DLL4 expression by liver sinusoidal
endothelial cells (LSECs) is graded toward higher expression
patterns in periportal zones (Halpern et al., 2018). Further,
DLL4 mediated-activation of RBPJ participates in driving differ-
entiation of monocytes to Kupffer-like cells in a repopulation
model (Sakai et al., 2019). During NASH, we confirmed the
peri-portal (E-cadherin+) enrichment of DLL4 on LSECs (Figures
S3D and S3E). Further, KC-N and KN-RM cells were significantly
closer to DLL4+ CD138+ LSECs compared to Ly6Chi&lo-RMs
(Figures S3F and S3G). These results provide support for DLL4
as a factor that may support niche specialization of distinct
macrophage populations during NASH, consistent with its
known function in instructing a KC program in recruited
monocytes.
NASH Diet Induces KC Death and Replacement
KCs are considered a self-renewing population that is normally
closed to HSC-derived cells (Perdiguero and Geissmann,
2016). However, the observation that the NASH diet resulted in
recruitment of HSC-derived KN-RM cells raised the question of
whether the NASH diet induces loss of embryonic KCs, providing
an open niche. TUNEL staining of NASH livers showed apoptosis
of Tim4+ KC-Ns during NASH diet, but not of KC-H cells from

Figure 2. Expanded Macrophage Diversity during NASH Is Supported by Monocyte Recruitment and Occupancy of Distinct Anatomical
Niches
(A) Confocal image of NASH liver showing E-cadherin (to demark the peri-portal regions) and tdTomato F4/80+ summed channel. Highlighted in the center by
white circles are the central veins (CV). Demarked at the periphery of the liver lobule, portal-venous or arterial vessels (PV). Data from n = 3–4 mice per condition.
Scale: 50 microns.
(B) Histo-cytometry analysis of NASH liver sample. Statistical information of segmented objects (F4/80+tdTomato+ surfaces) was imported into FlowJo and
subsequently gated on F4/80 and tdTomato expression to quantify KC-N, KN-RM, and Ly6Chi&lo-RM cells. Tim4 mean fluorescence intensity (MFI) for each gated
population are shown in the middle panel, and Mgl2 expression on Ly6Chi&lo-RM cells macrophage is shown at right to distinguish Ly6Chi (Mgl2 ) or Ly6Clo
(Mgl2+) cells.
(C) Confocal image of NASH liver showing distribution of rendered surfaces of KC-N, KN-RM, and Ly6Chi&lo-RM (Rec) cells. Dashed lines denote peri-portal
regions as in (A). Scale: 50 microns.
(D) Distance and phenotype (KN-RM, Mgl2lo-RM, or Mgl2hi-RM) of closest neighbor to KC-N cells in NASH livers. Data pooled from n = 4 mice. Wilcox two-sided
test; p < 0.001(***).
(E) Distance to nearest portal or central vein vasculature (large diameter vessels defined to be greater than 15 mm in diameter) of KC-N, KN-RM, and Ly6Chi&lo-RM
cells. Kruskal-Wallis with Dunn test; p < 0.001(***).
(F) Zoom-in representative confocal image of NASH liver showing distribution of rendered surfaces of Mgl2lo (Ly6Chi) and Mgl2hi (Ly6Clo)-RMs. E-cadherin
demarks the peri-portal regions, and highlighted by white circles are CV and PV blood vessels. Scale: 30 microns.
(G) Distance to nearest center of CV or PV of Ly6Chi and Ly6Clo RM cells (mm). Kruskal-Wallis with Dunn test; p < 0.01(**), p < 0.001(***).
(H) Immunofluorescence (IF) image assessing in situ cell death via TUNEL staining in addition to Tim4 and F4/80 staining. Image from NASH mice; maximum
intensity projection (MIP) of a 20-mm z stack. Scale: 5 microns for the top row and 10 microns for the bottom row.
(I) Quantification of total TUNEL+ hepatic macrophages per area (1000 mm2) in Ctrl and NASH mouse livers.
(J) Nearest neighbor distance of all TUNEL+ cells to KN-RM, Ly6Chi or lo-RM, and KC-N cells in NASH livers, data pooled from n = 4 mice. Kruskal-Wallis with Dunn
test; p < 0.001(***).
(K) Temporal assessment of percentage of total CD11bloF4/80+CD146 cells that are Tim4 KN-RM from mice fed a NASH diet as indicated.
Please see also Figure S2 and Figure S3.

Immunity 52, 1057–1074, June 16, 2020 1061

 ll
Article

A EMP Blood
C Mean expression
D Mean expression
Ly6Chi (log2(TPM+1)) (log2(TPM+1))
15 15
316 53

KN-RM
10 10

KC-H
Ly6Chi Ly6Clo 5 5
KC-H KC-N KN-RM -RM -RM

Ctrl NASH 521 40

0 0
Sinusoids Large vessels 0 5 10 15 0 5 10 15
KC-N KC-N
Cluster 1

E F

2,210 genes; Log2FC > 2; p.adj < 0.05; TPM > 4

15 15
1046 1442

Ly6CHi-RM
10 10

KN-RM
Cluster 3 Cluster 4

5 5

0 1193 0 685

0 5 10 15 0 5 10 15
Ly6Chi-RM Blood Ly6Chi Mono
Cluster 2

G Kupffer cell signature ROS metab. signature

C6 Arg2
1500
Cd163 200 Txnrd1
1000
100
500
Mean Expression (TPM)

z-scaled (Log2(TPM+1)) 0 0
−2 −1 0 1 2
K H
KN C-N
6C -RM

K H
Ly N- -N
6C RM
6C M

6C RM

o
6C M

6C M
-

-
-M

-M
KC

KC

K C
Ly h-iR

Ly -R

Ly -R
hi

hi
Ly h-i
lo

lo
B Ly6Chi-RM
Ly

30
Ly6Clo-RM Scar-assoc. mac (SAM) Response to wounding
20 800
RLM 12h Cd9 Clec10a
75
Trem2 Mgl2
PC2 (19%)

10 RLM 1d 600
Blood 50
0 KN-RM 400
Ly6Chi
200 25
−10 KC-N Monocytes
(Ctrl and NASH)
0 0
−20 RLM 14d
-H

Ly KN -N
Ly h-i M

-H

Ly KN -N
Ly h-i M
6C RM

6C RM

o
o
6C M

6C M
6C -R

6C -R

-M
-M
KC
KC

KC
KC
Ly -R

Ly -R

−30 KC-H
hi
hi
lo

lo

−20 0 20 40 60
PC1 (57%)
H I Mean expression J K NASH CRN
Clec4f+Tim+ Clec4f+Tim– (log2(TPM+1)) Score
15 6
DTR–

Total KC KC KC

6
DTR+, n= 12

10 3 10 4
Percent CD45+

4 2
5
DTR+

5 2
2 1 DTR–
0
DTR+
DTR –
0 5 10 15 0
0 DTR+ 0 0 DTR–, n = 10

(legend on next page)

1062 Immunity 52, 1057–1074, June 16, 2020

 ll
Article

mice fed a control diet (Figures 2H and 2I). In both control and
NASH livers, TUNEL+ KN-RM and Ly6Chi&lo-RM cells were rare
(Figure 2I). By nearest-neighbor analysis of TUNEL+ cells, KN-
RM cells were significantly enriched in nearby areas as
compared to Ly6Chi&lo-RM and KC-N cells (Figure 2J). Time
course experiments indicated that KN-RM cells were detected
by 10 weeks after initiation of the NASH diet and progressively
accumulated to account for more than 50% of the KC population
at 30 weeks (Figure 2K). These results indicate that Tim4+ KCs
undergo cell death during NASH, potentially resulting in partial
opening of the niche and enabling repopulation by HSC-derived
KN-RM cells.
Effect of NASH Diet on Resident and Recruited Myeloid
Cell Transcriptomes
The ability to sort distinct populations of hepatic myeloid cells
defined by scRNA-seq analysis enabled deep transcriptomic
profiling of each population (Table S2). Unsupervised hierar-
chical clustering of 2,210 DE genes among KC-H, KC-N,
KN-RM, Ly6Chi-RM, and Ly6Clo-RM tightly grouped KC-N
and KN-RM and distinguished KC-H, KC-N, and KN-RM cells
from Ly6Chi&lo-RMs (Figure 3A). With inclusion of DE genes in
blood monocytes, principal component analysis (PCA) also
indicated that KC-N and KN-RM cells from NASH mice group-
ed most closely with KC-H cells from healthy mice (Figure 3B).
After experimental KC ablation, blood Ly6Chi monocytes are
recruited to the liver and rapidly differentiate into KC-like liver
macrophages (Bonnardel et al., 2019; Sakai et al., 2019; Scott
et al., 2016). At 14 days after KC ablation, these repopulating
macrophages (GEO: GSE128662) also grouped closely with
KC-H, KC-N, and KN-RM (Figure 3B). Collectively, these re-
sults suggest that monocyte-derived KN-RMs follow a similar
developmental trajectory in the context of NASH to that of re-
populating liver macrophages (RLMs) in the KC abla-
tion model.
Pairwise comparisons of KC-H and KC-N cells indicated
that the NASH diet had a large impact on resident KC gene
expression (>800 DE genes), consistent with scRNA-seq
data (Figure 3C). Notably, fewer than 100 genes were identi-
fied as significantly altered when comparing KC-N and KN-
RM KCs from mice with NASH (Figure 3D), despite their
distinct origins (EMP versus HSC). In contrast, during NASH,
KN-RM cells exhibited more than 2,200 DE genes in compar-
ison to Ly6Chi-RM, despite both cells originating from an HSC
precursor (Figure 3E). The divergent differentiation programs
of KN-RM and Ly6Chi-RM were further reinforced by pairwise
comparisons with gene expression in Ly6Chi circulating
monocytes, indicating more than 2,000 DE genes in each
case (Figures 3F and S4A).
Prior studies defined a set of KC identity genes that distinguish
the KC from other tissue-resident macrophages (Lavin et al.,
2014). Twenty-eight of the genes downregulated during the tran-
sition of KC-H cells to KC-N cells in the NASH model are among
this set, exemplified by Cd163 and C6 (Figures 3G and S4B).
However, many KC-specific genes maintained similar expres-
sion in cells from mice on the NASH diet, including C1qa, Cd5l,
Id3, and Il18bp, indicating that only a subset of the KC-specific
gene expression program is altered during NASH. Furthermore,
KC identity genes such as Clec4f, Vsig4, and Cdh5 were highly
expressed in KN-RM, but neither Ly6Chi&lo-RM population.
Considering the differential spatial distribution of these cell pop-
ulations, such findings suggest that the KC niche is necessary to
promote induction of these genes (Figure S4B).
Gene ontology analyses of genes whose expression distin-
guishes KC-H, KC-N, and KN-RM cells from Ly6Chi&lo-RM
(clusters 1 and 2, Figure 3A) were consistent with a stronger
pro-inflammatory and wound repair phenotype of the Ly6-
Chi&lo-RMs (Figure S4C), in agreement with prior studies (Hey-
mann and Tacke, 2016). Examples of DE genes associated
with the functional categories of ‘‘ROS metabolism’’ and
‘‘response to wounding’’ are illustrated in Figure 3G. Ly6Clo-
RM preferentially expressed Mgl2 (CD301b), a gene recently re-
ported to be expressed in skin macrophages that activates a
specific myofibroblast population implicated in tissue repair
and aging (Shook et al., 2018). In addition, we observed
increased expression of Trem2 and Cd9 in both KC-N and
KN-RM cells, consistent with recent observations identifying
these genes as markers for macrophages associated with
increased lipid burden and the presence of tissue scarring
(Hill et al., 2018; Jaitin et al., 2019; Ramachandran et al.,
2019; Xiong et al., 2019). Collectively, these results indicate
that the NASH diet acts to reprogram the endogenous KC pop-
ulation and induce divergent programs of differentiation of KN-
RM cells and Ly6Chi&lo-RMs.

Figure 3. Highly Divergent Gene Expression Patterns across Myeloid Populations in NASH
(A) Unsupervised hierarchical clustering of DE genes in the indicated cell types in control and NASH liver.
(B) PCA of 2,000 most variable genes in RNA-seq data from myeloid cells in liver and blood from healthy and NASH-diet-fed mice, or recruited liver macrophages
from Sakai et al. (2019), n = 2–3 per group.
(C) Scatterplot of RNA-seq data in healthy (KC-H) or NASH (KC-N) Tim4+ KCs. DE genes identified by DESeq2 (FC > 2, p-adj < 0.05) are colored in red.
(D) Comparison of NASH Tim4+ KCs (KC-N) and Tim4 Kupffer niche recruited macrophages (KN-RM).
(E) Comparison of NASH Tim4 Kupffer niche recruited macrophages (KN-RM) and Ly6Chi-RM.
(F) Comparison of NASH blood Ly6Chi monocytes and Ly6Chi-RM.
(G) RNA-seq expression (mean TPM ± SD) of representative genes. TPM = transcripts per kilobase million).
(H) Percentage of hepatic CD45+F4/80hiCD11bloLiveSinglets from Clec4f-cre R26-iDTR / (n = 6) or Clec4f-cre R26i-DTR+ (n = 6) mouse livers. Mice were treated
with DT, allowed to rest for 4 weeks, then fed the choline-deficient, L-amino acid-defined, high-fat diet (CDAHFD) for 4 weeks.
(I) RNA-seq from whole liver tissue from Clec4f-cre R26-iDTR / (n = 10) or Clec4f-cre R26-iDTR+ mice (n = 12) as in (H). Zero DE genes (DESeq2). Boxes denote
quartiles, and whiskers denote data range.
(J) Representative (n > 5) H&E staining of liver tissue.
(K) NASH Clinical Research Network (CRN) scoring of stained liver sections. Scores > 5 are indicative of steatohepatitis. No significant difference was identified
using Kruskal-Wallis rank sum test (p = 0.71). Boxes denote quartiles, and whiskers denote data range.
Please see also Figure S4.

Immunity 52, 1057–1074, June 16, 2020 1063

 ll
Functional Convergence of Embryonic KCs and KN-RMs
in NASH
To clarify the functional biological roles of embryonic KCs and
monocyte origin KN-RM cells, we used the Clec4f-cre R26-
iDTR system (Sakai et al., 2019). We treated DTR+ or DTR
mice with diphtheria toxin (DT) to ablate embryonic KCs and
then allowed the KC niche to repopulate with monocyte precur-
sors. After a 4-week recovery period, the mice were subjected to
ad libitum feeding for 4 weeks with a rapid NASH-inducing model
(Matsumoto et al., 2013). Ablation of embryonic (KC-H) KCs
before feeding mice the NASH diet led to increased proportions
of Tim4 KN-RM cells, as expected (Figure 3H). To assess the
global molecular consequence of replacing embryonic KCs
with monocyte-derived KN-RMs, we performed RNA-seq on to-
tal liver tissue (Figure 3I). Notably, we found no DE genes be-
tween the two groups. We also found no differences in 23 circu-
lating inflammatory cytokines (Figure S4D). Finally, quantitative
histological assessment of liver sections indicated no effect of
replacing KCs with KN-RM. These results are consistent with
the largely convergent transcriptomes observed in comparison
of KC-N and KN-RM (Figures 3J and 3K).
Niche Occupancy Reprograms the Epigenetic
Landscapes
To investigate mechanisms responsible for environment-spe-
cific programs of gene expression, we identified accessible
chromatin defined by ATAC-seq in five populations of cells
from mice with NASH: KC-N and KN-RM cells, Ly6Chi&lo-RMs,
and Ly6Chi blood monocytes. Examples of ATAC-seq peaks in
these five populations of cells from mice fed the NASH diet in
the vicinity of the Clec4f and Itgam genes, which are highly differ-
entially expressed in KCs compared to Ly6Chi&lo-RMs, are illus-
trated in Figure 4A. Genome-wide comparisons of ATAC-seq
peak tag counts for Ly6Clo-RM versus Ly6Chi blood monocytes
are illustrated in Figure 4B, and the same comparison for KN-RM
and Ly6Chi blood monocytes is shown in Figure 4C. Differential
regions from the five myeloid populations in NASH mice, as
well as healthy KC-H cells and 24- and 48-h RLMs after KC abla-
tion (Sakai et al., 2019), were used for PCA illustrated as in
Figure 4D. PC1, accounting for ~65% of variance, primarily
distinguished KC-H, KC-N, and KN-RM populations from Ly6Chi
blood monocytes and Ly6Chi&lo-RM. RLMs 24 and 48 h after KC
ablation became incrementally closer to KCs along PC1, reflect-
ing chromatin remodeling after arriving at the KC niche. PC2, ac-
counting for ~13% of variance, primarily separated Ly6Chi blood
monocytes from Ly6Chi&lo-RM.
Hierarchical clustering of ATAC-seq data further support the
relationships suggested by PCA, with KC-N and KN-RM cells ex-
hibiting highly similar patterns of open chromatin that are distinct
from the transitional patterns observed in Ly6Chi&lo-RM (Fig-
ure S5A). Based on these findings and the results of lineage-
tracing experiments, we considered the open chromatin regions
of KN-RM and Ly6Clo-RM populations as divergent endpoints of
chromatin remodeling events following entry of Ly6Chi blood
monocytes into the NASH liver. ATAC-seq peaks specific for
KN-RM or Ly6Clo-RM in comparison to circulating Ly6Chi mono-
cytes (red data points in Figures 4B and 4C, respectively) indi-
cated that approximately 75% of these peaks were specific to
KN-RM or Ly6Clo-RM cells (Figure 4E). Peaks specific for KN-
1064 Immunity 52, 1057–1074, June 16, 2020
Article
RM cells were highly enriched for motifs recognized by LXR
and members of the MAF and TFE families (Figure 4E). In
contrast, ATAC-seq peaks specific for Ly6Clo-RM cells were
highly enriched for NF-kB motifs and for motifs recognized by
RUNX, ZEB, and KLF TFs. Open chromatin regions lost from
blood monocytes during acquisition of either the KC or Ly6Chi&lo
-RM niche signatures (Figures 4B and 4C, blue points) were en-
riched for KLF and C/EBP motifs (Figures S5B and S5C). The
open chromatin regions lost during acquisition of the KC niche
additionally included motifs for RUNX and CTCF (Figure S5B).
The pattern of motif enrichment in KC-N and KN-RM cells im-
plies that liver niche signals increase the expression and/or activ-
ities of TFs that bind to LXR, MAF, and TFE motifs. Consistent
with this possibility, Nr1h3 (encoding LXRa), Mafb, Tfec, and
Id3 (Lavin et al., 2014; Mass et al., 2016) are among a set of
KC LDTFs higher expressed in KC-N and KN-RM cells than in
Ly6Chi&lo-RM cells (Figure 4F; Table S2). Importantly, these
TFs are highly induced in RLMs within 12 h of entry into the
open KC niche of the liver (Sakai et al., 2019). These findings
further support a model in which KN-RM cells largely follow the
developmental program taken by RLMs following KC depletion
upon adherence to LSECs. In contrast, the lack of induction of
KC LDTFs in Ly6Chi&lo-RM cells is consistent with their location
outside of the sinusoidal space. NF-kB motifs are enriched in
KC enhancers in comparison to other tissue-resident macro-
phages (Sakai et al., 2019), but the particularly strong enrichment
for this motif in Ly6Clo-RM cells implies that the niche occupied
by these cells provides additional signals that activate NF-kB. In
addition, while RUNX factors are lower in KC-N and KN-RM cells
in comparison to blood Ly6Chi monocytes, their expression is
maintained or increased in Ly6Chi&lo-RMs (Figure S5D). In con-
cert, analysis of open chromatin provides evidence that the
divergent patterns of gene expression observed in HSC-derived
KN-RM and Ly6Chi&lo-RM cells are in part determined by
whether or not they receive niche-specific signals necessary to
adequately induce KC LDTFs.
NASH Diet Reprograms the KC Enhancer Landscape
We next sought to understand the basis for the altered expres-
sion of the more than 900 mRNAs during the transition of KC-H
cells in the healthy liver to KC-N cells in the NASH liver (Figures
3A and 3C). Corresponding changes in the regulatory land-
scapes during this transition would potentially enable inference
of TFs and upstream signaling pathways that mediate re-
sponses to the NASH diet. In contrast to changes in open chro-
matin observed in transition of Ly6Chi blood monocytes to KN-
RM cells (Figures 4C and 4D), relatively few differences in open
chromatin were observed comparing KC-H and KC-N cells
(Figure 5A, left). The enriched motifs associated with the
increased chromatin accessibility included sites for AP1 and
EGR factor binding, whereas sites of decreased chromatin
accessibility were enriched for PU.1 and SpiC motifs (Figure 5A
right).
To investigate potential changes in the transcriptional func-
tions of these regions, we performed chromatin immunoprecip-
itation sequencing (ChIP-seq) for acetylation of histone H3 lysine
27 (H3K27ac) in KC-H cells in the healthy liver and KC-N cells in
the NASH liver. H3K27ac is deposited by histone acetyltrans-
ferases (HATs) associated with transcriptional co-activators
 ll
Article

A ATAC-seq RNA ATAC-seq RNA B C

ATAC-seq ATAC-seq
Clec4f Itgam (log2(norm. tags +1)) (log2(norm. tags +1))
5 kb 10 kb 14 14
Ly6Chi 3e3
4e2
2e3
Blood 1e3 2e2
12 8507 12 6732
0 0
3e3
Ly6Chi 2e3
4e2 10 10

Ly6Clo-RM
-RM 1e3 2e2

KN-RM
0 0 8 8
Ly6Clo 3e3
4e2
2e3 6 6
-RM 1e3 2e2
0 0 4
3e3
4
4e2
KC-N 2e3
2 2
1e3 2e2 3509 2458
0 0
3e3 0 0
KN-RM 4e2
2e3
1e3 2e2 0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
0 0
Ly6Chi Blood Ly6Chi Blood

D Ly6Chi E
Blood
Control Gained during niche acquisition versus
40
KC-H ATAC-seq peaks Ly6Chi peripheral blood monocytes
hi
Ly6C
20 Blood Motif P- %targ/ Best
NASH value %bkgd Match
KC-N
0 GG
AA
GTCACA G
TGT AC GT 1e-101 30/17 TFE/MITF
PC2 (13%)

C A
TTAA
TG

RLM
T
CTG
CC G T A G T
C
G
AC G
CC
T A CA

48hr RLM TGACCT TGTGACCCT 1e-91

ATTG
C
AGATCG
AA
CGC
T
TT
CGAA
C
A
A
5.1/0.9 LXRE
24hr
G
CC T G A G
A
C G
C A T G T
C G T G T A G A

−20 4,500 only

C
G
AGTCAGCA 1e-79 33/21 MAF
KN-RM T C
A T G
C AG T
AG
C T C
T
G T A
A G
C T

Ly6Chi-RM up in KN-RM CGTGGGAGAGC

G
T
A
CC
T
T
C
A
G A
A
C
A
TTC
C
A
G
TT
CT
A C
A
TG
1e-22 12/7.6 RBPJ
−40 GGG T C
AATTCC C
A
GAG
TAC
TGC 1e-91 20/6.8 NF-κB
2,226 up C
AGT
A C T A C TT
C
GG
CCG A
G AG T AA G T G
ATA
TCCT

−60 in both GAAAAACGATGGAAAAT

C T
A G
C T G
C
T
C T GG T
C G
C TG T
A C T C
G AC A T C T A G T
C G T
C G T
C G T
C G
C A
1e-79 2.1/.02 NFAT
6,279 only up CA
G

C
T
G
TGACGTCATC
AC
AT

C
T

G T
C

G
C A
GCAA
G
C
TA
A G
C
G
C

G
C A
GC
T

G T
G

G T
C
T
GT
CG
AA
A G
C A G
C A G
C A G
C A G T A G
C A G
C A
1e-75 31/16 JUN-CRE
Ly6Clo-RM in Ly6Clo GG T CCCC
−80 AAAAC
T
A GT
TTCG
AT
AA
T
A
1e-522 34/10 NF-κB
-RM
C
TCCGC G GT GT
G
A
G

GG
C
CA
TA
AAGC
TCT
TC
GC
T
C
GTAT
C
G
AA
T
C
AGGTGA G
C
A
T 1e-148 37/23 ZEB
−100 −50 0 50 100 CAA
A ACCT
ACAG 1e-102 39/27 RUNX
GGC TG
TG
A
T
TC C TG A T G T AG C G T A G
C T C

PC1 (65%)

F Nr1h3 Mafb Tfec Id3

Mean Expression (TPM)

800 400 400

60
600 300 300

40
400 200 200

200 100 20 100

0 0 0 0
KC-H
KC-N
KN−RM
Ly6Chi-RM
Ly6Clo-RM
Ly6Chi Blood

KC-H
KC-N
KN−RM
Ly6Chi-RM
Ly6Clo-RM
Ly6Chi Blood

KC-H
KC-N
KN−RM
Ly6Chi-RM
Ly6Clo-RM
Ly6Chi Blood

KC-H
KC-N
KN−RM
Ly6Chi-RM
Ly6Clo-RM
Ly6Chi Blood

Figure 4. Niche-Specific Reprogramming of Epigenetic Landscapes

(A) UCSC genome browser tracks of ATAC-seq signals in the vicinities of the Clec4f and Itgam genes in the indicated cell types. Bar plots to the right of each track
represent the RNA-seq gene expression (mean TPM ± SD) in each population.
(B) Genome-wide comparison of normalized ATAC-seq peak tags at enhancer-like regions (>3 kb removed from transcription start site [TSS]) comparing Ly6Chi
blood monocytes and Ly6Clo-RM during NASH. Differential regions were identified using DESeq2 (FC > 2 and p-adj < 0.05 using independent biological du-
plicates).
(C) Genome-wide comparison of normalized ATAC-seq peak tags at enhancer-like regions (>3 kb removed from TSS) comparing open chromatin in Ly6Chi blood
monocytes and KN-RM.
(D) PCA of ATAC-seq datasets (n = 2–3) for the top 10,000 most variable distal (>3 kb from TSS) regions in myeloid cell populations during NASH, or recruited liver
macrophages (RLM) from Sakai et al. (2019).
(E) De novo motifs enriched in distal open chromatin regions (>3 kb from TSS) enriched in KN-RM (top), Ly6Clo-RM (bottom), or enriched in both populations
(middle) compared to Ly6Chi blood monocytes during NASH. The background for motif enrichment analysis is the distal open chromatin from Ly6Chi blood
monocytes.
(F) Expression (mean TPM ± SD) of KC LDTFs in the indicated cell types.
Please see also Figure S5.

Immunity 52, 1057–1074, June 16, 2020 1065

 ll
Article

A D
ATAC-seq Differential Kupffer ATAC-seq
(log2(norm. tags +1)) RNA- vs H3K27ac ChIP-seq
14 P- %targ/ Best
402 Motif 10 R = 0.70
value %bkgd match

log2(KC-N/KC-H) RNA-seq
12
G
CTGAGTCA
T
A G
C A C T A C
G T
C T
A G
C A G T AG T
C
1e-87 69/22 AP1/ATF
ACCCACGCG
CG
10 T 1e-15 19/7 EGR2 5
TT AAA
TTGC
ACGATCG
G C A T A G T G A G T T

8
ACTTCCTCTT
KC-N

G G
A
T
AA 1e-24 60/13 PU.1/SpiC
0
CG T T A G
C A G
C A G T
A G A
C
TCGG A C
G

6
4 C −5
2 Effect of NASH on H3K27ac
90 All enhancers KC signature
0
Up 10%
0 2 4 6 8 10 12 14 Up 3.5% −6 −4 −2 0 2 4 6
KC-H Down Down log2(KC-N/KC-H) H3K27ac ChIP-seq
8% 24%
B E
H3K27ac ChIP-seq
(log2(norm. tags +1)) Egr2 ***
Differential Kupffer H3K27ac ChIP-seq
43352 total 10452 KC sig Atf3 ***
14 P- %targ/ Best Fos
4201 Motif value %bkgd match Srebf1
365 KC sig ***
12 Jun *
GA
T
CC
TTGA TCA
AG
T
C
G
A
T A
C
G
A
T
1e-137 35/19 AP1/ATF Irf8 ***
TTTCTAATATTTTCC
G
C A C A G T
C G
C A G T G T
C

10 A
A

1e-25 0.7/.04 NFAT Runx1

G
C TGCCG CC AT

***
TGTGGTTA
G
C A A G
C A G A C G T G T C T G A G A C
G A G
C A G T G A

C A 1e-20 11/7.3 RUNX Mafb **

8 CGCCCACC
G
C A C T A G
C A C T A C T A G A G
C G
C
KC-N

G
T
A
C T
A G T A G A
T
G G
TTTTA G
A CG A
1e-19 16/11 EGR Nr1h2 ***
Mitf *
6
Irf7 *
A AGGTCA
TCG
A
1e-33 49/39 NR half-site
4
G
C G
T C T C T A C T A G
C A G T A G T
C

Nr1h3
AGTGGTCACAGGAC 1e-25 AATAT
T
AG C
T
G
A
0.4/0.0 MITF Irf5
TCTGTCAGCCTATCTG
G T
C C T A G
C C T C A G
C G A G T
C G T C C T A T A G T
C

C 1e-19 0.3/0.0 MAF

2 CA C G GAG GG Maf
GAAAGTAAAAGCAC 1e-17
G A G T G
C A T A G
C A T A G T
C C T A T A G T C A C T C A G T A G
C A T A

3583 C T A G T
C G T
C G T
C C
C
TAC A G G T G T
C G T
C G T
C CG
T A T A G T
C G T A
0.3/0.0 IRF Tfec **
0 2553 KC sig Irf1 **
G GGTTACTA AGGTCA 1e-15 3.5/1.6 LXRE
C
CG
A C T
TCG G T
Spic
0 2 4 6 8 10 12 14
C TC
T
ATCGTA
G
G
A C AG A
A G
C AC TC T
A
C
A TG
C ATG G A
T
C

(known) ***
KC-H 0 2 4
log2(NASH/Ctrl)
F G RNA-seq (TPM)
C6 Cd9
5 kb ATAC 20 kb ATAC
KC-H
RNA-seq (TPM)

KC-N 100 Atf3

Egr1
H3K27ac H3K27ac 50 Egr2
Spic
0

Cd163 Trem2
10 kb ATAC 5 kb ATAC
800 C6
RNA-seq (TPM)

Cd163
400 Cd9
H3K27ac H3K27ac
Trem2
0

0 1 4 10 20 30
Weeks on AMLN diet

Figure 5. The NASH Diet Alters the Activity States of Resident KC Enhancers
(A) Left: Normalized ATAC-seq signal at all distal open chromatin regions (>3 kb from TSS) in Tim4+ KCs from healthy mice (KC-H) or Tim4+ KCs from mice on
NASH diet (KC-N). Regions with significantly more chromatin accessibility during NASH are colored in red, while regions with less accessibility during NASH are
colored blue. Right: De novo motif enrichment from differentially accessible chromatin region shown at right.
(B) Left: H3K27ac ChIP-seq signal around distal (>3 kb from TSS) ATAC-seq peaks in a 2,000 bp window. Differentially acetylated regions were determined using
DESeq2 (FC > 2, p-adj < 0.05). Regions overlapping with KC signature enhancers (Figure S6A) are colored green. Enhancers with more acetylation during NASH
are colored red, or orange if also a KC signature enhancer. Enhancers with less acetylation during NASH are colored blue, or purple if also a KC signature
enhancer. Right: Representative motif enrichment from differentially acetylated chromatin regions.

(legend continued on next page)

1066 Immunity 52, 1057–1074, June 16, 2020
 ll
Article

and is highly correlated with regulatory element activity
(Creyghton et al., 2010). Out of 43,352 total distal open chro-
matin regions defined by ATAC-seq, 4,201 putative enhancers
gained H3K27ac and 3,583 putative enhancers lost H3K27ac
in response to the NASH diet (Figure 5B), affecting 18% of the
enhancer-like regions overall (Figure 5C).
A comparison of H3K27ac associated with ATAC-seq peaks in
KC-H cells versus in other macrophage populations enabled
definition of a set of 10,452 putative KC-H signature enhancers
(Figure S6A; Sakai et al., 2019). A profound overlap was
observed between KC signature enhancers and enhancers
downregulated by the NASH diet, in total identifying 2,553 KC
signature downregulated enhancers (24% of KC signature en-
hancers, 71% of all downregulated enhancers, Fisher’s exact
test p value < 0.001) (Figures 5B and 5C). Conversely, only 365
KC signature enhancers were upregulated by NASH diet (3.5%
of KC signature enhancers, 8.7% of all upregulated enhancers,
Fisher’s exact test p value = 1) (Figures 5B and 5C). Changes
in H3K27ac were highly correlated with changes in expression
of the nearest gene (Figure 5D). These findings suggest a prefer-
ential suppressive effect of the NASH diet on gene regulatory
networks governing the function of KC enhancers, in line with
the corresponding downregulation of KC identity genes
(Figure S4B).
ATAC-seq and H3K27ac ChIP-seq tracks associated with the
downregulated KC identity genes C6 and Cd163 illustrated sub-
stantial loss of H3K27ac at their respective promoters and distal
regions of open chromatin (Figure 5F). The opposite pattern was
observed at the upregulated lipid-associated macrophage (LAM)
and SAM-defining genes Trem2 and Cd9 (Jaitin et al., 2019;
Ramachandran et al., 2019). In addition, Trem2 and Cd9 both
provided examples of genes in which new regions of open chro-
matin are established that are associated with H3K27ac (Fig-
ure 5F), representing putative NASH-dependent enhancers.
Both pre-existing and NASH-induced regions associated with
Trem2 were shifted but conserved at the TREM2 locus in human
microglia (Figure S6C). Additional examples of NASH-induced
genes in KCs and associated gene ontologies are shown in
Figure S6B.
Motif enrichment analysis of regions of open chromatin
exhibiting loss of H3K27ac in response to the NASH diet
showed enrichment for binding sites recognized by LXR,
MAF, and IRF TFs (Figure 5B right). Each of these motifs are
recognized by TFs that are established or proposed to drive
KC identity (Lavin et al., 2014; Mass et al., 2016; Sakai
et al., 2019; Scott et al., 2018). In contrast, open chromatin
regions with gained H3K27ac were enriched for de novo
motifs matching AP1, NFAT, RUNX, and EGR TFs (Fig-
ure 5B right).
Comparison of KC-H to KC-N cells indicated significant induc-
tion of Atf3, Fos, Jun, Egr2, and Runx1 mRNAs (Figures 5E and
5G), suggesting that increased expression of these TFs during
NASH may contribute to activation of enhancers with corre-
sponding DNA binding elements (Figure 5B). Conversely,
mRNAs encoding a subset of KC LDTFs were downregulated
in response to the NASH diet, including Spic, Irf1, and Tfec.
Notably, SpiC (encoded by Spic) binds to a motif nearly identical
to that recognized by PU.1 that is enriched at genomic regions
exhibiting loss of open chromatin in NASH. In contrast, while
LXR motifs were highly enriched at genomic regions exhibiting
loss of open chromatin, expression of Nr1h3, encoding LXRa,
was not affected by the NASH diet, and expression of Nr1h2, en-
coding LXRb, was slightly increased (Figure 5E).
To relate changes in TF expression to global changes in gene
expression, we performed RNA-seq analysis of KCs following
1, 4, 10, 20, and 30 weeks of the NASH diet. Coordinated
changes in gene expression began to occur between 4 and
10 weeks of the NASH diet, with upregulation of Atf3 and
Egr2 associated with increases in Trem2 and Cd9 expression
and downregulation of Spic associated with decreases in
Cd163 and C6 expression (Figure 5G). These findings indicate
that acquisition of features corresponding to the LAM or SAM
phenotype characterized by high expression of Trem2 and
Cd9 requires prolonged exposure to the NASH diet, which ulti-
mately results in substantial reprogramming of the KC regula-
tory landscape.
NASH Diet Selectively Impairs LXR Regulation of KC
Identity Genes
The observation that LXR binding motifs were highly enriched in
genomic regions exhibiting diet-induced loss of H3K27ac led to
a focused analysis of LXR function. Comparison of effects of
LXR deletion with effects of the NASH diet on KC gene expres-
sion indicated that a subset of LXR target genes associated
with its role as a KC LDTF were strongly downregulated by
the NASH diet. For example, Cd5l, Timd4, Cd209l, Pcolce2,
and Plac8 require LXRa for expression (Sakai et al., 2019),
and all but Cd5l exhibited significantly reduced expression in
response to the NASH diet (Figure 6A). However, many canon-
ical LXR target genes such as Abca1, Abcg1, Mylip, and Srebf1
were significantly upregulated in response to the NASH diet
(Figure 6B; Table S3), arguing against a general loss of LXR
function. Targeted lipidomic analysis of known LXR ligands
(Peet et al., 1998) in liver indicated that the concentration of

(C) Summaries of percent representation of (left) differentially activated enhancers between control and NASH KCs, or (right) differentially acetylated enhancers
intersected with the KC signature enhancers.
(D) Ratio-ratio plot depicting fold change in H3K27ac ChIP-seq signal at enhancers (2,000 bp window centered on ATAC-seq peaks >3 kb from TSS) compared to
fold change in mRNA expression of closest gene annotated to enhancer region in Tim4+ KCs in NASH diet mice (KC-N) versus healthy mice (KC-H). Points colored
in blue are significantly different (FC > 2, p-adj < 0.05) for both H3K27ac ChIP-seq signal at enhancers and closest mRNA. Pearson correlation of 0.70 denotes
relationship between the highlighted differential data points.
(E) Log2FC of candidate TFs known to bind DNA elements found enriched in (B) for KCs from healthy mice (KC-H) and NASH mice (KC-N). *p-adj < 0.05, **p-adj <
0.01, and ***p-adj < 0.001 using DESeq2.
(F) UCSC genome browser tracks of ATAC-seq or H3K27ac ChIP-seq signals in the vicinities of the indicated genes.
(G) Mean TPM (LOESS fit) of the indicated genes in Tim4+ KCs from mice fed the AMLN NASH diet as indicated (0 week: n = 3; 1 week: n = 2; 4 week: n = 2; 10
week: n = 3; 20 week: n = 3; 30 week: n = 4).
Please see also Figure S6.

Immunity 52, 1057–1074, June 16, 2020 1067

 ll
Article

A B C

D E

Figure 6. Genome-wide Occupancy of LXR Binding in KCs during NASH

(A) RNA-seq expression (mean TPM ± SD) of LXR-dependent KC signature genes. *p-adj < 0.05, **p-adj < 0.01, and ***p-adj < 0.001 using DESeq2.
(B) RNA-seq expression (mean TPM ± SD) of canonical LXR target genes.
(C) Quantification of desmosterol, 24-, 25-, and 27-OHC and 24,25-EC in livers from control or NASH mice. *** = p < 0.001.
(D) Mean LXR ChIP-seq signal at merged irreproducibility discovery rate (IDR) peaks from KC nuclei. Nuclei were sorted from Clec4f-Cre-NLS-TdTomato mice fed
either a control diet (n = 2) or the NASH diet for 20 weeks (n = 2). Peaks with significantly altered LXR binding were determined using DESeq2 (FC > 2, p-adj < 0.05).
(E) ATAC-seq, H3K27ac ChIP-seq, and LXR ChIP-seq signals in the vicinities of the indicated genes.
(F) Normalized distribution of LXR ChIP-seq density or H3K27ac ChIP-seq density at enhancers with either significantly altered LXR binding during NASH or all
LXR bound enhancers.
(G) Left: De novo motif enrichment for NASH-gained LXR peaks (right) or NASH-lost LXR peaks (right) shown in Figure 7D.
Please see also Figure S7.

1068 Immunity 52, 1057–1074, June 16, 2020

 ll
Article

A B ATF3 ChIP-seq
IDR peaks
LXR gained LXR lost LXR total Control NASH
AP1/ATF3 7
Motifs per bp per peak

0.008 NR/LXR

Log2(Tags+1)
6
Egr2 0.004 0.003 5
PU1 25,507 4
0.006
0.003 3
0.002 2
0.004 0.002 1
2,808 0
0.002 0.001 0.001
13,291
0.000
0.000 0.000
−200 −100 0 100 200 −200 −100 0 100 200 −200 −100 0 100 200
1,000 bp
Distance from peak (bp)

C D
NASH specific ATF3 vs all ATF3 peaks NASH TF ocupancy at
P- %targ/ Best NASH gained enhancers
Motif value %bkgd match None 32%
ATF3 10%
1e-671 49.26/23.38 AP1/ATF3 LXR 3%

1e-111 11.06/5.27 Egr2/Egr1

1e-69 42.57/33.94 MITF/USF1

Both 54%

E
ChIP-seq tag densities centered on genomic regions co-bound by LXRs and ATF3
ATF3 LXR P300 H3K27ac
ChIP fragment depth

12 8 Control
20 10.0
per bp per peak

15 9 6 NASH
7.5

10 5.0 6 4

5 2.5 3
2
0
0 0 0 00 00 0 0 0 00 00 0 0 0 00 00 0 0 0 00 00
00 00 10 20 00 00 10 20 00 00 10 20 00 00 10 20
−2 −1 −2 −1 −2 −1 −2 −1
Distance from enhancer (bp)

F Control NASH
Trem2 Cd9 Arhgap22 Kcnn4
5 kb 10 kb 20 kb 10 kb
P300 ATF3 LXR

G H I
Mean expression Mean expression Arhgap22 Cd9 Itgax
(log2(TPM+1)) (log2(TPM+1))
*** * ***
12.5 80
10.0 43 significant 300
Mean Expression (TPM)

Gpnmb Mmp12 150 60

Clec4f-Cre+ − KN-RM

Itgax Trem2 100 40

200
(LXRα/β KC-DKO)

Arhgap22 10.0
Cd9 50 100
NASH − KC-N

7.5 20
0 0 0
7.5
146 significant

5.0 Kcnn4 Tgfb3 Trem2

5.0 *** ** n.s.
25 80
2.5 30 20
2.5 60
20 15
10 40
10 5 20
0.0 0.0 0 0 0
0.0 2.5 5.0 7.5 10.0 0.0 2.5 5.0 7.5 10.0 12.5 - + - + - +
re re re re re re
Control − KC-H Clec4f-Cre- − KN-RM f-C f-C f-C f-C f-C f-C
l e c4 lec4 l e c4 lec4 l e c4 lec4
(LXRα/β WT) C C C C C C

Figure 7. Combinatorial Actions of LXR and ATF3 Coordinate NASH Responsive Gene Expression in KCs during NASH
(A) Motif frequency within 500 bp of LXR binding sites described in Figure 6F.
(B) Normalized ATF3 ChIP-seq signal at merged IDR peaks from KC nuclei. Nuclei were sorted from Clec4f-Cre-NLS-TdTomato mice fed either a control diet or
the NASH diet for 20 weeks (n = 2). Significant ATF3 peaks had Poisson enrichment p value < 0.0001 and FC > 4.
(legend continued on next page)
Immunity 52, 1057–1074, June 16, 2020 1069
 ll
desmosterol, which is by far the most abundant endogenous
LXR ligand in the liver (Sakai et al., 2019; Yang et al., 2006),
was significantly increased in the setting of the cholesterol-
rich NASH diet in comparison to the control diet (Figure 6C).
All other natural LXR agonists were present at much lower con-
centrations (data not shown). The increase in desmosterol is
consistent with the upregulation of general LXR target genes
involved in cholesterol homeostasis. A corresponding increase
in desmosterol was reported in livers of human subjects with
NASH and cirrhosis (Gorden et al., 2015).
To investigate whether changes in gene expression are asso-
ciated with corresponding changes in LXR binding, we per-
formed ChIP-seq for LXRa + LXRb in the combination of KC-
N and KN-RM cells marked by sorting nuclei prepared from
rapidly fixed control or NASH livers for nuclear Clec4f-tdTo-
mato expression (Figure S7A). These studies yielded high qual-
ity datasets that could not be obtained using conventionally
sorted cells and indicated that the NASH diet induced LXR
binding by more than 2-fold at more than 1,700 locations and
resulted in a loss of binding at more than 1,000 locations (Fig-
ure 6D). Notably, DNA binding was relatively unchanged at
most of the LXR binding sites associated with canonical LXR
activity (e.g., Abca1), which are highly LXR-dependent but not
affected at the mRNA level by the NASH diet (Figure 6E). In
contrast, LXR binding in the vicinity of Timd4, Pcolce2, and
Plac8 was markedly reduced. Overall, LXR ChIP-seq peaks
were preferentially depleted from KC signature enhancers
(405/1046, or ~39% of total downregulated LXR peaks in en-
hancers, blue points, Fisher’s exact test p value < 0.001)
compared to LXR signal gained at KC signature enhancers
(128/1795, or ~7% of total upregulated LXR peaks in en-
hancers, red points, Fisher’s exact test p value = 1) (Figures
6D and S7B). Sites at which LXR binding was gained were
associated with increased H3K27ac, whereas the opposite
pattern was observed at sites at which LXR binding was lost,
consistent with a positive regulatory function of LXR at these lo-
cations (Figure 6F).
NASH-Induced Collaborative Binding Partners
Repurpose LXRs
Motif analysis of genomic regions exhibiting gain or loss of LXR
binding in response to the NASH diet indicated enrichment for
AP-1 or ATF, EGR, and MITF motifs at gained sites, and PU.1
or SPIC, GLIS3, and nuclear receptor half site motifs at lost sites
(Figure 6G). The distance relationships of these motifs to the cen-
ter of the LXR peak were consistent with collaborative binding in-
teractions observed for other LDTFs (Figure 7A; Heinz et al.,
2015). These findings suggested the possibility that NASH-
(C) De novo motif enrichment for NASH-specific ATF3 peaks as in (B).
(D) Intersection of ATF3 and/or LXR IDR ChIP-seq peaks from KC nuclei isolated from NASH-diet-fed mice with enhancers with increased activity during NASH
(from Figure 5B).
(E) Normalized distribution of ChIP-seq tag density at sites co-bound by LXRs and ATF3 indicated in (D).
(F) UCSC genome browser tracks of LXR, ATF3, and P300 ChIP-seq signals in the vicinities of the indicated genes. Peaks with NASH-gained (HOMER) binding for
both LXR and ATF3 are shaded yellow, and peaks with only gained ATF3 are shaded green.
(G) Mean expression (TPM) for genes nearest to NASH-specific LXR peaks co-bound by ATF3. DE genes identified using p-adj < 0.05 and FC > 2.
(H) RNA-seq (n = 2 per group) of KN-RM from control mice or Clec4fCre-LXRafl/flLXRbfl/fl mice fed the NASH diet for 20 weeks. DE genes identified as in (G).
(I) As in (H) for selected genes. **p-adj < 0.01; ***p-adj < 0.001.
Please see also Figure S7.
1070 Immunity 52, 1057–1074, June 16, 2020
Article
induced changes in the binding and function of LXRs were the
consequence of altered collaborative binding interactions with
AP-1 factors, EGRs, and members of the ETS TF family PU.1
and/or SpiC. To examine this possibility, we focused on ATF3,
as it was the most highly induced AP-1 factor in response to
the NASH diet (Figure 5E) and recognized the AP-1 motif en-
riched in NASH-specific enhancers (Figure 5B). Similar to
expression amounts of Atf3, we detected ~2.5 times as many
peaks in nuclei from mice with NASH (data not shown). This
observation correlated with an overall increased magnitude of
ATF3 ChIP-seq signal during NASH (Figures 7B and S7D). En-
hancers with gained ATF3 binding during NASH were enriched
for de novo motifs corresponding to AP-1, EGR, and MITF2
TFs (Figures 7C and S7C).
Intersection of the LXR and ATF3 ChIP-seq datasets indicated
~28,000 instances of overlapping peaks (Figure S7E). Of the
1,793 LXR binding sites increased by >2-fold in NASH, ~80%
overlapped with ATF3 peaks, consistent with a collaborative
binding relationship. Notably, the combination of LXR and
ATF3 was observed at 54% of NASH-induced enhancers (Fig-
ure 7D). The majority of these sites were occupied by LXRs under
control conditions, but exhibited substantial increases in
H3K27ac in the context of NASH (Figure 7E). This result sug-
gested that diet-induced binding of ATF3 to these sites resulted
in the subsequent recruitment of HATs. We therefore performed
ChIP-seq for the HAT P300 and found its recruitment to the LXR
and ATF3 peaks in response to the NASH diet, indicating a likely
role in contribution to acetylation of H3K27 and subsequent
enhancer function (Figures 7E and S7F–S7H). Representative
examples of NASH-induced co-binding of LXRs and ATF3 or
for NASH-induced binding of ATF3 at pre-existing LXR binding
sites are illustrated for Trem2, Cd9, Arhgap22, and Kcnn4 in
Figure 7F.
Analysis of the genes nearest to enhancers bound by ATF3
and LXR in the context of the NASH diet revealed a shift in
expression distribution toward a higher mean during NASH (Fig-
ure 7G). To directly evaluate NASH-specific functions of LXRs re-
sulting from these induced binding sites, we crossed
LXRafl/flLXRbfl/fl mice to the Clec4f-cre-TdTomato mouse to
generate KC-N and KN-RM-specific LXRa / LXRb / mice.
The progeny were then fed either a control diet or the NASH
diet for 20 weeks. Due to the near complete absence of KC-N
cells in the Clec4f-cre-TdTomato LXRafl/flLXRbfl/fl mice, compar-
isons were made of control and Clec4f-cre-TdTomato
LXRafl/flLXRbfl/fl KN-RMs. By comparing LXR-dependent gene
expression in KN-RMs under these two conditions, we identified
a set of 146 NASH-specific LXR-dependent genes (Figures 7H
and 7I), 33 of which were part of the set defined in Figure 7G.
 ll
Article

Notably, LXR dependency was observed both at genes
exhibiting NASH-induced binding of LXR (e.g., the SAM-associ-
ated gene Cd9) and genes exhibiting recruitment of ATF3 to pre-
existing LXR binding sites (e.g., Arhgap22 and Kcnn4). Trem2
expression was reduced in LXRa / LXRb / KN-RMs, but did
not retain statistical significance after correction for multiple
testing. Collectively, these findings provide evidence that
NASH-induced changes in collaborative TFs result in altered
binding and function of LXRs, contributing to the NASH-specific
program of KC gene expression.
DISCUSSION
populations provides a basis for understanding how disease-
promoting environmental signals instruct resident and recruited
macrophages to acquire distinct pathogenic programs of gene
expression. Application of the methods utilized in this study
should prove valuable in gaining a better understanding of
how distinct myeloid phenotypes are established in other dis-
ease contexts.
STAR+METHODS
Detailed methods are provided in the online version of this paper
and include the following:

Here, investigation of genomic and anatomic relationships d KEY RESOURCES TABLE

among phenotypically distinct macrophage populations in a d RESOURCE AVAILABILITY
mouse model of NASH revealed a strong association of niche B Lead Contact
occupancy with the transcriptional and phenotypic states of B Materials Availability
different EMP- and HSC-derived myeloid cells. HSC-derived B Data and Code Availability
macrophages observed in this model of NASH reside in at least d EXPERIMENTAL MODEL AND SUBJECT DETAILS
two distinct niches associated with divergent pathways of dif- B Mice
ferentiation. One pathway corresponds to the development of B NASH-Model Diets
Ly6Chi&lo-RMs, which reside proximal to large portal and cen- B Diphtheria toxin (DT)-mediated depletion of
tral vein vessels in contact with conventional (but most likely in- Kupffer cells
flamed) vascular endothelium. An alternative pathway results in d METHOD DETAILS
the accumulation of KN-RM cells, which colocalized with KC-N B Microscopy and Histo-Cytometry
in contact with the LSECs and adopted a similar pattern of B Nuclei Sorting
gene expression. This process occurred concordantly with pro- B Cell Sorting and Flow Cytometry
gressive loss of resident KC-Ns, most likely due to apoptosis. B Single-Cell RNA-Seq
These findings provide strong evidence for instructive roles of B ATAC-Seq
the LSECs and the NASH diet in determining the molecular B Poly A RNA-Seq
phenotypes of both KN-RM and KC-N cells. Of particular inter- B Chromatin Immunoprecipitation
est, we observed that a subset of KC-N and KN-RM cells ac- B ChIP-Seq Library Preparation
quire similarities to the recently identified SAM that resides in B Lipid Measurements
areas of fibrosis in human cirrhotic livers (Ramachandran d QUANTIFICATION AND STATISTICAL ANALYSIS
et al., 2019). B Sequencing Data Analysis
The transitions of open chromatin, H3K27ac and genome-
wide locations of LXRs and ATF3 provide evidence for SUPPLEMENTAL INFORMATION
NASH-induced alterations in expression and activities of
ATF3, resulting in reprogramming KC-specific functions of Supplemental Information can be found online at https://doi.org/10.1016/j.
LXRs. In the context of relatively constant LXR expression, immuni.2020.04.001.

increased ATF3 and likely other AP-1 TFs are suggested to

drive redistribution of LXR to NASH-specific enhancers as ACKNOWLEDGMENTS

well as activate poised enhancers bound by LXRs under control

conditions. The decreased expression of Spic in the context of
the NASH diet may also contribute to loss of collaborative LXR
binding at a subset of KC-specific enhancers. The delineation
of NASH-induced regulatory elements associated with the
Cd9 and Trem2 genes that are occupied by LXRs and ATF3
suggest roles of these TFs in establishing the SAM and/or
LAM phenotypes. A key remaining question is to determine
whether the altered functions of LXRs in KCs are pathogenic
or represent an adaptive response to the NASH diet.
Overall, these studies are consistent with a model in which
distinct microenvironments within the NASH liver drive diver-
gent patterns of differentiation of resident and infiltrating cells
by remodeling the open chromatin landscapes of recruited
monocytes and altering the activities of pre-existing enhancers
of the resident KC population. The inference of TFs and up-
stream signaling pathways associated with these distinct cell
These studies were supported by NIH grants DK091183, HL088083,
DK063491, and GM085764 and Fondation Leducq grant 16CVD01. J.S.S.
was supported by American Heart Association Fellowship 16PRE30980030
and NIH Predoctoral Training grant 5T32DK007541. T.D.T. was supported
by P30 DK063491, T32DK007044, and NRSA T32CA009523. M.S. was sup-
ported by the Manpei Suzuki Diabetes Foundation of Tokyo, Japan and the
Osamu Hayaishi Memorial Scholarship for Study Abroad, Japan. X.S. was
supported by the American Heart Association grant 18POST34060088.
This work was also supported in part by the Intramural Research program
of NIAID, NIH.
AUTHOR CONTRIBUTIONS
Conceptualization, J.S.S., T.D.T., M.S., A.G., R.N.G, and C.K.G.; Methodol-
ogy, J.S.S., T.D.T., M.S., A.G. and N.J.S.; Formal Analysis, J.S.S, T.D.T.,
A.G., Z.O., R.Z.L., and M.H.; Investigation, J.S.S., T.D.T., M.S., A.G.,
C.M.B., N.J.S., H.B., B.T.V., M.P.P., K.M.E., D.G., and L.-W.C.; Data Curation,
J.S.S., T.D.T, and Z.O; Writing – Original Draft, J.S.S., T.D.T, A.G., R.N.G, and
C.K.G.; Writing – Review & Editing, J.S.S, T.D.T., M.S., A.G., V.M.L., R.N.G.,

Immunity 52, 1057–1074, June 16, 2020 1071

 ll
and C.K.G.; Funding Acquisition, J.S.S., T.D.T., R.N.G, and C.K.G.; Visualiza-
tion, J.S.S., T.D.T., A.G., and Z.O.
DECLARATION OF INTERESTS
J.L.W. and X.S. are named inventors on patent applications or patents
related to the use of oxidation-specific antibodies that are held by UCSD.
J.L.W. is a consultant to Ionis Pharm and is a scientific founder of Oxi-
tope, Inc.
Received: February 3, 2020
Revised: April 1, 2020
Accepted: April 8, 2020
Published: May 1, 2020
REFERENCES
Baddeley, A., Rubak, E., and Turner, R. (2016). Spatial point patterns: method-
ology and applications with R (Boca Raton, London, New York: CRC Press,
Taylor & Francis Group).
Bain, C.C., Bravo-Blas, A., Scott, C.L., Perdiguero, E.G., Geissmann, F., Henri,
S., Malissen, B., Osborne, L.C., Artis, D., and Mowat, A.M. (2014). Constant
replenishment from circulating monocytes maintains the macrophage pool in
the intestine of adult mice. Nat. Immunol. 15, 929–937.
Bonnardel, J., T’Jonck, W., Gaublomme, D., Browaeys, R., Scott, C.L.,
Martens, L., Vanneste, B., De Prijck, S., Nedospasov, S.A., Kremer, A., et al.
(2019). Stellate Cells, Hepatocytes, and Endothelial Cells Imprint the Kupffer
Cell Identity on Monocytes Colonizing the Liver Macrophage Niche.
Immunity 51, 638–654.e9.
Buch, T., Heppner, F.L., Tertilt, C., Heinen, T.J., Kremer, M., Wunderlich, F.T.,
Jung, S., and Waisman, A. (2005). A Cre-inducible diphtheria toxin receptor
mediates cell lineage ablation after toxin administration. Nat. Methods 2,
419–426.
Buenrostro, J.D., Giresi, P.G., Zaba, L.C., Chang, H.Y., and Greenleaf, W.J.
(2013). Transposition of native chromatin for fast and sensitive epigenomic
profiling of open chromatin, DNA-binding proteins and nucleosome position.
Nat. Methods 10, 1213–1218.
Creyghton, M.P., Cheng, A.W., Welstead, G.G., Kooistra, T., Carey, B.W.,
Steine, E.J., Hanna, J., Lodato, M.A., Frampton, G.M., Sharp, P.A., et al.
(2010). Histone H3K27ac separates active from poised enhancers and
predicts developmental state. Proc. Natl. Acad. Sci. USA 107,
21931–21936.
De Schepper, S., Verheijden, S., Aguilera-Lizarraga, J., Viola, M.F., Boesmans,
W., Stakenborg, N., Voytyuk, I., Schmidt, I., Boeckx, B., Dierckx de Casterle, I.,
et al. (2018). Self-Maintaining Gut Macrophages Are Essential for Intestinal
Homeostasis. Cell 175, 400–415.e13.
Dick, S.A., Macklin, J.A., Nejat, S., Momen, A., Clemente-Casares, X.,
Althagafi, M.G., Chen, J., Kantores, C., Hosseinzadeh, S., Aronoff, L., et al.
(2019). Self-renewing resident cardiac macrophages limit adverse remodeling
following myocardial infarction. Nat. Immunol. 20, 29–39.
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, 15–21.
Eichenfield, D.Z., Troutman, T.D., Link, V.M., Lam, M.T., Cho, H., Gosselin, D.,
Spann, N.J., Lesch, H.P., Tao, J., Muto, J., et al. (2016). Tissue damage drives
co-localization of NF-kB, Smad3, and Nrf2 to direct Rev-erb sensitive wound
repair in mouse macrophages. eLife 5, 554–562.
Gautier, E.L., Shay, T., Miller, J., Greter, M., Jakubzick, C., Ivanov, S., Helft, J.,
Chow, A., Elpek, K.G., Gordonov, S., et al.; Immunological Genome
Consortium (2012). Gene-expression profiles and transcriptional regulatory
pathways that underlie the identity and diversity of mouse tissue macro-
phages. Nat. Immunol. 13, 1118–1128.
Gerner, M.Y., Kastenmuller, W., Ifrim, I., Kabat, J., and Germain, R.N. (2012).
Histo-cytometry: a method for highly multiplex quantitative tissue imaging
analysis applied to dendritic cell subset microanatomy in lymph nodes.
Immunity 37, 364–376.
1072 Immunity 52, 1057–1074, June 16, 2020
Article
Ginhoux, F., Greter, M., Leboeuf, M., Nandi, S., See, P., Gokhan, S., Mehler,
M.F., Conway, S.J., Ng, L.G., Stanley, E.R., et al. (2010). Fate mapping analysis
reveals that adult microglia derive from primitive macrophages. Science 330,
841–845.
Gomez Perdiguero, E., Klapproth, K., Schulz, C., Busch, K., Azzoni, E., Crozet,
L., Garner, H., Trouillet, C., de Bruijn, M.F., Geissmann, F., and Rodewald, H.R.
(2015). Tissue-resident macrophages originate from yolk-sac-derived erythro-
myeloid progenitors. Nature 518, 547–551.
Gorden, D.L., Myers, D.S., Ivanova, P.T., Fahy, E., Maurya, M.R., Gupta, S.,
Min, J., Spann, N.J., McDonald, J.G., Kelly, S.L., et al. (2015). Biomarkers of
NAFLD progression: a lipidomics approach to an epidemic. J. Lipid Res. 56,
722–736.
Gosselin, D., Link, V.M., Romanoski, C.E., Fonseca, G.J., Eichenfield, D.Z.,
Spann, N.J., Stender, J.D., Chun, H.B., Garner, H., Geissmann, F., and
Glass, C.K. (2014). Environment drives selection and function of
enhancers controlling tissue-specific macrophage identities. Cell 159,
1327–1340.
Gosselin, D., Skola, D., Coufal, N.G., Holtman, I.R., Schlachetzki, J.C.M., Sajti,
E., Jaeger, B.N., O’Connor, C., Fitzpatrick, C., Pasillas, M.P., et al. (2017). An
environment-dependent transcriptional network specifies human microglia
identity. Science 356, https://doi.org/10.1126/science.aal3222.
Haas, J.T., Francque, S., and Staels, B. (2016). Pathophysiology and
Mechanisms of Nonalcoholic Fatty Liver Disease. Annu. Rev. Physiol. 78,
181–205.
Hall, D., Poussin, C., Velagapudi, V.R., Empsen, C., Joffraud, M., Beckmann,
J.S., Geerts, A.E., Ravussin, Y., Ibberson, M., Oresic, M., and Thorens, B.
(2010). Peroxisomal and microsomal lipid pathways associated with resis-
tance to hepatic steatosis and reduced pro-inflammatory state. J. Biol.
Chem. 285, 31011–31023.
Halpern, K.B., Shenhav, R., Massalha, H., Toth, B., Egozi, A., Massasa, E.E.,
Medgalia, C., David, E., Giladi, A., Moor, A.E., et al. (2018). Paired-cell
sequencing enables spatial gene expression mapping of liver endothelial cells.
Nat. Biotechnol. 36, 962–970.
Hardy, T., Oakley, F., Anstee, Q.M., and Day, C.P. (2016). Nonalcoholic Fatty
Liver Disease: Pathogenesis and Disease Spectrum. Annu. Rev. Pathol. 11,
451–496.
Heinz, S., Benner, C., Spann, N., Bertolino, E., Lin, Y.C., Laslo, P., Cheng, J.X.,
Murre, C., Singh, H., and Glass, C.K. (2010). Simple combinations of lineage-
determining transcription factors prime cis-regulatory elements required for
macrophage and B cell identities. Mol. Cell 38, 576–589.
Heinz, S., Romanoski, C.E., Benner, C., and Glass, C.K. (2015). The selection
and function of cell type-specific enhancers. Nat. Rev. Mol. Cell Biol. 16,
144–154.
Heinz, S., Texari, L., Hayes, M.G.B., Urbanowski, M., Chang, M.W., Givarkes,
N., Rialdi, A., White, K.M., Albrecht, R.A., Pache, L., et al. (2018).
Transcription Elongation Can Affect Genome 3D Structure. Cell 174, 1522–
1536.e22.
Heymann, F., and Tacke, F. (2016). Immunology in the liver–from homeostasis
to disease. Nat. Rev. Gastroenterol. Hepatol. 13, 88–110.
Hill, D.A., Lim, H.W., Kim, Y.H., Ho, W.Y., Foong, Y.H., Nelson, V.L., Nguyen,
H.C.B., Chegireddy, K., Kim, J., Habertheuer, A., et al. (2018). Distinct macro-
phage populations direct inflammatory versus physiological changes in adi-
pose tissue. Proc. Natl. Acad. Sci. USA 115, E5096–E5105.
Holtman, I.R., Skola, D., and Glass, C.K. (2017). Transcriptional control of
microglia phenotypes in health and disease. J. Clin. Invest. 127,
3220–3229.
Jaitin, D.A., Adlung, L., Thaiss, C.A., Weiner, A., Li, B., Descamps, H.,
Lundgren, P., Bleriot, C., Liu, Z., Deczkowska, A., et al. (2019). Lipid-
Associated Macrophages Control Metabolic Homeostasis in a Trem2-
Dependent Manner. Cell 178, 686–698.e14.
Ju, C., and Tacke, F. (2016). Hepatic macrophages in homeostasis and liver
diseases: from pathogenesis to novel therapeutic strategies. Cell. Mol.
Immunol. 13, 316–327.

Article
Karlmark, K.R., Weiskirchen, R., Zimmermann, H.W., Gassler, N., Ginhoux, F.,
Weber, C., Merad, M., Luedde, T., Trautwein, C., and Tacke, F. (2009). Hepatic
recruitment of the inflammatory Gr1+ monocyte subset upon liver injury pro-
motes hepatic fibrosis. Hepatology 50, 261–274.
Kent, W.J., Sugnet, C.W., Furey, T.S., Roskin, K.M., Pringle, T.H., Zahler, A.M.,
and Haussler, D. (2002). The human genome browser at UCSC. Genome Res.
12, 996–1006.
Kiss, M., Van Gassen, S., Movahedi, K., Saeys, Y., and Laoui, D. (2018).
Myeloid cell heterogeneity in cancer: not a single cell alike. Cell. Immunol.
330, 188–201.
Kleiner, D.E., Brunt, E.M., Van Natta, M., Behling, C., Contos, M.J., Cummings,
O.W., Ferrell, L.D., Liu, Y.C., Torbenson, M.S., Unalp-Arida, A., et al.;
Nonalcoholic Steatohepatitis Clinical Research Network (2005). Design and
validation of a histological scoring system for nonalcoholic fatty liver disease.
Hepatology 41, 1313–1321.
Krenkel, O., Puengel, T., Govaere, O., Abdallah, A.T., Mossanen, J.C.,
Kohlhepp, M., Liepelt, A., Lefebvre, E., Luedde, T., Hellerbrand, C., et al.
(2018). Therapeutic inhibition of inflammatory monocyte recruitment reduces
steatohepatitis and liver fibrosis. Hepatology 67, 1270–1283.
Langmead, B., and Salzberg, S.L. (2012). Fast gapped-read alignment with
Bowtie 2. Nat. Methods 9, 357–359.
Lavin, Y., Winter, D., Blecher-Gonen, R., David, E., Keren-Shaul, H., Merad,
M., Jung, S., and Amit, I. (2014). Tissue-resident macrophage enhancer land-
scapes are shaped by the local microenvironment. Cell 159, 1312–1326.
Li, Q., Brown, J.B., Huang, H., and Bickel, J. (2011). Measuring reproducibility
of high-throughput experiments. The Annals of Applied Statistics 5,
1752–1779.
Link, V.M., Duttke, S.H., Chun, H.B., Holtman, I.R., Westin, E., Hoeksema,
M.A., Abe, Y., Skola, D., Romanoski, C.E., Tao, J., et al. (2018). Analysis of
Genetically Diverse Macrophages Reveals Local and Domain-wide
Mechanisms that Control Transcription Factor Binding and Function. Cell
173, 1796–1809.e17.
Love, M.I., Huber, W., and Anders, S. (2014). Moderated estimation of fold
change and dispersion for RNA-seq data with DESeq2. Genome Biol. 15, 550.
Machado, M.V., and Diehl, A.M. (2016). Pathogenesis of Nonalcoholic
Steatohepatitis. Gastroenterology 150, 1769–1777.
€nther, P., Crozet, L.,
Mass, E., Ballesteros, I., Farlik, M., Halbritter, F., Gu
Jacome-Galarza, C.E., Ha €ndler, K., Klughammer, J., Kobayashi, Y., et al.
(2016). Specification of tissue-resident macrophages during organogenesis.
Science 353, https://doi.org/10.1126/science.aaf4238.
Matsumoto, M., Hada, N., Sakamaki, Y., Uno, A., Shiga, T., Tanaka, C., Ito, T.,
Katsume, A., and Sudoh, M. (2013). An improved mouse model that rapidly de-
velops fibrosis in non-alcoholic steatohepatitis. Int. J. Exp. Pathol. 94, 93–103.
McGrath, K.E., Frame, J.M., Fegan, K.H., Bowen, J.R., Conway, S.J.,
Catherman, S.C., Kingsley, P.D., Koniski, A.D., and Palis, J. (2015). Distinct
Sources of Hematopoietic Progenitors Emerge before HSCs and Provide
Functional Blood Cells in the Mammalian Embryo. Cell Rep. 11, 1892–1904.
Mederacke, I., Dapito, D.H., Affò, S., Uchinami, H., and Schwabe, R.F. (2015).
High-yield and high-purity isolation of hepatic stellate cells from normal and
fibrotic mouse livers. Nat. Protoc. 10, 305–315.
Mitchell, C., Couton, D., Couty, J.-P., Anson, M., Crain, A.-M., Bizet, V., Rénia,
L., Pol, S., Mallet, V., and Gilgenkrantz, H. (2009). Dual role of CCR2 in the
constitution and the resolution of liver fibrosis in mice. Am. J. Pathol. 174,
1766–1775.
Muse, E.D., Yu, S., Edillor, C.R., Tao, J., Spann, N.J., Troutman, T.D.,
Seidman, J.S., Henke, A., Roland, J.T., Ozeki, K.A., et al. (2018). Cell-specific
discrimination of desmosterol and desmosterol mimetics confers selective
regulation of LXR and SREBP in macrophages. Proc. Natl. Acad. Sci. USA
115, E4680–E4689.
Musso, G., Cassader, M., and Gambino, R. (2016). Non-alcoholic steatohepa-
titis: emerging molecular targets and therapeutic strategies. Nat. Rev. Drug
Discov. 15, 249–274.
Oishi, Y., Spann, N.J., Link, V.M., Muse, E.D., Strid, T., Edillor, C., Kolar, M.J.,
Matsuzaka, T., Hayakawa, S., Tao, J., et al. (2017). SREBP1 Contributes to
ll
Resolution of Pro-inflammatory TLR4 Signaling by Reprogramming Fatty
Acid Metabolism. Cell Metab. 25, 412–427.
Peet, D.J., Janowski, B.A., and Mangelsdorf, D.J. (1998). The LXRs: a new
class of oxysterol receptors. Curr. Opin. Genet. Dev. 8, 571–575.
Perdiguero, E.G., and Geissmann, F. (2016). The development and mainte-
nance of resident macrophages. Nat. Immunol. 17, 2–8.
Pirzgalska, R.M., and Domingos, A.I. (2018). Macrophages in obesity. Cell.
Immunol. 330, 183–187.
Poussin, C., Ibberson, M., Hall, D., Ding, J., Soto, J., Abel, E.D., and Thorens,
B. (2011). Oxidative phosphorylation flexibility in the liver of mice resistant to
high-fat diet-induced hepatic steatosis. Diabetes 60, 2216–2224.
Ramachandran, P., Dobie, R., Wilson-Kanamori, J.R., Dora, E.F., Henderson,
B.E.P., Luu, N.T., Portman, J.R., Matchett, K.P., Brice, M., Marwick, J.A., et al.
(2019). Resolving the fibrotic niche of human liver cirrhosis at single-cell level.
Nature 575, 512–518.
Rinella, M.E., and Sanyal, A.J. (2016). Management of NAFLD: a stage-based
approach. Nat. Rev. Gastroenterol. Hepatol. 13, 196–205.
Sakai, M., Troutman, T.D., Seidman, J.S., Ouyang, Z., Spann, N.J., Abe, Y.,
Ego, K.M., Bruni, C.M., Deng, Z., Schlachetzki, J.C.M., et al. (2019). Liver-
Derived Signals Sequentially Reprogram Myeloid Enhancers to Initiate and
Maintain Kupffer Cell Identity. Immunity 51, 655–670.e8.
Satija, R., Farrell, J.A., Gennert, D., Schier, A.F., and Regev, A. (2015). Spatial
reconstruction of single-cell gene expression data. Nat. Biotechnol. 33,
495–502.
Schulz, C., Gomez Perdiguero, E., Chorro, L., Szabo-Rogers, H., Cagnard, N.,
Kierdorf, K., Prinz, M., Wu, B., Jacobsen, S.E.W., Pollard, J.W., et al. (2012). A
lineage of myeloid cells independent of Myb and hematopoietic stem cells.
Science 336, 86–90.
Scott, C.L., T’Jonck, W., Martens, L., Todorov, H., Sichien, D., Soen, B.,
Bonnardel, J., De Prijck, S., Vandamme, N., Cannoodt, R., et al. (2018). The
Transcription Factor ZEB2 Is Required to Maintain the Tissue-Specific
Identities of Macrophages. Immunity 49, 312–325.e5.
Scott, C.L., Zheng, F., De Baetselier, P., Martens, L., Saeys, Y., De Prijck, S.,
Lippens, S., Abels, C., Schoonooghe, S., Raes, G., et al. (2016). Bone marrow-
derived monocytes give rise to self-renewing and fully differentiated Kupffer
cells. Nat. Commun. 7, 10321.
Seki, E., De Minicis, S., Osterreicher, C.H., Kluwe, J., Osawa, Y., Brenner, D.A.,
and Schwabe, R.F. (2007). TLR4 enhances TGF-beta signaling and hepatic
fibrosis. Nat. Med. 13, 1324–1332.
Seki, E., De Minicis, S., Gwak, G.-Y., Kluwe, J., Inokuchi, S., Bursill, C.A.,
Llovet, J.M., Brenner, D.A., and Schwabe, R.F. (2009a). CCR1 and CCR5 pro-
mote hepatic fibrosis in mice. J. Clin. Invest. 119, 1858–1870.
Seki, E., de Minicis, S., Inokuchi, S., Taura, K., Miyai, K., van Rooijen, N.,
Schwabe, R.F., and Brenner, D.A. (2009b). CCR2 promotes hepatic fibrosis
in mice. Hepatology 50, 185–197.
Shaw, T.N., Houston, S.A., Wemyss, K., Bridgeman, H.M., Barbera, T.A.,
Zangerle-Murray, T., Strangward, P., Ridley, A.J.L., Wang, P.,
Tamoutounour, S., et al. (2018). Tissue-resident macrophages in the intestine
are long lived and defined by Tim-4 and CD4 expression. J. Exp. Med. 215,
1507–1518.
Shook, B.A., Wasko, R.R., Rivera-Gonzalez, G.C., Salazar-Gatzimas, E.,
López-Giráldez, F., Dash, B.C., Muñoz-Rojas, A.R., Aultman, K.D., Zwick,
R.K., Lei, V., et al. (2018). Myofibroblast proliferation and heterogeneity are
supported by macrophages during skin repair. Science 362, https://doi.org/
10.1126/science.aar2971.
Tacke, F., and Zimmermann, H.W. (2014). Macrophage heterogeneity in liver
injury and fibrosis. J. Hepatol. 60, 1090–1096.
Theurl, I., Hilgendorf, I., Nairz, M., Tymoszuk, P., Haschka, D., Asshoff, M., He,
S., Gerhardt, L.M.S., Holderried, T.A.W., Seifert, M., et al. (2016). On-demand
erythrocyte disposal and iron recycling requires transient macrophages in the
liver. Nat. Med. 22, 945–951.
Tripathi, S., Pohl, M.O., Zhou, Y., Rodriguez-Frandsen, A., Wang, G., Stein,
D.A., Moulton, H.M., DeJesus, P., Che, J., Mulder, L.C.F., et al. (2015).
Immunity 52, 1057–1074, June 16, 2020 1073
 ll
Meta- and Orthogonal Integration of Influenza ‘‘OMICs’’ Data Defines a Role
for UBR4 in Virus Budding. Cell Host Microbe 18, 723–735.
Xiong, X., Kuang, H., Ansari, S., Liu, T., Gong, J., Wang, S., Zhao, X.Y., Ji, Y., Li,
C., Guo, L., et al. (2019). Landscape of Intercellular Crosstalk in Healthy and
NASH Liver Revealed by Single-Cell Secretome Gene Analysis. Mol. Cell 75,
644–660.e5.
Yang, C., McDonald, J.G., Patel, A., Zhang, Y., Umetani, M., Xu, F., Westover,
E.J., Covey, D.F., Mangelsdorf, D.J., Cohen, J.C., and Hobbs, H.H. (2006).
Sterol intermediates from cholesterol biosynthetic pathway as liver X receptor
ligands. J. Biol. Chem. 281, 27816–27826.
Yona, S., Kim, K.W., Wolf, Y., Mildner, A., Varol, D., Breker, M., Strauss-Ayali,
D., Viukov, S., Guilliams, M., Misharin, A., et al. (2013). Fate mapping reveals
1074 Immunity 52, 1057–1074, June 16, 2020
Article
origins and dynamics of monocytes and tissue macrophages under homeo-
stasis. Immunity 38, 79–91.
Younossi, Z.M., Blissett, D., Blissett, R., Henry, L., Stepanova, M., Younossi,
Y., Racila, A., Hunt, S., and Beckerman, R. (2016). The economic and clinical
burden of nonalcoholic fatty liver disease in the United States and Europe.
Hepatology 64, 1577–1586.
Zigmond, E., Samia-Grinberg, S., Pasmanik-Chor, M., Brazowski, E., Shibolet,
O., Halpern, Z., and Varol, C. (2014). Infiltrating monocyte-derived macro-
phages and resident kupffer cells display different ontogeny and functions in
acute liver injury. J. Immunol. 193, 344–353.
Cell Metabolism

Article

NanoSIMS Analysis of Intravascular

Lipolysis and Lipid Movement
across Capillaries and into Cardiomyocytes
Cuiwen He,1 Thomas A. Weston,1 Rachel S. Jung,1 Patrick Heizer,1 Mikael Larsson,1 Xuchen Hu,1 Christopher M. Allan,1
Peter Tontonoz,3,4 Karen Reue,2 Anne P. Beigneux,1 Michael Ploug,5,6 Andrea Holme,7 Matthew Kilburn,7
Paul Guagliardo,7 David A. Ford,8 Loren G. Fong,1 Stephen G. Young,1,2,9,10,* and Haibo Jiang1,7,9,*
1Department of Medicine, University of California, Los Angeles, Los Angeles, CA 90095, USA
2Department of Human Genetics, University of California, Los Angeles, Los Angeles, CA 90095, USA
3Department of Pathology and Laboratory Medicine, University of California, Los Angeles, Los Angeles, CA 90095, USA
4Howard Hughes Medical Institute, University of California, Los Angeles, Los Angeles, CA 90095, USA
5Finsen Laboratory, Rigshospitalet, Copenhagen, Denmark
6Biotech Research and Innovation Centre, University of Copenhagen, Copenhagen, Denmark
7Centre for Microscopy, Characterisation and Analysis, University of Western Australia, Perth 6009, Australia
8Department of Biochemistry and Molecular Biology, Center for Cardiovascular Research, Saint Louis University, St. Louis, MO 63104, USA
9Senior author
10Lead Contact

*Correspondence: sgyoung@mednet.ucla.edu (S.G.Y.), haibo.jiang@uwa.edu.au (H.J.)

https://doi.org/10.1016/j.cmet.2018.03.017

SUMMARY et al., 2016). LPL is synthesized by parenchymal cells and

The processing of triglyceride-rich lipoproteins
(TRLs) in capillaries provides lipids for vital tissues,
but our understanding of TRL metabolism is limited,
in part because TRL processing and lipid movement
have never been visualized. To investigate the move-
ment of TRL-derived lipids in the heart, mice were
given an injection of [2H]triglyceride-enriched TRLs,
and the movement of 2H-labeled lipids across capil-
laries and into cardiomyocytes was examined by
NanoSIMS. TRL processing and lipid movement in
tissues were extremely rapid. Within 30 s, TRL-
derived lipids appeared in the subendothelial spaces
and in the lipid droplets and mitochondria of cardio-
myocytes. Enrichment of 2H in capillary endothelial
cells was not greater than in cardiomyocytes,
implying that endothelial cells may not be a control
point for lipid movement into cardiomyocytes.
Remarkably, a deficiency of the putative fatty acid
transport protein CD36, which is expressed highly
in capillary endothelial cells, did not impede entry
of TRL-derived lipids into cardiomyocytes.
INTRODUCTION
The lipolytic processing of triglyceride-rich lipoproteins (TRLs) by
lipoprotein lipase (LPL) is the central event in plasma lipid meta-
bolism, providing lipid nutrients for adipose tissue, heart, skeletal
muscle, and other tissues (Havel, 2010; Havel and Kane, 2001;
Kane and Havel, 2001). The processing of TRLs has been stud-
ied for decades (Havel and Gordon, 1960; Korn, 1955a, 1955b),
but only recently have key mechanisms fallen into place (Fong
secreted into the interstitial spaces, where it is captured by gly-
cosylphosphatidylinositol-anchored high-density lipoprotein-
binding protein 1 (GPIHBP1) on endothelial cells and transported
to the lumen of capillaries (Beigneux et al., 2007; Davies et al.,
2010). GPIHBP1 moves bidirectionally across endothelial cells
(Davies et al., 2012), with each excursion providing an opportu-
nity to pick up LPL and move it to the capillary lumen. In the
setting of GPIHBP1 deficiency, LPL is stranded within the inter-
stitial spaces (Davies et al., 2010), where it is useless for intravas-
cular TRL processing. GPIHBP1- and GPIHBP1-bound LPL are
located only on capillary endothelial cells and are absent from
endothelial cells of arterioles and venules (Beigneux et al.,
2007; Davies et al., 2010).
Aside from shuttling LPL to the capillary lumen, GPIHBP1
plays two additional functions in intravascular lipolysis. First,
the GPIHBP1-LPL complex is crucial for the margination of
TRLs along capillaries (Goulbourne et al., 2014), making it
possible for LPL-mediated TRL processing to proceed. Second,
the binding of LPL to GPIHBP1 stabilizes the structural integrity
of LPL’s hydrolase domain and preserves its enzymatic activity,
even in the presence of physiologic inhibitors (e.g., ANGPTL4),
and by doing so GPIHBP1 serves to focus LPL activity in capil-
laries (Fong et al., 2016; Mysling et al., 2016a, 2016b).
Although recent studies have clarified functions of GPIHBP1
and LPL in TRL processing, other topics in the realm of intravas-
cular lipolysis have remained enigmatic. In particular, no one
understands the mechanisms by which the fatty acid products
of TRL processing move across endothelial cells and into paren-
chymal cells (Hagberg et al., 2013). An early electron microscopy
study (Scow et al., 1976) suggested that the products of lipolysis
move across endothelial cells in ‘‘channels,’’ but the existence of
such channels is questionable and has never been substanti-
ated. In recent years, several groups have proposed that fatty
acid movement across endothelial cells is regulated and requires
transporter proteins (e.g., CD36 and fatty acid transporters)

Cell Metabolism 27, 1055–1066, May 1, 2018 ª 2018 Elsevier Inc. 1055
(Hagberg et al., 2013). The movement of fatty acids across cap-
illaries has been proposed to involve fatty acid entry into the
endothelial cell cytoplasm and formation of acyl-coenzyme A
(acyl-CoA) derivatives (Hagberg et al., 2013). Also, endothelial
cells have been proposed to be a ‘‘control point’’ for regulating
fatty acid movement, protecting parenchymal cells from exces-
sive lipid uptake (Hagberg et al., 2013). According to the latter
view, one could easily imagine that TRL-derived lipids might
accumulate in endothelial cells before moving to parenchymal
cells. On the other hand, others have proposed that fatty acids
simply diffuse freely across mammalian cells without assistance
from protein transporters (Guo et al., 2006; Hamilton et al., 2001;
Xu et al., 2013). Another enigma has been the itinerary of TRL-
derived fatty acids after entering parenchymal cells. It remains
controversial whether fatty acids can enter mitochondria directly
or whether they first must enter cytosolic triglyceride droplets.
Understanding how lipids move across capillaries and into
surrounding parenchymal cells represents an important issue
in lipoprotein physiology, but thus far the insights into this
process have been limited and have depended on cell culture
models or indirect evidence (e.g., drawing inferences from tran-
script levels for putative lipid-transport proteins or measuring
lipid stores within tissue extracts) (Bharadwaj et al., 2010;
Hagberg et al., 2010, 2013; Jang et al., 2016). We suspected
that it would be possible to gain additional insights into TRL pro-
cessing and lipid movement by imaging TRL processing within
capillaries.
In the current studies, we used nanoscale secondary ion mass
spectrometry (NanoSIMS) to visualize both TRL processing in
heart capillaries and the fate of TRL-derived lipids in adjacent
cardiomyocytes. NanoSIMS imaging uses a Cs+ beam to
bombard a tissue section, releasing secondary ions that can
be analyzed by mass spectrometry and used to create images
based solely on the isotopic content of the tissue (He et al.,
2017a, 2017b; Jiang et al., 2014a, 2014b). After giving mice an
injection of TRLs enriched in [2H]triglycerides, NanoSIMS imag-
ing makes it possible to visualize the 2H-labeled products of TRL
processing in endothelial cells, the subendothelial spaces, the
mitochondria, and/or cytosolic lipid droplets of cardiomyocytes.
In NanoSIMS imaging studies, all of the secondary ion data are
available for quantitative analyses. In the current studies, we
show that the movement of TRL-derived lipids across endothe-
lial cells and into cytosolic lipid droplets and mitochondria of
cardiomyocytes is extraordinarily rapid, occurring within sec-
onds. We found no evidence that capillary endothelial cells are
a substantial barrier to the movement of TRL-derived lipids
into cardiomyocytes. Also, we found no evidence that CD36
deficiency impedes the entry of TRL-derived lipids into
cardiomyocytes.
RESULTS
The Lipolytic Processing of TRLs and Lipid Movement
into Cardiomyocytes Are Rapid
2
H-Enriched TRLs (2H-TRLs) were harvested from the plasma of
Gpihbp1 / mice after administering 2H-labeled fatty acids by
gastric gavage. The 2H content of TRLs harvested from
Gpihbp1 / mice was quite high, as judged by the fact that their
density exceeded 1.006 g/mL (i.e., virtually all of the 2H-TRLs
1056 Cell Metabolism 27, 1055–1066, May 1, 2018
were on the bottom of the ultracentrifuge tube when spun at
d = 1.006 g/mL). Quantitative lipidomics studies confirmed that
large percentages of the fatty acyl chains in the 2H-TRLs were
deuterated (Table S1). Indeed, the majority of the linoleic acid
was deuterated. We observed consistent findings by NanoSIMS.
When the 2H-TRL particles were analyzed by NanoSIMS, the
2
H/1H ratio in the larger TRLs was 3,000–6,000 times the natu-
ral abundance of 2H, such that about one-third of all hydrogens in
the lipoprotein particles were 2H (Figure S1).
The 2H-TRLs (40 mg triglycerides in a volume of 200 mL) were
injected intravenously into wild-type mice. After 30 s, 2 min, or
30 min, the mice were euthanized, perfused extensively with
PBS, and then perfusion-fixed with carbodiimide/glutaralde-
hyde. (Carbodiimide is effective in fixing carboxylate-containing
small molecules in cells or tissues [Dallwig and Deitmer, 2002;
Takei et al., 2012; Tymianski et al., 1997], and our studies, in
which [3H]oleic acid was added to CHO-K1 cells, revealed
60%–75% greater [3H]oleic acid fixation with carbodiimide/
glutaraldehyde than with glutaraldehyde alone.) After fixation,
the left ventricular apex was embedded in epoxy resin,
sectioned, and analyzed by NanoSIMS. Images from 12C14N–
ions (a cluster secondary ion that reveals tissue nitrogen content)
were useful for visualizing tissue morphology; composite
12 14 –
C N and 2H/1H images identified regions of 2H enrichment
(Figure 1). In the composite images, the 2H/1H scale ranges
from 0.0002 to 0.01 (from slightly above 2H natural abundance
to 70 times natural abundance). Margination of 2H-TRLs along
the luminal surface of capillary endothelial cells was detectable
30 s and 2 min after the injection of 2H-TRLs (Figures 1A and
1B, white arrows), but ‘‘marginated TRLs’’ were not encountered
in capillaries at the 30-min time point. The 2H/1H ratio in the
marginated lipoproteins at 30 s and 2 min ranged from 15 to
450 times natural abundance, suggesting that processing had
removed some of the 2H-labeled triglycerides from the particles.
Enrichment of 2H within cardiomyocyte lipid droplets (40–50
times natural abundance) was observed as early as 30 s after
injecting 2H-TRLs (Figure 1, yellow arrows).
The margination of TRLs along capillaries is mediated by the
LPL-GPIHBP1 complex (Goulbourne et al., 2014); thus, there is
little doubt that the marginated TRLs in Figure 1 were undergoing
processing by LPL. Interestingly, we were unable to document
streaming of 2H-labeled lipids away from marginated TRLs and
into endothelial cells or cardiomyocytes (Figure 2). When we
quantified the 2H content of marginated TRLs and the 2H content
of immediately adjacent regions of endothelial cells and cardio-
myocytes, we observed a ‘‘fall-off’’ in 2H enrichment with
increasing distance from the TRL (Figure 2). One might have
easily jumped to the conclusion that the fall-off in 2H enrichment
represented streaming of fatty acids away from the TRL, but we
do not believe that this was the case, simply because we
observed a similar fall-off in 2H enrichment with increasing dis-
tance into the capillary lumen. In our tissue sections, the capillary
lumen is filled with epoxy resin, which is rich in 12C but devoid of
14
N. We suspect that the appearance of 2H enrichment in the
lumen of heart capillaries (which had been thoroughly perfused
before fixation) was the result of short-distance diffusion of tri-
glycerides into the resin-filled capillary lumen during tissue pro-
cessing. Because of this observation, we do not believe that we
observed bona fide streaming of fatty acids away from TRLs and

into endothelial cells or cardiomyocytes. We suspect that the
streaming of fatty acids away from TRLs, which obviously must
occur physiologically, is so rapid that it was not detectable by
NanoSIMS at the 30-s or 2-min time points. In support of that
reasoning, we never observed higher levels of 2H enrichment in
segments of cardiomyocytes that were immediately adjacent
to capillaries (Figure 1), implying that movement of 2H-labeled
lipids into and across cardiomyocytes is quite rapid.
The Entry of 2H-TRL-Derived Lipids into Cardiomyocytes
Depends on Intravascular Processing of 2H-TRLs by LPL
To be confident that the rapid appearance of 2H in cytosolic lipid
droplets of cardiomyocytes (Figure 1) was secondary to LPL-
mediated 2H-TRL processing, we performed experiments in
isolated, perfused mouse hearts, which allowed us to compare
2
H entry into cardiomyocytes in the presence or absence of an
LPL inhibitor. We also examined 2H uptake into hearts of
Gpihbp1 / mice (where LPL is absent from the luminal surface
of capillaries) (Davies et al., 2010, 2012). In perfused hearts of
wild-type mice treated with a nonspecific lipase inhibitor (tetra-
hydrolipistatin), the appearance of 2H into cardiomyocytes was
lower than in the absence of the inhibitor (Figures 3A and 3B).
Enrichment of 2H in cardiomyocytes was also lower in the setting
of Gpihbp1 deficiency (Figures 3A and 3B). In wild-type hearts
that had been treated with the LPL inhibitor, numerous
2
H-TRLs were found along endothelial cells, reflecting an accu-
mulation of marginated but unprocessed 2H-TRLs in capillaries
(Figure 3A). As expected, 2H enrichment was lower in endothelial
cells of Gpihbp1 / hearts (Figure 3), where TRL margination
does not occur (Goulbourne et al., 2014). When we measured
the 2H/1H ratio in endothelial cells of wild-type mice (focusing
exclusively on segments devoid of marginated TRLs) with the
2
H/1H ratio in endothelial cells of Gpihbp1 / mice, we found
a lower 2H/1H ratio in Gpihbp1 / endothelial cells, reflecting
Figure 1. NanoSIMS Images of TRLs along
Heart Capillary Endothelial Cells
Four-month-old wild-type female mice were fas-
ted for 4 hr and then injected with 200 mL 2H-TRLs
containing 40 mg triglycerides (produced in
Gpihbp1 / mice after administering 2H-labeled
fatty acids by gavage). After 30 s, 2 min, or 30 min,
mice were euthanized and perfusion-fixed with
carbodiimide/glutaraldehyde; sections were pre-
pared for NanoSIMS imaging. Three mice were
used in this experiment, one for each time point.
(A) 12C14N– NanoSIMS images were generated to
visualize tissue morphology; composite 12C14N–
(blue) and 2H/1H ratio (red) images revealed
2
H-TRLs that had marginated along capillary
endothelial cells (white arrows) and 2H-enriched
cytosolic lipid droplets in cardiomyocytes (yellow
arrows). The scale shows the 2H/1H ratio multiplied
by 10,000. The 2H natural abundance, relative to
1
H, is 0.000156. Scale bars, 6 mm.
(B) The boxed areas in (A) images are shown at
higher magnification to reveal capillary endothelial
cells. Composite 12C14N– (gray) and 2H/1H ratio
(red) images show marginated lipoproteins (ar-
rows) in the capillary lumen and 2H-enriched
cytosolic lipid droplets in cardiomyocytes. Scale
bars, 2 mm. See also Figure S1.
reduced TRL processing (Figure 3B). In both wild-type and
Gpihbp1 / mice, the 2H/1H ratio in endothelial cells was similar
to the 2H/1H ratio in surrounding cardiomyocytes (Figure 3B).
Investigating the Fate of TRL-Derived Lipids within
Cardiomyocytes
The mitochondria of cardiomyocytes could be identified in back-
scattered electron (BSE) images and in 12C14N–, 16O–, and 31P–
NanoSIMS images (Figure 4A). Visualizing mitochondria with
12 14 –
C N images was particularly helpful because the NanoSIMS
50L instrument is capable of recording 12C14N–, 2H–, and 1H– ions
simultaneously. In Figure 4B, we show high-resolution 12C14N–
and 2H/1H NanoSIMS images of the heart after injecting
2
H-TRLs. The 2H/1H scale in these images differs from those in
Figure 1, ranging from slightly above 2H natural abundance to
7 times natural abundance. When the mice were killed 30 s
or 2 min after the injection of 2H-TRLs, marginated TRLs could
be identified in the capillary lumen, and there was 2H enrichment
in both mitochondria and cytosolic lipid droplets of cardiomyo-
cytes (Figure 4C). At the 30-min time point, we did not find
marginated TRLs in capillaries; 2H enrichment persisted in cardi-
omyocyte mitochondria and lipid droplets, but the amount of 2H
enrichment in mitochondria was lower in the 30-min images than
in the 30-s or 2-min images (Figures 4C and 4D), almost certainly
reflecting mitochondrial catabolism of 2H-labeled fatty acids in
the 30-min time span after the injection of 2H-TRLs. We found
no evidence, at any of the time points, that 2H enrichment in
endothelial cells was greater than in adjacent cardiomyocytes
(Figures 1, 2, and 3; data not shown), implying that TRL-derived
lipids do not accumulate in endothelial cells before moving into
parenchymal cells.
In some experiments, 13C-labeled fatty acids were given to
mice by gastric gavage for several days before administering
the intravenous injection of 2H-TRLs. In these studies, we once
Cell Metabolism 27, 1055–1066, May 1, 2018 1057
Figure 2. Movement of Lipids from 2H-TRLs to Surrounding Cardiomyocytes
Four-month-old female wild-type mice were fasted for 4 hr and then given an intravenous injection of 200 mL of 2H-TRLs containing 40 mg triglycerides. After 30 s
or 2 min, mice were euthanized and perfusion-fixed with carbodiimide/glutaraldehyde; tissue sections were prepared for NanoSIMS. Images from three regions
(I, II, III) of the left ventricular apex were recorded. Two mice were used in this experiment.
(A) 12C14N– NanoSIMS images to show tissue morphology. Section I was prepared from tissue that had been stained with imidazole.
(B and C) Composite 12C14N–, blue in (B) and gray in (C), and 2H/1H ratio (red) images were generated to visualize 2H-enriched TRLs in the capillary lumen and
2
H-enriched lipid droplets within cardiomyocytes. The boxed area of the composite image is shown at higher magnification in (C). 12C14N– ions and the 2H/1H ratio
were measured in a TRL particle (purple circle) along the luminal surface of a capillary endothelial cell and in regions (white circles) extending into the endothelial
cell (e-1, e-2), an adjacent cardiomyocyte (c), or into the capillary lumen (l). Scale bars, 2 mm.
(D) Graph showing a progressive decrease in the 2H/1H ratio—starting with the TRL particle in the capillary lumen and extending into the endothelial cell, the
adjacent cardiomyocyte, and the capillary lumen.
(E) Graph showing 12C14N– ions in the TRL particle and in regions extending into the endothelial cell, the cardiomyocyte, and the capillary lumen.

again observed 2H-TRLs margination in endothelial cells and 2H
enrichment in cardiomyocyte lipid droplets 30 s after injecting
2
H-TRLs (Figure 5A). 2H-Labeled lipids invariably entered cyto-
solic lipid droplets that were already labeled with 13C (Figure 5A),
and the levels of 2H and 13C enrichment in lipid droplets were
positively correlated (p < 0.001) (Figure 5B). Of note, however,
the percentages of the labels in cytosolic lipid droplets were
different. Approximately 50% of cardiomyocyte 2H was located
in cytosolic lipid droplets, whereas only 5% of the 13C was in
lipid droplets (Figures 5C and 5D). This difference was not partic-
ularly surprising because 13C-labeled fatty acids (administered a
day earlier) would be expected to contribute to membrane phos-
pholipids. Also, fatty acid metabolites are used for synthesis of
nonessential amino acids (Sidossis et al., 1995). Indeed, the
fact that metabolites from 13C-labeled fatty acids contribute to
protein synthesis was illustrated by NanoSIMS images of brown
adipose tissue. As expected, the cytosolic lipid droplets in brown
adipocytes were enriched in 13C but devoid of nitrogen (i.e., as
judged by the 12C14N– images) (Figure S2). Interestingly, the he-
moglobin in one erythrocyte (almost certainly a newly released
erythrocyte) was enriched in 13C, but 13C enrichment was negli-
gible in older blood cells (two leukocytes and one erythrocyte)
(Figure S2). The nitrogen-rich cytoplasm of brown adipocytes
was also enriched in 13C, but to a lesser degree than cytosolic
lipid droplets (Figure S2).
The adult mouse brain primarily uses glucose for fuel, and TRL
margination is absent in the capillaries of the brain (Goulbourne
et al., 2014). For those reasons, we were skeptical that we would
detect significant amounts of 13C or 2H enrichment in the brain af-
ter administering 13C-labeled fatty acids or 2H-TRLs. Indeed, after
administering 13C-labeled fatty acids by gastric gavage and sub-
sequently administering 2H-TRLs intravenously, we were unable
to detect 13C or 2H enrichment in the cerebral cortex (Figure S3A).
NanoSIMS instruments have extraordinarily accurate mass
resolution, making it nearly impossible to mistake one isotope
for another. However, to be confident that we identified 2H–
and 13C– ions correctly, we created NanoSIMS images from
the heart of a mouse that had not been given 13C-labeled fatty
acids or 2H-TRLs. We had no difficulty in generating 12C14N–
NanoSIMS images of heart tissue, but as expected there was
no enrichment in 13C or 2H (Figure S3B).
Detecting 2H-Labeled Lipids in the Subendothelial
Spaces
The TRL-derived lipids that enter cardiomyocytes obviously
must have traversed the subendothelial spaces around

1058 Cell Metabolism 27, 1055–1066, May 1, 2018

Figure 3. The 2H-Enriched Lipids that Enter Cardiomyocytes Are Derived from LPL-Mediated Intravascular Lipolysis, as Judged by Studies in
Isolated, Perfused Hearts
Hearts from 4-month-old female wild-type or Gpihbp1 / mice were isolated and cannulated. Beating hearts were perfused with Tyrode’s buffer containing CaCl2
and either an LPL inhibitor (1 mM tetrahydrolipistatin) or vehicle (DMSO) alone. Next, hearts were perfused with 2H-TRLs. After 5 min, the hearts were perfused
with PBS (to remove nonmarginated TRLs) and then perfusion-fixed with carbodiimide/glutaraldehyde. Sections of hearts were prepared for NanoSIMS. Three
mice were used in this experiment, one for each time point.
(A) 12C14N– images were created to visualize morphology (left panels); composite 12C14N– (blue) and 2H/1H ratio (red) images (right panels) were used
to visualize 2H-TRLs in capillaries and 2H-enriched lipid droplets in cardiomyocytes. The scale shows the 2H/1H ratio multiplied by 10,000. Scale
bars, 2 mm.
(B) Bar graphs show 2H/1H ratios in cardiomyocytes, the sum of 2H/1H ratios in cytosolic lipid droplets, 2H/1H ratios in capillary endothelial cells, 2H/1H ratios in
regions of cardiomyocytes devoid of lipid droplets (LD), and 2H/1H ratios in regions of endothelial cells devoid of marginated lipoproteins (LP). In each graph, the y
axis starts at 0.00015 (natural abundance of 2H). *p < 0.05; **p < 0.01; ***p < 0.0005. Each graph represent quantification of four 30 3 30 mm images. Data are
presented as means ± SD and were analyzed with one-way ANOVA with multiple comparisons.

capillaries. After administering an intravenous injection of TRLs
and then staining heart tissue with OsO4 and imidazole, we
observed irregular, darkly stained structures in the subendo-
thelial spaces around capillaries. (Imidazole is a nitrogen-rich
mordant that increases OsO4 binding to lipids [Angermu €ller
and Fahimi, 1982].) The irregular, darkly stained structures
were frequent at the 30-s and 2-min time points but were
uncommon at the 30-min time point (Figure 6A). To determine
if the darkly stained subendothelial structures contained TRL-
derived lipids, we performed correlative BSE/NanoSIMS imag-
ing on OsO4/imidazole-stained heart tissue harvested 30 s after
an intravenous injection of 2H-TRLs. BSE images revealed
irregular, darkly stained structures in the subendothelial
spaces (Figures 6B and S4A); NanoSIMS imaging of the
same sections revealed that these structures were enriched
in 2H (Figures 6B and S4A) from 2H-TRL processing. Because
these structures were stained with imidazole, they were en-
riched in nitrogen, as judged by 12C14N– images (Figures 6B
and S4A).
Staining of lipids by imidazole was also evident in NanoSIMS
images of cardiomyocytes from mice that had been given
13
C-labeled fatty acids by gastric gavage (Figure S4B). Normally,
13
C-enriched cytosolic lipid droplets in cardiomyocytes are
devoid of nitrogen and appear ‘‘black’’ in a 12C14N– NanoSIMS
image. When the same tissue was stained with imidazole, the
cytosolic lipid droplets became nitrogen rich and were ‘‘white’’
in a 12C14N– image (Figure S4B).
The irregular OsO4/imidazole-stained structures were less
common around larger blood vessels, where GPIHBP1 and
LPL are absent (Davies et al., 2010; Fong et al., 2016). In the
electron micrograph shown in Figure 6C, many irregular, darkly
stained structures were found in and around a capillary, but
not in the vicinity of an arteriole (Figures 6C and S4C). By
NanoSIMS, we could not detect 2H enrichment in the subendo-
thelial spaces surrounding arterioles, even though 2H-labeled
lipids from TRL processing were present in adjacent cardio-
myocytes and in the smooth muscle cells of nearby arterioles
(Figures 6D).

Cell Metabolism 27, 1055–1066, May 1, 2018 1059

Figure 4. NanoSIMS Imaging to Visualize Lipids Released from 2H-Enriched TRLs
Three-month-old female wild-type mice were fasted for 4 hr and then given an intravenous injection of 200 mL 2H-TRLs (40 mg triglycerides). After 30 s, 2 min, or
30 min, the mice were euthanized and perfusion-fixed with carbodiimide/glutaraldehyde. Tissue sections were prepared for NanoSIMS and backscattered
electron (BSE) imaging. Three mice were used in this experiment, one for each time point.
(A) NanoSIMS images (12C14N–, 16O–, 31P–) from tissue not stained with imidazole. A BSE image of the same region was useful for morphology. Several mito-
chondria are outlined in red.
(B) NanoSIMS images of heart sections 30 s, 2 min, or 30 min after the injection of 2H-TRLs. 12C14N– images to visualize cell and tissue morphology (top panels);
2
H/1H ratio images (bottom panels) to visualize the fate of lipids released from 2H-TRLs. LP, lipoproteins; L, capillary lumen; M, mitochondria; LD, lipid droplets.
The scale shows the 2H/1H ratio multiplied by 10,000.
(C) Bar graph showing the 2H/1H ratio in endothelial cells, mitochondria, myocytes, regions of myocytes lacking lipid droplets (LD), and lipid droplets at different
time points. Three 30 3 30 mm images (each containing a minimum of two to three cardiomyocytes) were analyzed at each time point (endothelial cells, n = 8–13;
mitochondria, n = 24–73; cardiomyocytes, n = 7–18; regions of myocytes lacking lipid droplets, n = 40; lipid droplets, n = 40–90). Data are presented as means ±
SD Differences were assessed with one-way ANOVA. *p < 0.05; **p < 0.01; ns, nonsignificant.
(D) Bar graph depicting 2H/1H in lipid droplets divided by 2H/1H in mitochondria at each time point (mean ± SD). Each graph represent data from two 10 3 10 mm
images and three 30 3 30 mm images. Data are presented as means ± SD and were analyzed with one-way ANOVA with multiple comparisons. **p < 0.01. See
also Figure S2.

Assessing the Impact of CD36 Deficiency on the Entry of
TRL-Derived Lipids into Cardiomyocytes
NanoSIMS imaging opens the door to examining the functional
relevance of specific proteins in the movement of TRL-derived
lipids into and across capillaries. CD36, a putative fatty acid
transporter, is expressed at high levels in heart capillary endo-
thelial cells (Greenwalt et al., 1995) (Figure S5) and has been
hypothesized to play a role in the movement of fatty acids into
cardiomyocytes (Bharadwaj et al., 2010; Hagberg et al., 2013).
In the current study, we focused on testing whether a deficiency
of CD36 might impede the movement of 2H-TRL-derived lipids
into cardiomyocytes. Wild-type and Cd36 / mice were injected
intravenously with equivalent amounts of 2H-TRLs, and heart
sections were prepared for NanoSIMS imaging. At 1.5- and
2-min time points, 2H enrichment in cardiomyocytes was equiv-
alent in wild-type and Cd36 / mouse hearts (Figures 7A and
7B), but the amount of 2H in cytosolic lipid droplets was lower
in Cd36 / hearts (as a percentage of total 2H in cells) (Figures
7A and 7C). When we administered 13C-labeled fatty acids to
wild-type and Cd36 / mice by gastric gavage, we observed
consistent findings—similar amounts of 13C in wild-type and
Cd36 / cardiomyocytes (Figures 7D and 7E), but reduced
amounts of 13C in cytosolic lipid droplets of Cd36 / cardiomyo-
cytes (as a percentage of total 13C in cells) (Figures 7D and 7F). In
addition to reduced amounts of total 13C in lipid droplets, the
13
C/12C ratio in lipid droplets was lower in Cd36 / cardiomyo-
cytes (Figure 7G). Interestingly, the 13C/12C ratio was slightly
higher in Cd36 / mitochondria (Figure 7G). In another series
of NanoSIMS studies, we once again found similar amounts of
13
C enrichment in wild-type and Cd36 / cardiomyocytes (Fig-
ures S6A and S6B), but the percentage of total 13C in lipid drop-
lets was lower in Cd36 / cardiomyocytes (Figure S6C). CD36
deficiency has been reported to be associated with lower levels
of acyl-CoA in cells (Bharadwaj et al., 2010), but whether that
finding is due to the reduced amounts of 2H in cytosolic lipid
droplets of Cd36 / mice is unknown. We considered the

1060 Cell Metabolism 27, 1055–1066, May 1, 2018

Figure 5. NanoSIMS Images of Heart Tissue from Mice that Had Been Given 13C-Labeled Fatty Acids by Gastric Gavage and on the Following
Day an Intravenous Injection of 2H-TRLs
13
C-Labeled fatty acids were administered to 3- to 4-month-old female wild-type mice by gastric gavage (two 80-mg doses 12 hr apart). At 22 hr after the second
dose, mice were fasted for 4 hr and then injected with 200 mL 2H-TRLs containing 40 mg triglycerides. After periods of 30 s or 2 min post-injection, the mice were
euthanized and perfusion-fixed with carbodiimide/glutaraldehyde; sections of heart were prepared for NanoSIMS imaging. Two mice were used in this exper-
iment, one for each time point.
(A) NanoSIMS images of heart tissues 30 s after the injection of 2H-TRLs. 12C14N– images were useful for morphology; 2H/1H ratio images revealed marginated
2
H-TRLs in capillaries (white arrows) and 2H-enriched lipids in lipid droplets of cardiomyocytes (boxed regions). 13C/12C ratio images show 13C distribution. 2H/1H
and 13C/12C ratio scales are multiplied by 10,000. Scale bars, 2 mm.
(B) Scatterplots comparing 2H/1H and 13C/12C ratios in the center of cardiomyocyte lipid droplets. Data were obtained from lipid droplets (n = 50–60) in two
NanoSIMS images (31 3 31 mm).
(C) NanoSIMS images showing 2H enrichment in marginated lipoproteins in capillaries (white arrows) and cardiomyocyte lipid droplets (several indicated by
yellow arrows). 12C14N– images show tissue morphology, 2H/1H ratio images show the distribution of 2H 30 s or 2 min after the injection of 2H-TRLs, and 13C/12C
ratio images show 13C distribution 1 day after giving 13C-labeled fatty acids by gastric gavage. The same lipid droplets were identified in both 2H/1H and 13C/12C
images. 2H/1H and 13C/12C ratio scales are multiplied by 10,000. Scale bars, 5 mm.
(D) Plot depicting the percentage of cellular 2H and 13C in cytosolic lipid droplets, relative to 1H and 12C, respectively (two 35 3 35 mm NanoSIMS images analyzed
at each time point). See also Figure S3.

possibility that the lipid droplets in Cd36 / cardiomyocytes DISCUSSION

were simply too small to be visualized by NanoSIMS (where
the lateral resolution is 40–50 nm). By electron microscopy,
the average area of cytosolic lipid droplets in Cd36 / cardio-
myocytes was lower than in wild-type cardiomyocytes (Fig-
ure S7), but this difference was small, probably too small to
explain the paucity of lipid droplets in the NanoSIMS images of
Cd36 / cardiomyocytes.
We used NanoSIMS imaging along with electron microscopy to
visualize intravascular TRL processing and the fate of TRL-
derived lipids in cardiomyocytes. We gleaned three insights
from our studies. First, TRL processing and the uptake of TRL
lipids by cardiomyocytes are extremely rapid—occurring within
seconds after TRLs enter the bloodstream. Second, during

Cell Metabolism 27, 1055–1066, May 1, 2018 1061

Figure 6. The Lipid Products of TRL Processing Can Be Identified in the Subendothelial Spaces
(A) Transmission electron micrographs of the left ventricular apex from 4-month-old female wild-type mice 30 s, 2 min, or 30 min after an intravenous injection of
200 mL TRLs. Irregular, darkly stained structures were observed in the subendothelial spaces (red arrows) in OsO4/imidazole-stained tissue; these structures were
more common at the 30-s and 2-min time points. Scale bars, 1 mm. 2H/1H ratio scales were multiplied by 10,000. Three mice were used in this experiment, one for
each time point.
(B) Visualizing irregular, darkly stained structures in the subendothelial spaces by NanoSIMS and BSE imaging. A 4-month-old female wild-type mouse was
fasted for 4 hr, injected with 200 mL 2H-TRLs, and killed 30 s later. Heart tissue was stained with OsO4/imidazole and processed for NanoSIMS and BSE imaging.
The irregular, darkly stained subendothelial structures (white arrows) in the BSE images (left panel) were enriched in nitrogen (12C14N– ions, middle panels),
reflecting the high nitrogen content of imidazole, and were enriched in 2H, as judged by composite 12C14N– (blue) and 2H/1H ratio (red) images (right panels).
(C) The irregular, darkly stained subendothelial structures were found in the subendothelial spaces around capillaries (a), but not in the vicinity of a large blood
vessel (b and c).
(D) NanoSIMS images of an arteriole and an adjacent cardiomyocyte 2 min after an injection of 2H-TRLs. Enrichment of 2H was observed in mitochondria and
cytosolic lipid droplets of cardiomyocytes and in the mitochondria of arteriolar smooth muscle cells, but 2H enrichment was not observed in the interstitial spaces
around the arteriole. See also Figure S4.

active TRL processing, we observed no evidence that 2H-TRL-
derived lipids accumulate to higher levels in capillary endothelial
cells than in adjacent cardiomyocytes. Third, we found no evi-
dence that CD36 deficiency substantially impedes entry of
TRL-derived lipids into cardiomyocytes.
For decades, the lipid metabolism field has recognized
that the half-life of TRLs is measured in minutes (Havel and
Kane, 2001), but we were nevertheless quite surprised by how
rapidly TRL-derived lipids traverse capillary endothelial cells
and enter cardiomyocytes. NanoSIMS imaging showed that
2
H-TRL-derived lipids appear in cardiomyocytes within seconds
(preferentially in cytosolic lipid droplets but also in mitochondria).
Other features of the NanoSIMS images were consistent with
extremely rapid movement of fatty acids into and across tissues.
For example, we never observed more 2H enrichment in capillary
endothelial cells than in cardiomyocytes, nor did we observe
greater 2H enrichment in regions of cardiomyocytes adjacent
to capillaries. Also, 2H rapidly entered arteriolar smooth muscle
cells, despite the fact that TRL processing in arterioles is almost
certainly negligible, owing to an absence of GPIHBP1 and LPL
(Davies et al., 2010). All of these observations appear to be
consistent with the proposal that fatty acids diffuse rapidly
across cells (Guo et al., 2006; Hamilton et al., 2001; Xu et al.,
2013) or, alternatively, that there is a yet-to-be-identified abun-
dant high-capacity transport system for fatty acid movement
across endothelial cells. Before reaching cardiomyocytes, the
products of lipolysis obviously must traverse subendothelial
spaces. In imidazole-stained heart tissue, we observed darkly
staining structures in the subendothelial spaces, and those
structures contained 2H-labeled lipids from 2H-TRL processing.
An attractive feature of NanoSIMS analysis is that the second-
ary ion data are available for quantitative analyses, pixel by pixel,
with >260,000 pixels/image. In our studies, quantitative analyses
revealed that an LPL inhibitor drug or a deficiency of GPIHBP1 is

1062 Cell Metabolism 27, 1055–1066, May 1, 2018

Figure 7. TRL-Derived Lipids Move across Endothelial Cells and Enter Cardiomyocytes in Both Wild-Type and Cd36-Deficient Mice
NanoSIMS analyses of 2H– ions in the heart after administering an intravenous injection of 200 mL 2H-TRLs to 4-month-old female wild-type (Cd36+/+) and Cd36 /
mice. At 1.5 or 2 min after the injection, hearts were harvested and sections were prepared for NanoSIMS. Differences in quantitative NanoSIMS data were
analyzed with a Student’s t test with Welch’s correction. Four mice were used in this experiment, one for each time point.
(A) 12C14N– images (left) were used to define tissue morphology; the 2H/1H ratio images (middle) and the 2H/1H hue saturation images (right) show sites of
2
H enrichment. White arrows indicate examples of cytosolic lipid droplets. Ratio scales were multiplied by 10,000. Scale bars, 10 mm.
(B) 2H/1H ratios in cardiomyocytes of Cd36+/+ and Cd36 / mice. Each data point represents one cardiomyocyte in two or three 30 3 30 mm NanoSIMS images.
*p < 0.05; ns, nonsignificant.
(C) 2H in cytosolic lipid droplets as a percentage of total 2H in cardiomyocytes. Graphs were generated by quantifying two or three 30 3 30 mm NanoSIMS images
for each time point. NanoSIMS analyses of 13C– ions in the heart after feeding 13C-labeled fatty acids by gavage (three 80-mg doses 12 hr apart). Twenty-four
hours after the last dose, mice were fasted for 4 hr, and hearts were harvested and prepared for NanoSIMS. *p < 0.05; ns, nonsignificant.
(D) 12C14N– NanoSIMS images were used to visualize morphology; 31P images were used to visualize mitochondria; 13C/12C ratio images show sites of
13
C enrichment. Arrows indicate examples of cytosolic lipid droplets. Ratio scales were multiplied by 10,000. Scale bars, 10 mm.
(legend continued on next page)
Cell Metabolism 27, 1055–1066, May 1, 2018 1063
accompanied by reduced uptake of 2H-TRL-derived lipids into
cardiomyocytes. Quantitative analyses also allowed us to quan-
tify the distribution of 2H-TRL-derived lipids within cardiomyo-
cytes. Approximately 50% of the 2H in cardiomyocytes was
located in cytosolic lipid droplets 30 s or 2 min after an injection
of 2H-TRLs—4- to 6-fold more than in mitochondria. After
30 min, 2H enrichment was 12-fold greater in lipid droplets
than in mitochondria, likely reflecting intervening oxidation of
2
H-labeled fatty acids in mitochondria. We also examined
13
C distribution in cardiomyocytes 1 day after administering
13
C-labeled fatty acids by gavage. Only 5% of the 13C in cardi-
omyocytes was located in cytosolic lipid droplets, likely reflect-
ing the use of fatty acids for membrane phospholipids and the
use of fatty acid metabolites in the synthesis of nonessential
amino acids.
CD36 is expressed at high levels in capillary endothelial cells
(Greenwalt et al., 1995), and is often assumed to play a role in
uptake of fatty acids by cells and tissues (Tanaka et al., 1997;
Yang et al., 2007). When radioiodinated fatty acid derivatives
are injected intravenously into humans, they are taken up by
the heart. Uptake of those radiopharmaceuticals by the heart
is reduced in subjects with CD36 deficiency (Hwang et al.,
1998; Tanaka et al., 1997). Also, when Cd36 / mice are
fasted, fatty acids accumulate in the plasma, and triglyceride
accumulation in the heart is reduced (Bharadwaj et al., 2010).
These observations suggest that CD36 could be important for
the uptake of free fatty acids in the plasma compartment, but
it is unclear whether CD36 deficiency influences uptake of
TRL-derived nutrients by cardiomyocytes. Bharadwaj et al.
(2010) loaded human VLDL with 14C-labeled triglycerides
in vitro and then injected the TRLs into wild-type and
Cd36 / mice; they found reduced amounts of 14C in heart tis-
sue extracts from Cd36 / mice. On the other hand, Cd36 defi-
ciency appeared to have little effect on the uptake of lipids from
TRLs that had been endogenously labeled with 14C (14C-TRLs
prepared in Lpl-deficient mice). In the current study, we used
NanoSIMS to assess the effect of Cd36 deficiency on the fate
of 2H-TRL-derived lipids in the heart. After 1.5 or 2 min, 2H entry
into wild-type and Cd36 / cardiomyocytes was nearly iden-
tical, indicating that CD36 deficiency has little effect on the
movement of TRL-derived lipids into cardiomyocytes. (This
observation seems consistent with an absence of impaired
cardiac performance in Cd36 / mice [Koonen et al., 2007].)
However, we observed reduced entry of 2H into cardiomyocyte
lipid droplets in Cd36 / mice. The cytosolic lipid droplets were
slightly smaller in Cd36 / mice, but we are skeptical that their
smaller size prevented us from detecting them by NanoSIMS.
Cd36 deficiency has complex effects on cardiomyocytes,
including effects on insulin and calcium signaling, fatty acyl-
CoA levels, and fatty acid esterification (Abumrad and Gold-
berg, 2016; Bharadwaj et al., 2010; Xu et al., 2013), and those
effects presumably could contribute, directly or indirectly, to
(E) 13C/12C ratio in cardiomyocytes of Cd36+/+ and Cd36 / mice. Each data point represents one cardiomyocyte in six 30 3 30 mm NanoSIMS images.
***p < 0.0005.
(F) 13C enrichment in cardiomyocyte lipid droplets as a percentage of total 13C in cardiomyocytes. Graphs were generated by quantifying six 30 3 30 mm images
for each sample. **p < 0.01. See also Figures S5–S7.
(G) 13C/12C ratio in mitochondria and lipid droplets of Cd36+/+ and Cd36 / mice. Six 30 3 30 mm NanoSIMS images were analyzed. Each data point shows the
mean 13C/12C ratio in the mitochondria and lipid droplets in each image (>50 mitochondria and 5–50 lipid droplets were analyzed per image). ***p < 0.0005.
1064 Cell Metabolism 27, 1055–1066, May 1, 2018
changes in the distribution of 2H-TRL-derived lipids within
cardiomyocytes.
NanoSIMS analyses are particularly helpful when, as in the
current study, there are anatomical questions to address (i.e.,
an interest in defining the distribution of an isotope in different
cell types or different subcellular compartments). NanoSIMS
imaging will continue to be a useful approach in sorting out fac-
tors that regulate lipolysis and lipid movement within tissues. In
the past, others have concluded, based largely on perturbations
of lipid stores in knockout mice, that vascular endothelial growth
factor-B is crucial for regulating fatty acid transport across capil-
lary endothelial cells (Hagberg et al., 2010). Fatty acid-binding
proteins and fatty acyl-CoA synthetases have also been pro-
posed to play a role in the delivery of lipids to myocytes (Hagberg
et al., 2010, 2013). Recently, a valine metabolite, 3-hydroxyiso-
butyrate, was proposed to regulate fatty acid transport across
endothelial cells (Jang et al., 2016), based largely on in vitro
models and measurements of lipid stores in tissues. NanoSIMS
imaging could be useful for defining the impact of both specific
genes and small molecules in lipid movement into and across
capillaries. The utility of NanoSIMS imaging for metabolism
research will extend well beyond TRL metabolism. For example,
NanoSIMS could simultaneously quantify the uptake and sub-
cellular distribution of 2H-labeled glucose (or 2-deoxyglucose),
15
N-labeled glutamine, and 13C-labeled fatty acids in solid
tumors, making it possible to define metabolic heterogeneity in
tumor cells, stromal cells, and endothelial cells.
Limitations of Study
There are certain limitations to our study. One is that the number
of mice that we studied was low, and the number of images
generated per mouse was rather limited. NanoSIMS imaging is
very expensive, slow, and technically demanding. In a 24-hr
period, only 8 to 12 images could be obtained from one very small
region of the heart. Although each image is generated from an
extraordinary amount of quantitative data on secondary ions, it
is simply not feasible to image many mice per group, many
different lines of mice, or many different tissues. That is why we
believe that NanoSIMS is most suitable for addressing an
anatomical issue and less suitable when scientific questions
can be addressed by measuring tritium-labeled lipids in tissue
extracts. Another limitation of our study is that it is conceivable
that there was some fatty acid diffusion within tissues during
the PBS perfusion (prior to perfusion-fixation with carbodiimide
and glutaraldehyde) or some fatty acid diffusion after fixation dur-
ing the preparation of tissues for NanoSIMS imaging. However,
this type of lipid diffusion could not explain the rapid entry of
deuterium into neutral lipids of cytosolic lipid droplets of
cardiomyocytes. Our NanoSIMS images uncovered deuterium-
labeled lipids in the subendothelial spaces of the heart, but it
is likely that the patchy localization of those lipids was due
to imidazole and osmium binding. Also, even though the
deuterium-enriched, imidazole-staining patches were primarily
located around small capillaries, we cannot fully exclude the pos-
sibility that some of the deuterium-labeled lipids moved into the
subendothelial spaces during tissue processing. Finally, we
would point out that the 13C NanoSIMS images (obtained at least
22 hr after the administration of 13C-labeled fatty acids) should
not be interpreted as exclusively showing lipid distribution, given
that some of the 13C-labeled fatty acids were metabolized and
used to generate amino acid substrates for protein synthesis.
STAR+METHODS
Detailed methods are provided in the online version of this paper
and include the following:
d KEY RESOURCES TABLE
d CONTACT FOR REAGENT AND RESOURCE SHARING
d EXPERIMENTAL MODEL AND SUBJECT DETAILS
B Mice
d METHOD DETAILS
2 Beigneux, A.P., Gin, P., Davies, B.S.J., Weinstein, M.M., Bensadoun, A., Fong,
B H-labeled Triglyceride-rich Lipoproteins
13 2
B Administration of [ C]Fatty Acids and H-TRLs to Mice
B Isolated Mouse Heart Experiments
B Preparation of Tissue Sections for Electron Micro-
scopy and NanoSIMS Imaging
B NanoSIMS and Backscattered Electron (BSE) Imaging
B Transmission Electron Microscopy
B Transmission Electron Microscopy of Negatively
Stained Lipoproteins
B Immunohistochemistry
B Statistics
SUPPLEMENTAL INFORMATION
Supplemental Information includes seven figures and one table and can be
found with this article online at https://doi.org/10.1016/j.cmet.2018.03.017.
ACKNOWLEDGMENTS
We are grateful for the NanoSIMS facilities at California Institute of Technology
(Pasadena, CA) and the University of Western Australia (Perth). We acknowl-
edge support from the Leducq Foundation Transatlantic Network grant
(12CVD04) (to S.G.Y.), the NIH (P01 HL090553, R01 HL087228, and
HL125335 to S.G.Y.), and a Ruth L. Kirschstein National Research Service
Award, F32 HL132471 (to C.H.). We also acknowledge the Australian Micro-
scopy & Microanalysis Research Facility and the Science and Industry Endow-
ment Fund for supporting the Ion Probe Facility at the Centre for Microscopy,
Characterisation and Analysis at the University of Western Australia. We thank
Dr. Marianne Cilluffo for help in preparing tissue sections.
AUTHOR CONTRIBUTIONS
C.H. designed, performed, and analyzed the experiments, and prepared fig-
ures and the initial draft of the paper. T.A.W. performed electron microscopy
and prepared tissue sections for NanoSIMS. R.S.J. performed the experi-
ments and analyzed NanoSIMS data. P.H. managed the mouse colony and
performed mouse experiments. M.L. performed isolated heart studies. X.H.
performed NanoSIMS studies. C.M.A. performed immunohistochemistry
studies. P.T., K.R., M.P., and A.P.B. advised on experimental design and
data analysis. A.H., M.K., and P.G. assisted with NanoSIMS. D.A.F. performed
lipidomics studies. L.G.F. and S.G.Y. analyzed the data and prepared the
paper. H.J. performed NanoSIMS and BSE imaging, analyzed the data, and
assisted with figures and the paper.
DECLARATION OF INTERESTS
The authors declare no competing interests.
Received: December 14, 2017
Revised: February 16, 2018
Accepted: March 24, 2018
Published: May 1, 2018
REFERENCES
Abumrad, N.A., and Goldberg, I.J. (2016). CD36 actions in the heart: lipids,
calcium, inflammation, repair and more? Biochim. Biophys. Acta 1861,
1442–1449.
Angermu €ller, S., and Fahimi, H.D. (1982). Imidazole-buffered osmium tetrox-
ide: an excellent stain for visualization of lipids in transmission electron micro-
scopy. Histochem. J. 14, 823–835.
Beigneux, A.P., Davies, B., Gin, P., Weinstein, M.M., Farber, E., Qiao, X.,
Peale, P., Bunting, S., Walzem, R.L., Wong, J.S., et al. (2007).
Glycosylphosphatidylinositol-anchored high density lipoprotein-binding
protein 1 plays a critical role in the lipolytic processing of chylomicrons. Cell
Metab. 5, 279–291.
L.G., and Young, S.G. (2009). Highly conserved cysteines within the Ly6
domain of GPIHBP1 are crucial for the binding of lipoprotein lipase. J. Biol.
Chem. 284, 30240–30247.
Bharadwaj, K.G., Hiyama, Y., Hu, Y., Huggins, L.A., Ramakrishnan, R.,
Abumrad, N.A., Shulman, G.I., Blaner, W.S., and Goldberg, I.J. (2010).
Chylomicron- and VLDL-derived lipids enter the heart through different path-
ways: in vivo evidence for receptor- and non-receptor-mediated fatty acid
uptake. J. Biol. Chem. 285, 37976–37986.
Bligh, E.G., and Dyer, W.J. (1959). A rapid method of total lipid extraction and
purification. Can. J. Biochem. Physiol. 37, 911–917.
Dallwig, R., and Deitmer, J.W. (2002). Cell-type specific calcium responses in
acute rat hippocampal slices. J. Neurosci. Methods 116, 77–87.
Davies, B.S., Goulbourne, C.N., Barnes, R.H., 2nd, Turlo, K.A., Gin, P.,
Vaughan, S., Vaux, D.J., Bensadoun, A., Beigneux, A.P., Fong, L.G., et al.
(2012). Assessing mechanisms of GPIHBP1 and lipoprotein lipase movement
across endothelial cells. J. Lipid Res. 53, 2690–2697.
Davies, B.S.J., Beigneux, A.P., Barnes, R.H., II, Tu, Y., Gin, P., Weinstein,
M.M., Nobumori, C., Nyrén, R., Goldberg, I.J., Olivecrona, G., et al. (2010).
GPIHBP1 is responsible for the entry of lipoprotein lipase into capillaries.
Cell Metab. 12, 42–52.
Febbraio, M., Abumrad, N.A., Hajjar, D.P., Sharma, K., Cheng, W., Pearce,
S.F.A., and Silverstein, R.L. (1999). A null mutation in murine CD36 reveals
an important role in fatty acid and lipoprotein metabolism. J. Biol. Chem.
274, 19055–19062.
Fong, L.G., Young, S.G., Beigneux, A.P., Bensadoun, A., Oberer, M., Jiang, H.,
and Ploug, M. (2016). GPIHBP1 and plasma triglyceride metabolism. Trends
Endocrinol. Metab. 27, 455–469.
Goulbourne, C.N., Gin, P., Tatar, A., Nobumori, C., Hoenger, A., Jiang, H.,
Grovenor, C.R., Adeyo, O., Esko, J.D., Goldberg, I.J., et al. (2014). The
GPIHBP1-LPL complex is responsible for the margination of triglyceride-rich
lipoproteins in capillaries. Cell Metab. 19, 849–860.
Greenwalt, D.E., Scheck, S.H., and Rhinehart-Jones, T. (1995). Heart CD36
expression is increased in murine models of diabetes and in mice fed a high
fat diet. J. Clin. Invest. 96, 1382–1388.
Guo, W., Huang, N., Cai, J., Xie, W., and Hamilton, J.A. (2006). Fatty acid trans-
port and metabolism in HepG2 cells. Am. J. Physiol. Gastrointest. Liver
Physiol. 290, G528–G534.
Hagberg, C., Mehlem, A., Falkevall, A., Muhl, L., and Eriksson, U. (2013).
Endothelial fatty acid transport: role of vascular endothelial growth factor B.
Physiology (Bethesda) 28, 125–134.
Hagberg, C.E., Falkevall, A., Wang, X., Larsson, E., Huusko, J., Nilsson, I., van
Meeteren, L.A., Samen, E., Lu, L., Vanwildemeersch, M., et al. (2010). Vascular
Cell Metabolism 27, 1055–1066, May 1, 2018 1065
endothelial growth factor B controls endothelial fatty acid uptake. Nature 464,
917–921.
Hamilton, J.A., Johnson, R.A., Corkey, B., and Kamp, F. (2001). Fatty acid
transport: the diffusion mechanism in model and biological membranes.
J. Mol. Neurosci. 16, 99–108, discussion 151–107.
Havel, R.J. (2010). Triglyceride-rich lipoproteins and plasma lipid transport.
Arterioscler. Thromb. Vasc. Biol. 30, 9–19.
Havel, R.J., and Gordon, R.S., Jr. (1960). Idiopathic hyperlipemia: metabolic
studies in an affected family. J. Clin. Invest. 39, 1777–1790.
Havel, R.J., and Kane, J.P. (2001). Introduction: structure and metabolism of
plasma lipoproteins. In The Metabolic and Molecular Bases of Inherited
Disease, C.R. Scriver, A.L. Beaudet, W.S. Sly, D. Valle, B. Childs, K.W.
Kinzler, and B. Vogelstein, eds. (McGraw-Hill), pp. 2705–2716.
He, C., Fong, L.G., Young, S.G., and Jiang, H. (2017a). NanoSIMS imaging: an
approach for visualizing and quantifying lipids in cells and tissues. J. Investig.
Med. 65, 669–672.
He, C., Hu, X., Jung, R.S., Weston, T.A., Sandoval, N.P., Tontonoz, P., Kilburn,
M.R., Fong, L.G., Young, S.G., and Jiang, H. (2017b). High-resolution imaging
and quantification of plasma membrane cholesterol by NanoSIMS. Proc. Natl.
Acad. Sci. USA 114, 2000–2005.
Hwang, E.H., Taki, J., Yasue, S., Fujimoto, M., Taniguchi, M., Matsunari, I.,
Nakajima, K., Shiobara, S., Ikeda, T., and Tonami, N. (1998). Absent myocar-
dial iodine-123-BMIPP uptake and platelet/monocyte CD36 deficiency.
J. Nucl. Med. 39, 1681–1684.
Jang, C., Oh, S.F., Wada, S., Rowe, G.C., Liu, L., Chan, M.C., Rhee, J.,
Hoshino, A., Kim, B., Ibrahim, A., et al. (2016). A branched-chain amino acid
metabolite drives vascular fatty acid transport and causes insulin resistance.
Nat. Med. 22, 421–426.
Jiang, H., Favaro, E., Goulbourne, C.N., Rakowska, P.D., Hughes, G.M.,
Ryadnov, M.G., Fong, L.G., Young, S.G., Ferguson, D.J., Harris, A.L., et al.
(2014a). Stable isotope imaging of biological samples with high resolution sec-
ondary ion mass spectrometry and complementary techniques. Methods 68,
317–324.
Jiang, H., Goulbourne, C.N., Tatar, A., Turlo, K., Wu, D., Beigneux, A.P.,
Grovenor, C.R., Fong, L.G., and Young, S.G. (2014b). High-resolution imaging
of dietary lipids in cells and tissues by NanoSIMS analysis. J. Lipid Res. 55,
2156–2166.
Kane, J.P., and Havel, R.J. (2001). Disorders of the biogenesis and secretion of
lipoproteins containing the B apolipoproteins. In The Metabolic and Molecular
Bases of Inherited Disease, C.R. Scriver, A.L. Beaudet, W.S. Sly, D. Valle, B.
Childs, K.W. Kinzler, and B. Vogelstein, eds. (McGraw-Hill), pp. 2717–2752.
Koonen, D.P., Febbraio, M., Bonnet, S., Nagendran, J., Young, M.E.,
Michelakis, E.D., and Dyck, J.R. (2007). CD36 expression contributes to
age-induced cardiomyopathy in mice. Circulation 116, 2139–2147.
Korn, E.D. (1955a). Clearing factor, a heparin-activated lipoprotein lipase. I.
Isolation and characterization of the enzyme from normal rat heart. J. Biol.
Chem. 215, 1–14.
Korn, E.D. (1955b). Clearing factor, a heparin-activated lipoprotein lipase. II.
Substrate specificity and activation of coconut oil. J. Biol. Chem. 215, 15–26.
1066 Cell Metabolism 27, 1055–1066, May 1, 2018
Mysling, S., Kristensen, K.K., Larsson, M., Beigneux, A.P., Gardsvoll, H., Fong,
L.G., Bensadouen, A., Jorgensen, T.J., Young, S.G., and Ploug, M. (2016a).
The acidic domain of the endothelial membrane protein GPIHBP1 stabilizes
lipoprotein lipase activity by preventing unfolding of its catalytic domain.
eLife 5, e12095.
Mysling, S., Kristensen, K.K., Larsson, M., Kovrov, O., Bensadouen, A.,
Jorgensen, T.J., Olivecrona, G., Young, S.G., and Ploug, M. (2016b). The
angiopoietin-like protein ANGPTL4 catalyzes unfolding of the hydrolase
domain in lipoprotein lipase and the endothelial membrane protein GPIHBP1
counteracts this unfolding. eLife 5, e20958.
Quehenberger, O., Armando, A.M., Brown, A.H., Milne, S.B., Myers, D.S.,
Merrill, A.H., Bandyopadhyay, S., Jones, K.N., Kelly, S., Shaner, R.L., et al.
(2010). Lipidomics reveals a remarkable diversity of lipids in human plasma.
J. Lipid Res. 51, 3299–3305.
Scow, R.O., Blanchette-Mackie, E.J., and Smith, L.C. (1976). Role of capillary
endothelium in the clearance of chylomicrons. A model for lipid transport from
blood by lateral diffusion in cell membranes. Circ. Res. 39, 149–162.
Shao, F., and Ford, D.A. (2014). Elaidic acid increases hepatic lipogenesis by
mediating sterol regulatory element binding protein-1c activity in huh-7 cells.
Lipids 49, 403–413.
Sidossis, L.S., Coggan, A.R., Gastaldelli, A., and Wolfe, R.R. (1995). Pathway
of free fatty acid oxidation in human subjects. Implications for tracer studies.
J. Clin. Invest. 95, 278–284.
Takei, S., Hasegawa-Ishii, S., Uekawa, A., Chiba, Y., Umegaki, H., Hosokawa, M.,
Woodward, D.F., Watanabe, K., and Shimada, A. (2012). Immunohistochemical
demonstration of increased prostaglandin F(2)alpha levels in the rat hippocam-
pus following kainic acid-induced seizures. Neuroscience 218, 295–304.
Tanaka, T., Okamoto, F., Sohmiya, K., and Kawamura, K. (1997). Lack of
myocardial iodine-123 15-(p-iodiphenyl)-3-R,S-methylpentadecanoic acid
(BMIPP) uptake and CD36 abnormality–CD36 deficiency and hypertrophic
cardiomyopathy. Jpn. Circ. J. 61, 724–725.
Tymianski, M., Bernstein, G.M., Abdel-Hamid, K.M., Sattler, R., Velumian, A.,
Carlen, P.L., Razavi, H., and Jones, O.T. (1997). A novel use for a carbodiimide
compound for the fixation of fluorescent and non-fluorescent calcium indica-
tors in situ following physiological experiments. Cell Calcium 21, 175–183.
Weinstein, M.M., Yin, L., Tu, Y., Wang, X., Wu, X., Castellani, L.W., Walzem,
R.L., Lusis, A.J., Fong, L.G., Beigneux, A.P., et al. (2010). Chylomicronemia
elicits atherosclerosis in mice. Arterioscler. Thromb. Vasc. Biol. 30, 20–23.
Xu, S., Jay, A., Brunaldi, K., Huang, N., and Hamilton, J.A. (2013). CD36
enhances fatty acid uptake by increasing the rate of intracellular esterification
but not transport across the plasma membrane. Biochemistry 52, 7254–7261.
Yang, J., Sambandam, N., Han, X., Gross, R.W., Courtois, M., Kovacs, A.,
Febbraio, M., Finck, B.N., and Kelly, D.P. (2007). CD36 deficiency rescues
lipotoxic cardiomyopathy. Circ. Res. 100, 1208–1217.
Yao, G. (2000). Effect of LPL activity on fatty acid composition in mouse
plasma, liver, skeletal and cardiac muscles, adipose and brain tissues,
M.Sc. thesis (Université Laval).
Neuron

Article

TREM2 Regulates Microglial Cholesterol Metabolism

upon Chronic Phagocytic Challenge
Alicia A. Nugent,1,2 Karin Lin,1,2 Bettina van Lengerich,1 Steve Lianoglou,1 Laralynne Przybyla,1 Sonnet S. Davis,1
Ceyda Llapashtica,1 Junhua Wang,1 Do Jin Kim,1 Dan Xia,1 Anthony Lucas,1 Sulochanadevi Baskaran,1,3
Patrick C.G. Haddick,1 Melina Lenser,1 Timothy K. Earr,1 Ju Shi,1,4 Jason C. Dugas,1 Benjamin J. Andreone,1 Todd Logan,1
Hilda O. Solanoy,1 Hang Chen,1,5 Ankita Srivastava,1 Suresh B. Poda,1 Pascal E. Sanchez,1 Ryan J. Watts,1
Thomas Sandmann,1 Giuseppe Astarita,1 Joseph W. Lewcock,1 Kathryn M. Monroe,1,* and Gilbert Di Paolo1,6,*
1Denali Therapeutics, South San Francisco, CA 94080, USA
2These authors contributed equally
3Current address: Frontier Medicines, South San Francisco, CA 94080, USA
4Current address: Jazz Pharmaceuticals, Palo Alto, CA 94304, USA
5Current address: BridGene Biosciences, Sunnyvale, CA 94089, USA
6Lead Contact

*Correspondence: monroe@dnli.com (K.M.M.), dipaolo@dnli.com (G.D.P.)

https://doi.org/10.1016/j.neuron.2019.12.007

SUMMARY ceptors, and enhanced cytokine secretion (Deczkowska et al.,

Loss-of-function (LOF) variants of TREM2, an im-
mune receptor expressed in microglia, increase
Alzheimer’s disease risk. TREM2 senses lipids and
mediates myelin phagocytosis, but its role in micro-
glial lipid metabolism is unknown. Combining
chronic demyelination paradigms and cell sorting
with RNA sequencing and lipidomics, we find that
wild-type microglia acquire a disease-associated
transcriptional state, while TREM2-deficient micro-
glia remain largely homeostatic, leading to neuronal
damage. TREM2-deficient microglia phagocytose
myelin debris but fail to clear myelin cholesterol, re-
sulting in cholesteryl ester (CE) accumulation. CE in-
crease is also observed in APOE-deficient glial cells,
reflecting impaired brain cholesterol transport. This
finding replicates in myelin-treated TREM2-deficient
murine macrophages and human iPSC-derived mi-
croglia, where it is rescued by an ACAT1 inhibitor
and LXR agonist. Our studies identify TREM2 as a
key transcriptional regulator of cholesterol transport
and metabolism under conditions of chronic myelin
phagocytic activity, as TREM2 LOF causes patho-
genic lipid accumulation in microglia.
INTRODUCTION
Microglia constantly surveil the brain parenchyma to eliminate
dead cells, dysfunctional synapses, and other cellular debris
(Deczkowska et al., 2018; Hammond et al., 2018; Ransohoff,
2016). With aging or in pathological conditions, parenchyma
damage can facilitate the transition of microglia from homeostat-
ic into reactive states. This transition profoundly alters the
microglial transcriptome, leading to morphological changes,
increased phagocytic activity, expression of various immune re-
2018; Hammond et al., 2018). Recent advances in single-cell
RNA sequencing (scRNA-seq) have revealed multiple states of
microglia. These studies suggest that functional sub-popula-
tions of microglia co-exist within the brain in various mouse
models of neurodegenerative diseases, including homeostatic
and disease-associated microglia (DAM), the microglial neuro-
degenerative phenotype (MGnD), and activated response micro-
glia (ARM) (Deczkowska et al., 2018; Keren-Shaul et al., 2017;
Krasemann et al., 2017; Sala Frigerio et al., 2019). Collectively,
we refer to responsive microglial states as damage-associated
microglia, because different disease or aging models show sig-
nificant overlap but distinct transcriptional profiles. Whether
transition to damage-associated microglia states is beneficial
or deleterious remains unclear and likely depends on the disease
type, stage, and inherent pathology. Identifying the molecular
and cellular basis underlying microglial states in healthy and dis-
ease conditions may facilitate the development of immune ther-
apies for the treatment of neurodegenerative diseases.
Microglial dysfunction appears central to the etiology of late-
onset Alzheimer’s disease (LOAD), as large-scale genetic
studies have uncovered variants in LOAD risk-associated genes
that are highly expressed in microglia (Carmona et al., 2018;
Efthymiou and Goate, 2017). One of the LOAD genes encodes
triggering receptor expressed on myeloid cells 2 (TREM2), a sin-
gle-pass transmembrane immune receptor selectively ex-
pressed in microglia within the CNS. Individuals carrying rare
heterozygous variants of TREM2, such as R47H, have higher
LOAD risk (average odds ratio [OR] 4.5) (Guerreiro et al.,
2013; Jonsson et al., 2013; Ulland and Colonna, 2018). The
immunoglobulin-like ectodomain (ECD) of TREM2 binds various
ligands, including lipids, and the AD variant R47H has reduced
affinity for apolipoproteins (e.g., APOE) or lipid ligands, suggest-
ing that the increased LOAD risk reflects a partial loss of function
(LOF) in lipid recognition and/or lipid-induced signaling (Atagi
et al., 2015; Bailey et al., 2015; Sudom et al., 2018; Ulland and
Colonna, 2018; Wang et al., 2016; Yeh et al., 2016). Trem2–/–
and Trem2R47H mutant microglia fail to surround and clear
amyloid plaques in vivo, resulting in accumulation of dystrophic

Neuron 105, 837–854, March 4, 2020 ª 2019 Elsevier Inc. 837

 A C F
Trem2 + /+ + /- -/- Microglial DAM genes
Control CPZ Control CPZ Control CPZ
+/+
5 week
+/- -/- +/+
12 week
+/- -/-
Ctl CPZ Ctl CPZ Ctl CPZ Ctl CPZ Ctl CPZ Ctl CPZ

5 week CPZ

Down
Trem2 + /+ + /- -/-
Ctl CPZ Control CPZ Control CPZ
12 week CPZ

Trem2 +/+ Control

Trem2 +/- Cuprizone
Trem2 -/-

B Trem2 genotype effect 12 wk Cuprizone

effect
+/+ vs -/- +/+ vs +/- +/+ +/-

6 0 0 199 159 740

Up
0 8
0 0 1 6
0 2

+/- vs -/- -/-

D E
Lysosomal function Lipid metabolism
control treatment (log2)
control treatment (log2)

Change from +/+

Change from +/+

+/+ +/- -/- +/+ +/- -/- +/+ +/- -/- +/+ +/- -/-
Trem2 genotype Trem2 genotype log2 FC
Cst7 Ctse Galns Hexa Apoc1
A Fabp3 Lpl Olr1 4 2 0 -2 -4
Ctsb Ctsl Gla Prcp Apoe Fabp5
F Nceh1 Soat1
Ctsd Ctsz Gusb Ch25h Lipa Npc2
G H I
Homeostatic genes Stage 1 DAM genes Stage 2 DAM genes
control treatment (log2)

control treatment (log2)

control treatment (log2)

Change from +/+

Change from +/+

Change from +/+

+/+ +/- -/- +/+ +/- -/- +/+ +/- -/- +/+ +/- -/- +/+ +/- -/- +/+ +/- -/-
Trem2 genotype Trem2 genotype Trem2 genotype
Cx3cr1 P2ry12 Tmem119
T Apoe Ctsb Fth1 Tyrobp Axl Cd9 Csf1 Ctsl Lpl
B2m Ctsd Lyz2 Ccl6 Clec7a Csf7 Itgax Timp2

Figure 1. TREM2 Deficiency Prevents DAM Conversion during Chronic Demyelination

(A) Principal-component analysis of top 500 differentially expressed genes in bulk microglia isolated from brains of control versus 5 week and control versus
12 week cuprizone (CPZ)-treated Trem2+/+, Trem2+/–, and Trem2–/– mice (n = 2–4 mice per condition).
(B) Number of differentially expressed genes in bulk microglia after control (left) or 12 week CPZ diet (right). ANOVA; FDR < 0.1; absolute log2 FC > 0.5.
(C) Heatmap of top gene expression changes (log2 fold change) in bulk microglia after 5 (top) and 12 (bottom) week CPZ diet. Orange marks indicate previously
identified DAM genes. ANOVA, FDR < 0.1, absolute log2 FC > 0.5; columns represent individual mice.

(legend continued on next page)

838 Neuron 105, 837–854, March 4, 2020

neurites near diffuse plaques (Jay et al., 2017; Parhizkar et al.,
2019; Ulland and Colonna, 2018; Wang et al., 2015, 2016;
Yuan et al., 2016). Additionally, TREM2 LOF modulates tau seed-
ing, spreading, and tau-associated neuroinflammation and
neurotoxicity (Bemiller et al., 2017; Leyns et al., 2017, 2019;
Sayed et al., 2018). Trem2 overexpression reduces amyloid
burden and neuritic dystrophy in an Alzheimer’s disease (AD)
mouse model (Lee et al., 2018). Rare homozygous TREM2 LOF
mutations cause Nasu-Hakola disease, a syndrome with recur-
rent bone fractures, myelin loss, neurodegeneration, and early-
onset dementia, likely resulting from microglial dysfunction
(Paloneva et al., 2002).
Consistent with a role in microglia proliferation and survival,
TREM2 signals via receptor tyrosine kinases DNAX-activation
protein 10 (DAP10) and DAP12 to modulate proliferation, sur-
vival, immune response, calcium mobilization, cytoskeletal dy-
namics, mTOR signaling, autophagy, and energy metabolism
(Ulland and Colonna, 2018; Ulland et al., 2017; Yeh et al.,
2017). A proposed function of TREM2 is to mediate microglial
response and transition to a damage-associated microglia state
via control of gene expression (Deczkowska et al., 2018; Ulland
and Colonna, 2018). scRNA-seq analyses of TREM2-deficient
5XFAD microglia revealed the inability to acquire a late-stage
DAM profile (DAM stage 2), instead halting at an intermediate
DAM state (DAM stage 1) (Keren-Shaul et al., 2017). Other
studies showed a key role for TREM2 and APOE in the transition
to damage-associated microglia states (Krasemann et al., 2017;
Götzl et al., 2019). Microarray studies in Trem2–/– mice subjected
to a demyelinating cuprizone (CPZ) diet also showed that TREM2
controls expression of many microglial genes controlling lipid
transport (Apoe) or catabolism (Lpl) (Poliani et al., 2015).
Trem2–/– mice showed normal demyelination in the acute phase
but partial remyelination during the recovery phase, suggesting
that TREM2 may bind to myelin debris for engulfment and clear-
ance, a process required for proper remyelination (Cantoni et al.,
2015; Poliani et al., 2015). However, whether TREM2 mediates
myelin lipid processing in microglia is unknown.
We sought to assess the physiological role of TREM2 in
microglial lipid metabolism by subjecting mice to a demyelin-
ating CPZ diet (Praet et al., 2014). Myelin contains the majority
(>80%) of the brain’s free cholesterol (Martı́n et al., 2014),
allowing the assessment of microglial response to a chronic
and extensive cholesterol challenge that is selective to
the brain and independent of peripheral lipid metabolism.
CPZ triggered the TREM2-dependent transition from a
homeostatic to damage-associated microglia state with
upregulation of genes controlling lipid transport and meta-
bolism. Chronic myelin phagocytosis caused accumulation
of cholesteryl ester (CE) and oxidized CE (oxCE) in Trem2–/–
brain. Fluorescence-activated cell sorting (FACS)-based
(D and E) Expression changes in individual genes associated with (D) lysosomal function and (E) lipid metabolism in bulk microglia with control (left inset) or CPZ
diet (right inset) for 5 (top) or 12 (bottom) weeks.
(F) Heatmap of bulk microglial expression changes in the top 69 DAM genes downregulated (top) or upregulated (bottom) in 5XFAD compared with wild-type
microglia from Karen-Shaul et al. 2017 after 5 and 12 week control versus CPZ diet. Camera; p < 1 3 10 41.
(G–I) Expression changes in individual (G) homeostatic, (H) stage 1, and (I) stage 2 DAM genes (identified in Keren-Shaul et al., 2017) in Trem2+/+, Trem2+/–, and
Trem2–/– bulk microglia with control (left inset) or CPZ diet (right inset) for 5 (top) or 12 (bottom) weeks.
See also Figure S1 and Data S1.
lipidomic techniques revealed elevation of CE in Trem2–/– mi-
croglia, a phenotype also observed in Apoe–/– microglia after
CPZ diet. This could be reproduced in vitro, where CE accu-
mulation occurs in Trem2–/– bone marrow-derived macro-
phages (BMDMs) and human induced pluripotent stem cell
(iPSC)-derived microglia subjected to myelin challenge. CE
accumulation was rescued by inhibition of acetyl-CoA acetyl-
transferase 1 (ACAT1), which converts cholesterol to CE, and
by a liver X receptor (LXR) agonist, suggesting that TREM2
LOF causes an impairment in cholesterol efflux. Our results
establish a key role for TREM2 in the control of microglial
gene expression and cholesterol transport upon chronic
myelin phagocytosis. Failure to properly execute this program
results in extensive neuronal damage in the brain. Because CE
is known to accumulate in AD patient brain and AD mouse
models (Astarita et al., 2011; Chan et al., 2012; Morel et al.,
2013; Shibuya et al., 2015), and LOAD-linked TREM2 variants
result in a partial LOF (Ulland and Colonna, 2018), our study
strongly suggests that enhancing TREM2 function may be
beneficial in AD, in part by facilitating lipid clearance in
microglia.
RESULTS
TREM2 Deficiency Prevents DAM Conversion during
Chronic Demyelination
To characterize the effects of acute and chronic demyelination
on TREM2-dependent gene expression in microglia, Trem2+/+,
Trem2+/–, and Trem2–/– mice were fed a 0.2% CPZ diet for 5 or
12 weeks. CD11b+ microglia were isolated from hemibrain using
FACS (Figures S1A–S1E). CPZ induced transcriptional changes
in Trem2+/+ and Trem2+/– microglia, whereas CPZ-challenged
Trem2–/– microglia clustered with those of untreated mice (prin-
cipal-component analysis [PCA]; Figure 1A). Few genotype-
dependent differences were found under normal diet conditions
(Figure 1B), but CPZ caused significant changes in hundreds of
genes after 5 and 12 weeks (Figures 1B and 1C; absolute log2
fold change [FC] > 0.5, false discovery rate [FDR] < 0.2; Data
S1). These changes were restricted primarily to Trem2+/+ and
Trem2+/– microglia, while Trem2–/– microglia largely failed to
respond to CPZ (Figures 1B and 1C).
Analysis of gene sets from the Reactome database (Fabregat
et al., 2018) revealed Trem2-dependent upregulation of genes
involved in lysosome and phagosome function, AD, oxidative
phosphorylation, and cholesterol metabolism. Key lysosome
cathepsin genes, such as Ctse and Ctsl, were upregulated
2-fold upon chronic demyelination in Trem2+/+ and Trem2+/– mi-
croglia (FDR < 0.05) but unchanged in Trem2–/– microglia (Fig-
ure 1D; Data S1; interaction p < 0.05). Previous microarray
studies reported a failure of Trem2–/– microglia to upregulate
Neuron 105, 837–854, March 4, 2020 839
 A B
All cells Sample enrichment DAM activity
1
Cluster
1 5
2 30
6 20
3
10
4 2 0
5 S
6 8
7
8 4

3
0 5 10 15 20 Trem2 Trem2 Trem2 Trem2 DAM1 DAM1 DAM2
% of sample +/+ +/+ +/- -/- down up up
(average) Ctl 12wk 12wk 12wk
CPZ CPZ CPZ
C D
Cluster gene markers
Cluster gene marker expression
3000
Apoe

2000
1000
1 down

0
50
40
30
Lpl

20
10
0
1 up

200
Ctsb

100
0
600
2 up

Ctsd

400
200
0
90
Ctsz
3 down

60
30
0
100
P2ry12

75
50
25
3 up

0
200
150
B2m

100
50
0

Normalized UMI
4 down

40
Cd68

20
0
100
Cx3cr1

75
50
4 up

25
0
60
Lgals3

40
5u 5d

20
0
600
Spp1

400
200
6d

0
50
6u

40
Cst7

30
20
10
7d

0
120
Cd63

80
40
7 up

0
20
Capg

15
10
5
0
8 up

300
Fth1

200
100
0

Figure 2. scRNA-Seq Confirms that Trem2–/– Microglia Exhibit Attenuated Transition to DAM upon Demyelination
(A) tSNE (t-distributed stochastic neighbor embedding) plot of the 3,023 single microglia sorted from Trem2+/+ with control diet and Trem2+/+, Trem2+/–, and
Trem2–/– with 12 week CPZ, colored by cluster assignment (n = 2 pooled mice per condition).
(B) Percentage composition of all cells within each cluster across all samples (left), within each cluster per condition (middle), and aggregated scores of DAM-
related gene sets per cluster (right). Gene set scores per cell are normalized to zero mean and unit variance and visualized in the heatmap as their average over all
cells per cluster.
(legend continued on next page)

840 Neuron 105, 837–854, March 4, 2020

genes involved in phagocytosis and lipid metabolism (Poliani
et al., 2015). We confirmed these results and expanded the list
of lipid metabolism-related genes (Figure 1E). Importantly,
several Trem2-dependent genes control cholesterol transport
and metabolism, including Apoe, Apoc1, Ch25h, Lipa, Nceh1,
Npc2, and Soat1. Gene set enrichment analysis also showed
that DAM genes (Keren-Shaul et al., 2017) were significantly up-
regulated with CPZ in Trem2+/+ and Trem2+/– microglia at both
time points but attenuated in Trem2–/– microglia (Figures 1C,
1F, and S1F). This suggests that chronic demyelination induces
expression changes related to those observed in 5XFAD and
SOD1 microglia (Keren-Shaul et al., 2017).
Keren-Shaul et al. (2017) described the microglial transition
from homeostasis into DAM as a two-step process: a
TREM2-independent transition to DAM stage 1, followed by a
TREM2-dependent transition to DAM stage 2. We compared
our TREM2 CPZ dataset with published homeostatic, DAM 1,
and DAM 2 gene profiles. Homeostatic genes P2ry12 and
Tmem119 were significantly downregulated in Trem2+/+ and
Trem2+/– (FDR < 0.01) but not Trem2–/– microglia in response
to CPZ (Figure 1G; genotype-diet interaction p < 0.05; Data
S1). Although described as a TREM2-independent phenome-
non in an AD model, we observed reduced induction of DAM
1 genes such as Apoe (interaction p < 0.001) and Fth1 (interac-
tion p < 0.005) in Trem2–/– compared with Trem2+/+ and
Trem2+/– microglia following CPZ treatment (Figure 1H). Apoe
expression was 8-fold higher with CPZ treatment of Trem2+/+
and Trem2+/– mice after 12 weeks but was attenuated in
Trem2–/– mice, similar to DAM 2 genes, such as Axl (interaction
p < 0.05) and Cd9 (interaction p < 0.001) (Figure 1I; Data S1).
Our data indicate a role of TREM2 in endolysosomal function
and lipid metabolism, with a clear implication of cholesterol
transport and metabolism. Additionally, they suggest that
CPZ elicits a damage-associated microglia state that fails to
be initiated in Trem2–/– microglia.
TREM2 Deficiency Reduces Age-Dependent Conversion
to DAM
Microglial dysfunction is a hallmark of the aging brain, as cells
accumulate excessive myelin debris and acquire a DAM tran-
scriptional state (Mecca et al., 2018; Safaiyan et al., 2016;
Keren-Shaul et al., 2017). We sorted microglia derived from
young (2-month-old) and aged (15- to 17-month-old) wild-type
and Trem2–/– mice to assess whether aged Trem2–/– microglia
exhibit a DAM-like state. Aged wild-type microglia downregu-
lated a number of homeostatic genes and upregulated DAM 1
and DAM 2 genes compared to young microglia, albeit to a lesser
degree than modulation in the CPZ model (Figures S1G and
S1H). As with CPZ, activation of the DAM 2 gene set was atten-
uated in aged Trem2–/– microglia compared with aged controls
(Figures S1G and S1H). This effect was most striking for Lpl
and Spp1, consistent with the Trem2 dependency of the DAM
2 profile exhibited in 5XFAD microglia (Keren-Shaul et al.,
(C) Heatmap showing the relative gene expression profiles (normalized to zero mean and unit variance) of the top 15 up- and downregulated genes per cluster.
(D) Expression profiles for selected marker genes plotted as normalized counts per cell. Left legend denotes upregulated (up arrow) versus downregulated (down
arrow) marker genes in indicated clusters.
See also Figure S2 and Data S2.
2017). Unlike with CPZ, microglia from aged mice only showed
a mild increase in expression of cholesterol metabolism-related
genes compared with young mice, suggesting that aged micro-
glia face only a minor cholesterol burden. Thus, gene expression
changes in microglia derived from CPZ-challenged Trem2–/–
mice are also present in aged Trem2–/– microglia, although the
latter are less striking.
scRNA-Seq Confirms that Trem2–/– Microglia Exhibit
Attenuated Transition to DAM upon Demyelination
To determine if the population-based CPZ-induced transcrip-
tional changes observed were homogeneous across microglia,
we conducted scRNA-seq on CD11b+/CD45low microglia iso-
lated from CPZ-treated mice (Figures S2A–S2D). We used a
graph-based clustering approach (Xu and Su, 2015) to divide
3,023 cells into eight microglia sub-populations (Figure 2A),
each accounting for 2%–19% of analyzed cells (Figure 2B,
left). Quantification of cluster membership across groups (Fig-
ure 2B, middle) identified two treatment- and genotype-depen-
dent clusters (4 and 8). Cluster 4 mainly contained microglia
exposed to CPZ with at least one copy of Trem2 (1% of
Trem2+/+ controls; 16%–19% of Trem2+/+ CPZ and Trem2+/–
CPZ; 1.5% of Trem2–/– CPZ). Microglia from cluster 8 were
largely absent in Trem2+/+ controls (< 2.3%), mildly increased
in the Trem2+/+ CPZ and Trem2+/– CPZ mice (10%), and
most abundant in the Trem2–/– CPZ mice (20%) (Data S2).
Cluster 3 consists of very few cells and is discussed in STAR
Methods.
To further characterize cells within each cluster, we identified
marker genes with cluster-specific over- and under-expression
(Data S2). Relative expression of the top 15 up- and downregu-
lated genes per cluster confirmed that these clusters are distinct
(Figure 2C), although genes exclusive to a single cluster are rare.
Consistent with bulk RNA-seq data, the top upregulated marker
genes in the Trem2+/+ CPZ and Trem2+/– CPZ-enriched cluster
4 are lysosomal genes (e.g., Ctsb, Ctsd, Ctsz) and genes
involved in lipid metabolism (e.g., Apoe, Lpl) (Figure 2D). The
downregulated marker genes in cluster 4 include homeostatic
genes, such as P2ry12 and Tmem119, suggesting microglia in
this cluster are in a reactive state (Figures 2D and S2E). Cluster
8 similarly consisted of cells with upregulated lysosome- and
lipid metabolism-related genes, such as Ctsb, Ctsd, and Apoe,
but their expression was lower than that seen in cluster 4 (Figures
2D and S2F). About 20% of the total microglia population upre-
gulate the above genes; thus not all Trem2+/+ and Trem2+/–
microglia alter gene expression with CPZ.
Consistent with our bulk microglial findings, the transcriptome
from Trem2+/+ and Trem2+/– CPZ cluster 4 largely contained up-
regulated marker genes that previously have been characterized
in DAM 2 expression (Keren-Shaul et al., 2017), including Lgals3,
Cd63, Spp1, Cst7, Cd68, Capg, and Fth1 (Figure 2D). To deter-
mine if clusters 4 and 8 relate to DAM stages, we aggregated
the expression of established DAM stage 1 and 2 marker genes
Neuron 105, 837–854, March 4, 2020 841
(Data S2) for each cluster. Cluster 4 showed the highest enrich-
ment for DAM 2 gene expression, followed by cluster 8 (Fig-
ure 2B, right), suggesting that Trem2+/+ CPZ and Trem2+/– CPZ
microglia exhibit a DAM 2-like transition in response to CPZ
that is attenuated in Trem2–/– CPZ microglia. As in Keren-Shaul
et al. (2017), DAM 1 genes were upregulated in Trem2+/+
CPZ, Trem2+/– CPZ, and Trem2–/– CPZ microglia, although
Trem2–/– showed reduced upregulation. Thus, upon chronic
demyelination, Trem2–/– microglia appear to be arrested in their
transition to expression states seen in Trem2+/+ and Trem2+/– mi-
croglia and are unable to upregulate transcription of DAM genes,
including lysosome- and lipid metabolism-related genes, similar
to reports of Trem2–/–:5XFAD microglia (Keren-Shaul et al.,
2017). Proper conversion to reactive states during chronic
demyelination may enable lysosomal degradation and/or efflux
of myelin lipids.
TREM2 Deficiency Causes Neuronal Damage during
Chronic Demyelination
Neurite dystrophy occurs in CPZ-treated Trem2–/– mice (Cantoni
et al., 2015; Poliani et al., 2015). Likewise, we found accumula-
tion of dystrophic APP-positive puncta in the hippocampus
and corpus callosum of Trem2–/– mice after 5 or 12 weeks of
CPZ (Figures S3A and S3B). Quantification of APP-positive
dystrophic neurite puncta number, intensity, or area showed a
significant interaction between genotype and treatment for the
5 and/or 12 week CPZ diet (Figure S3C). Accordingly, plasma
neurofilament-light chain (Nf-L) levels were higher in CPZ-
treated Trem2–/– mice (Figure 3A; 12 week CPZ treatment-geno-
type interaction p < 0.001, two-way ANOVA). Similarly, aged
Trem2–/– plasma showed higher Nf-L levels (Figure S3D; age-ge-
notype interaction p < 0.05, two-way ANOVA), indicating that
TREM2 is neuroprotective.
TREM2 Deficiency Causes CE Accumulation in the Brain
Lipid metabolism-related genes are strongly induced upon
chronic demyelination in wild-type but not Trem2–/– microglia
(Figures 1E and 2D), including seven genes coding for proteins
directly involved in cholesterol transport (Apoe, Apoc1), hydroly-
sis of CE in lysosomes (Lipa), egress of cholesterol from lyso-
somes (Npc2), CE synthesis and storage in lipid droplets
(Soat1), CE hydrolysis in lipid droplets (Nceh1), and 25-hydroxyl-
ation (Ch25h). Thus, both intracellular and extracellular choles-
terol transport may be defective in Trem2–/– microglia with
CPZ. To test this, we conducted liquid chromatography-mass
spectrometry (LCMS) analysis of lipids from coronal forebrain
sections containing the corpus callosum. Brain lipidomic profiles
of the three genotypes under control diet were similar (Figure 3B;
Data S3). Upon 5 week CPZ, minimal changes were detected,
with an enhancement in CE and oxCE levels in all three geno-
types (Figure S3E). With 12 week CPZ, CE and oxCE lipid spe-
cies were significantly elevated in Trem2–/– brain compared
with Trem2+/+ and Trem2+/– brain (Figures 3B–3D; FDR < 0.05;
interaction p < 0.01 for 12 week CPZ, two-way ANOVA). Further
comparison revealed that Trem2–/– brain significantly accumu-
lated CE species containing polyunsaturated fatty acids, such
as CE22:6 (docosahexaenoic acid) and CE20:4 (arachidonic
acid) (Figure 3C; interaction p < 0.0001, two-way ANOVA).
842 Neuron 105, 837–854, March 4, 2020
CE22:6 showed the most striking increase in Trem2–/– brain
with chronic demyelination, upward of 38-fold compared with
Trem2–/– control brain and 2.5-fold compared with Trem2+/+
12 week CPZ brain (Figure 3C). Likewise, oxCE species, previ-
ously reported only in atherosclerotic lesions (Choi et al., 2017;
Hutchins et al., 2011), were significantly enhanced in Trem2–/–
brain compared with Trem2+/+ with chronic demyelination,
although they were found at much lower levels than CE
(Figure 3D; interaction p < 0.01 for 12 week CPZ, two-way
ANOVA). Despite the significant accumulation of CE, brain
cholesterol levels remained unaltered (Figure 3E). Other neutral
lipids, such as triacylglycerol (TG), were elevated in Trem2–/–
brain upon CPZ treatment (Figure 3F; interaction p < 0.05, two-
way ANOVA). Ganglioside GM3d38:1 and d40:1 were also
increased in Trem2–/– brain with chronic demyelination (Fig-
ure 3G; interaction p < 0.05, two-way ANOVA), reminiscent of
lysosomal storage disorders. LCMS of mouse plasma did not
reveal any genotype- or CPZ-specific differences in cholesterol
and CE levels (Data S3; two-way ANOVA). These data indicate
that chronic demyelination causes a profound alteration of
cholesterol metabolism selectively in Trem2–/– CNS.
TREM2 Deficiency Causes CE Accumulation in Isolated
Microglia
The increase in CE found in Trem2–/– brain upon chronic demy-
elination may be primarily intracellular, resulting from phagocy-
tosis of cholesterol-rich myelin by brain phagocytes and storage
in lipid droplets after conversion of cholesterol to CE by ACAT.
Alternatively, if Trem2–/– phagocytes are unable to properly
engulf myelin debris, the latter may accumulate in the interstitial
space and/or the cerebral spinal fluid (CSF), potentially gener-
ating CE extracellularly. To test this, we developed cell-type-
specific lipidomics via isolation of microglia (CD11b+/CD45low)
and astrocytes (ACSA2+) using FACS (Figure S4A) and collected
CSF to assess circulating levels of lipids in the CNS.
As observed in forebrain, 12 week CPZ increased levels of
certain lipid species in microglia of all genotypes compared
with untreated genotype controls, but Trem2–/– microglia ex-
hibited more dramatic increases compared with Trem2+/+ and
Trem2+/– microglia (Figure 4A; Data S3; two-way ANOVA,
FDR < 0.05). There were no genotype effects in microglia without
CPZ. Lipid changes were unique to microglia, as astrocytes or
CSF did not display any genotype- or CPZ treatment-specific al-
terations (Figures 4B and 4C; two-way ANOVA, FDR < 0.05).
Lipids selectively increased in Trem2–/– microglia upon chronic
demyelination included many species dysregulated in the fore-
brain (e.g., CE20:4, CE22:6) (Figures 4D and 4E; interaction p <
0.001 for 12 week CPZ, two-way ANOVA), which were elevated
8- to 20-fold in Trem2–/– microglia upon 12 week CPZ diet
compared with controls. Myelin-enriched lipids (Podbielska
et al., 2011) or metabolites thereof, such as hexosylceramide
(HexCer) d18:1/22:0, d18:1/24:0, and d18:1/24:1, galactosylcer-
amide (GalCer) d18:1/22:0 and d18:1/24:0, ceramide (Cer)
d18:1/24:1, and sulfatide d18:1/24:0, d18:1/24:0h, and d18:1/
24:1, were significantly elevated upon 5 week CPZ treatment
without genotype effects, suggesting comparable microglial
uptake of myelin upon acute demyelination (treatment effect
p < 0.05 for 5 week CPZ, two-way ANOVA; Data S3). HexCer
 A C
Cholesteryl esters
Plasma neurofilament light chain
100 *** *** ** **
1000 ***
Plasma Nf-L (pg/mL)

10

ng lipid/ug protein
Trem2 +/+
100 Trem2 +/-
1
Trem2 -/-

0.1
10
Control 5 week CPZ 12 week CPZ
0.01
B Trem2 +/+ +/- -/- Z-score
CE 18:1 CE 20:4 CE 20:5 CE 22:6
Control CPZ Control CPZ Control CPZ
D
Oxidized cholesteryl esters
100
*** ** **** **

pg lipid/ug protein 10

0.1

0.01
CE oxoODE CE HODE CE HpODE CE oxoHETE

E F
Cholesterol Triacylglycerols
1000 1
*
ng lipid/ug protein
ng lipid/ug protein

0.1

100 0.01
Cholesterol TG 58:8/22:6

G Gangliosides
100
* * **
Trem2 +/+ Control
Trem2 +/+ 5 wk CPZ
pg lipid/ug protein

10 Trem2 +/+ 12 wk CPZ

Trem2 +/- Control
Trem2 +/- 5 wk CPZ
1 Trem2 +/- 12 wk CPZ
Trem2 -/- Control
Trem2 -/- 5 wk CPZ
0.1
Trem2 -/- 12 wk CPZ
GM3 d36:1 GM3 d38:1 GM3 d40:1

Figure 3. TREM2 Deficiency Causes CE Accumulation in the Brain

(A) Neurofilament light chain (Nf-L) levels in plasma isolated from Trem2+/+, Trem2+/–, and Trem2–/– mice with control, 5 week, or 12 week CPZ diet. Genotype-
treatment interaction with 12 week CPZ: ***p < 0.001, two-way ANOVA; n = 7–16 mice per condition. Data represent median ± max/min.
(B) Heatmap of lipids significantly altered by genotype and/or 12 week CPZ in mouse forebrain. Two-way ANOVA, FDR < 0.05; columns represent individual mice;
n = 3–7 mice per condition.
(C–G) Concentration of (C) CE, (D) oxidized CE, (E) cholesterol, (F) triacylglycerol (TG), and (G) GM3 lipid species from mouse forebrain with control, 5 week, or
12 week CPZ diet. Data represent median ± max/min. Two-way ANOVA fitted for 5 week and 12 week CPZ diet; genotype-treatment interaction with 12 week
CPZ: *p < 0.05, **p < 0.01, and ***p < 0.001. No significant genotype-treatment interaction with 5 week CPZ diet.
See also Figure S3 and Data S3.

Neuron 105, 837–854, March 4, 2020 843

 Microglia lipidomics Trem2 +/+ Control Trem2 +/- Control Trem2 -/- Control
A Trem2 +/+ +/- -/- Trem2 +/+ 5 wk CPZ Trem2 +/- 5 wk CPZ Trem2 -/- 5 wk CPZ
Ctl 5wk 12wk Ctl 5wk 12wk Ctl 5wk 12wk Z-score
Trem2 +/+ 12 wk CPZ Trem2 +/- 12 wk CPZ Trem2 -/- 12 wk CPZ

Microglia sterols Astrocyte sterols

D +
E
10 **** *** 10

pg lipid/cell

pg lipid/cell
1
1

0.1
0.1

0.01
0.01
10-5

10-10
0.001
Cholesterol CE 18:1 CE 20:4 CE 22:6 Cholesterol CE 18:1 CE 20:4 CE 22:6

F Microglia hexosylceramides G Astrocyte hexosylceramides

100 1000
*** **** **

100

fg lipid/cell
fg lipid/cell

10

10
B Astrocyte lipidomics
Trem2 +/+ +/- -/-
Ctl 5wk 12wk Ctl 5wk 12wk Ctl 5wk 12wk Z-score
1 1
d18:1/22:0 d18:1/24:0 d18:1/24:1 d18:1/22:0 d18:1/24:0 d18:1/24:1
H Microglia galactosylceramides I Astrocyte galactosylceramides
100 100
* *
10

fg lipid/cell
10
fg lipid/cell

1
0.1

0.01 0.1
d18:1/22:0 d18:1/24:0 d18:1/24:1 d18:1/22:0 d18:1/24:0 d18:1/24:1
J 100 Microglia ceramides
K Astrocyte ceramides
100
*** ***

10
fg lipid/cell
fg lipid/cell

10

C CSF lipidomics
Trem2 +/+ +/- -/- 1 0.1
Ctl 5wk 12wk Ctl 5wk 12wk Ctl 5wk 12wk d18:1/18:0 d18:1/24:0 d18:1/24:1 d18:1/18:0 d18:1/24:0 d18:1/24:1
L M
1000 Microglia sulfatides Astrocyte sulfatides
+ + 1000
** ** *
fg lipid/cell

100
fg lipid/cell

100

10
10

1 1
d18:1/24:0 d18:1/24:0h d18:1/24:1 d18:1/24:0 d18:1/24:0h d18:1/24:1

N Microglia gangliosides O Astrocyte gangliosides

100
* * 1000
fg lipid/cell

100
fg lipid/cell

10

10

1 1
Z-score GM3 d36:1 GM3 d38:1 GM3 d40:1 GM3 d36:1 GM3 d38:1 GM3 d40:1

Figure 4. TREM2 Deficiency Causes CE Accumulation in Isolated Microglia

(A) Heatmap of lipids significantly altered by treatment and/or genotype in microglia isolated from Trem2+/+, Trem2+/–, and Trem2–/– mouse brain upon control,
5 week, or 12 week CPZ diet. Two-way ANOVA, FDR < 0.05; columns represent individual mice; n = 6 mice per condition.
(B and C) Heatmap comparison of lipids detected in (B) astrocytes or (C) cerebral spinal fluid (CSF) isolated from Trem2 mice upon control, 5 week, or 12 week
CPZ diet. Columns represent individual mice; n = 5 or 6 mice per condition.
(D–O) Concentrations of significantly altered lipid species in (A) from microglia or astrocytes isolated from mouse brain with control, 5 week, or 12 week CPZ diet:
(D and E) sterols, including free cholesterol and CE, from (D) microglia or (E) astrocytes; (F and G) HexCer from (F) microglia or (G) astrocytes; (H and I) GalCer from

(legend continued on next page)

844 Neuron 105, 837–854, March 4, 2020

species, GalCer d18:1/24:0 and d18:1/24:1, Cer d18:1/18:0 and
d18:1/24:1, and sulfatides were further enriched in Trem2–/– mi-
croglia compared with Trem2+/+ upon 12 week CPZ (Figures 4F–
4M; genotype-treatment interaction p < 0.05, two-way ANOVA).
Additionally, GM3d36:1, GM3d38:1, and BMP36:2 were
elevated in Trem2–/– microglia upon 12 week CPZ diet (Figures
4N, 4O, S4B, and S4C), potentially indicating lysosomal dysfunc-
tion (Bissig and Gruenberg, 2013; Miranda et al., 2018). LCMS
analysis of CSF from the three genotypes with or without CPZ
did not show any lipid changes (Figures S4D–S4H), suggesting
that lipid accumulation observed in bulk forebrain tissue does
not reflect extracellular accumulation, although changes in inter-
stitial fluid cannot be ruled out. Thus, Trem2–/– microglia are able
to phagocytose myelin debris during demyelination but are un-
able to properly metabolize or mediate the efflux of myelin lipids.
Microglia sorted from the brain of aged Trem2+/+ and Trem2–/–
mice were also analyzed for lipid content, but no genotype-spe-
cific alteration of CE was found, consistent with minor alterations
in cholesterol metabolism-related genes compared with CPZ-
treated Trem2–/– microglia (Figures S1G and S1H).
APOE Deficiency Results in a Broader CE Accumulation
in the Brain
Apoe is one of the most striking TREM2-dependent damage-
associated microglia genes (Götzl et al., 2019; Krasemann
et al., 2017; Parhizkar et al., 2019) and is commonly believed
to be the main cholesterol transporter in the CNS. Additionally,
microglial responses in the Apoe–/– phenocopy those of
Trem2–/– in specific paradigms, such as phagocytosis of
apoptotic neurons (Krasemann et al., 2017). APOE deficiency
may thus also produce microglial CE accumulation with CPZ, re-
flecting reduced CNS cholesterol transport. However, unlike
TREM2, APOE is expressed in microglia and astrocytes, raising
the possibility that APOE deficiency may more broadly affect
brain cholesterol metabolism. We subjected Apoe+/+ and
Apoe–/– mice to a control or 12 week CPZ diet prior to LCMS an-
alyses of forebrain, isolated microglia, astrocytes, and CSF. Lack
of APOE generally increased brain CE levels (Figure 5A; Data S4).
Levels of CE18:1 and CE22:6 were increased by 2.7- and 4-fold,
respectively, in Apoe–/– forebrain relative to wild-type forebrain
with control diet. CPZ led to an increase in CE18:1, 20:4, and
22:6 species in Apoe–/– versus Apoe+/+ forebrain (Figure 5B;
main effect from two-way ANOVA, p < 0.001; FCs were 6.6,
1.4, and 6.7, respectively), and APOE deficiency significantly
exacerbated the treatment effects for CE18:1 and CE20:4 (geno-
type-treatment interaction p < 0.05). In addition, levels of BMP
40:8 and 44:12 were higher in the Apoe–/– forebrain, consistent
with lysosomal defects (Figure S5A; p < 0.05). There was a geno-
type-specific increase in all CE species in sorted microglia
(Figures 5C and 5D; Data S4; p < 0.0001, two-way ANOVA),
with 3.2-, 6.9-, and 2.5-fold increases in CE18:1, CE20:4, and
CE22:6, respectively, in Apoe–/– microglia relative to wild-type
microglia on control diet. With CPZ, changes were 2.8-, 4.2-,
(H) microglia or (I) astrocytes; (J and K) ceramides from (J) microglia or (K) astrocytes; (L and M) sulfatides from (L) microglia or (M) astrocytes; and (N and O)
gangliosides from (N) microglia or (O) astrocytes. Data represent median ± max/min. Two-way ANOVA fitted for 5 week and 12 week CPZ diet; genotype-
treatment interaction with 5 (+p < 0.05) or 12 week CPZ diet (*p < 0.05, **p < 0.01, and ***p < 0.001). Independent genotype or treatment effects noted in Data S3.
See also Figure S4.
and 2.3-fold, respectively, compared with wild-type microglia,
suggesting that CPZ increases the levels of most CE species
similarly in wild-type and Apoe–/– microglia, thus preserving ge-
notype differences. Genotype-specific increases in CE were pre-
sent in sorted astrocytes (Figures 5E and 5F; Data S4; p < 0.001,
two-way ANOVA), with 2.1- and 4.5-fold increases in CE18:1 and
CE22:6, respectively, in Apoe–/– astrocytes relative to wild-type
astrocytes with CPZ. Interaction between genotype and CPZ
was observed for CE18:1 (p < 0.05). Remarkably, no changes
were found for cholesterol in the forebrain or sorted glial cells
(Figures 5B, 5D, and 5F). Myelin lipids HexCer and sulfatides
were similarly increased in wild-type and Apoe–/– microglia
with 12 week CPZ, but there were no treatment effects in astro-
cytes (Figures S5B and S5C; Data S4). Unlike in Trem2–/– CSF,
CEs were elevated in Apoe–/– CSF, pointing to a widespread in-
crease of these sterols in mutant brain (Figures S5D and S5E).
Therefore, impaired cholesterol transport resulting from APOE
deficiency causes massive, primarily CPZ-independent, accu-
mulation of CE in forebrain, glial cells, and CSF. The fact that
Trem2–/– microglia exhibit similar CE storage and have lower
APOE levels suggests that the lipid phenotype stems from
cholesterol transport defects.
Myelin Sulfatide Binds TREM2 and Promotes
Downstream Signaling
Next, we sought to delineate the mechanisms underlying lipid
dysregulation in Trem2–/– cells in vitro. Myelin lipid accumulation
in mutant cells could result from either increased phagocytosis
or reduced lipid clearance post-phagocytosis. Because
Trem2–/– myeloid cells generally show reduced phagocytosis
(Kleinberger et al., 2014; Poliani et al., 2015), Trem2–/– microglia
lipid phenotypes are unlikely to reflect increased myelin phago-
cytosis. Prior to investigating myelin phagocytosis in Trem2–/–
cells, we assessed the ability of specific myelin lipids to bind
and signal via TREM2, which may in turn regulate the phagocytic
clearance of myelin.
TREM2 binds to various lipids, including phosphatidylcholine
(PC), sphingomyelin (SM), phosphatidylserine (PS), phosphati-
dylinositol (PI), and sulfatide (Kober et al., 2016; Sudom et al.,
2018; Wang et al., 2015), some of which (e.g., sulfatide) are
myelin enriched (Du and Grabowski, 2004). We generated lipo-
somes harboring candidate TREM2 ligands composed of 70
molar % PC with 30 molar % of a test lipid to assess TREM2
stimulation. Upon lipid ligand binding, TREM2 recruits SYK via
DAP12’s ITAM domain, leading to SYK phosphorylation (Hamer-
man et al., 2006). We stably overexpressed human DAP12 in
the presence or absence of human TREM2 (hTREM2) in
HEK293 cells (Figure S6A) and monitored phospho-SYK
(pSYK) levels as a readout for receptor activation, using a
TREM2 agonist antibody as a positive control. PS-, PI-,
GalCer-, and sulfatide-containing liposomes increased pSYK
levels in TREM2/DAP12 HEK293 cells (Figure 6A). To examine
lipid signaling via endogenous TREM2, we differentiated human
Neuron 105, 837–854, March 4, 2020 845
 A Forebrain lipidomics C Microglia lipidomics E Astrocyte lipidomics
Apoe +/+ -/- Apoe +/+ -/- Apoe +/+ -/-
Ctl 12wk Ctl 12wk Z-score Ctl 12wk Ctl 12wk Z-score Z-score
3
Ctl 12wk Ctl 12wk 3
Sulfatide d181/ CE161 Sulfatide d181/240h
LPC180 2 CE226 2
Sulfatide d181/240
PCO 180/20 CE182 Sulfatide d181/241
PE361 1 CE204 1
LacCerd181/180
PG160_181 CE181 HexCerd181/241
PG181/181 0 PE386 0
HexCerd181/240
Sulfatide d181/ PE406 SMd181/240
HexCerd181/160 1 PE364 1
PEP 180/181
PA180_204 PE384 Sulfatide d181/180
PA181/181 CE205
2 HexCerd181/180 2
PA160_181 LacCerd181/241 PE386
PA180_181 HexCerd181/180
3 PE364 3
DG181/81 HexCerd181/241 PC361
PI180_181 LacCerd181/180 PEP 180/204
DG180_181 HexCerd181/240 CE204
PS180_181 Sulfatide d181/240 TG 587/204
Sulfatide d181/ Sulfatide d181/241 TG 546/204
PEP 180/181 Sulfatide d181/160 Stearic acid
PI180_204 Sulfatide d181/180h Palmitic acid
PI160_181 Sulfatide d181/180 Palmitoleic acid
PI181/181 Sulfatide d181/240h PA181_181
PC361 Sulfatide d181/241h PC386
PC362 Cerd181/160 PI180_181
Cholesteryl glucos SMd181/160 TG 585/204
SMd181/241 HexCerd181/160 LPG180
Cholesterol PG160_181 GM3d361
PEP 181/204 PI204/204 Sulfatide d181/180h
Cerd181/240 PI180_181 DG180/204
GM3d361 LBPA181/181 Sulfatide d181/241h
GM3d341 PG180_204 PI181/181
GM3d401 Cerd181/241 TG 542/180
PC406 PI181/181 TG 523/181
PC386 LPEP 180 CE HpODE
BMP204/204 PC386 PA180_204
CE204 PC364 DG181/181
PS180_204 TG 587/204 CE205
SMd181/240 LPC204 CE226
SMd181/160 LPC181 CE181
SMd181/180 TG 524/181 CE161
CE182 LPE181 CE182
CE181 LPC226 LBPA181/181
CE161 LPG180 GM3d381
CE205 PEP 180/181 PCO 180/20
CE226 PC361 PI160_226
BMP226/226 GM3d381 PEP 160/226
TG564/204 GM3d361 LPC180
LPI160 PEP 180/182 LacCerd181/241
LPI180 TG 546/204 LPC260
Sitosteryl glucosi PI160_226 Cerd181/240
DG160_181 PG180_181 Cerd181/160

B Forebrain sterols D Microglia sterols F Astrocyte sterols

1000 * * 10 * 10 *
#### ### #### #### #### ### ###
ng lipid/mg tissue

100
pg lipid/cell

1 pg lipid/cell 1

10

1 0.1 0.1
Cholesterol CE 18:1 CE 20:4 CE 22:6 Cholesterol CE 18:1 CE 20:4 CE 22:6 Cholesterol CE 18:1 CE 20:4 CE 22:6

Apoe +/+ Control Apoe -/- Control

Apoe +/+ 12 wk CPZ Apoe -/- 12 wk CPZ

Figure 5. APOE Deficiency Causes CE Accumulation in the Brain, Sorted Microglia, and Astrocytes
(A) Heatmap of top 50 lipid species (ranked by ANOVA p value) altered by genotype and/or 12 week CPZ diet in Apoe+/+ and Apoe–/– mouse forebrain.
(B) Concentration of free cholesterol and CE species from Apoe+/+ and Apoe–/– mouse forebrain extracts with control or 12 week CPZ diet.
(C) Heatmap of top 50 lipid species (ranked by ANOVA p value) altered by genotype and/or 12 week CPZ diet in Apoe+/+ and Apoe–/– sorted microglia.
(D) Concentration of free cholesterol and CE species from Apoe+/+ and Apoe–/– sorted microglia with control or 12 week CPZ diet.
(E) Heatmap of top 50 lipid species (ranked by ANOVA p value) altered by genotype and/or 12 week CPZ diet in Apoe+/+ and Apoe–/– sorted astrocytes.
(F) Concentration of free cholesterol and CE species from Apoe+/+ and Apoe–/– sorted astrocytes with control or 12 week CPZ diet.
In (B), (D), and (F), data represent median ± max/min (n = 6). Two-way ANOVA; genotype effects: ###p < 0.001 and ####p < 0.0001; genotype-treatment in-
teractions: *p < 0.05. Treatment effects are noted in Data S4. See also Figure S5.

peripheral blood monocytes into macrophages (Figure S6B). (RU) = 704, and KD = 5.6 mM, RU = 631, respectively (Figures
Cells showed liposome dose-dependent increases in pSYK 6D and 6E). In comparison, hTREM2 R47H showed reduced
levels selectively for sulfatide and PS (Figure 6B). Increased affinity and lower binding response (i.e., RU) for sulfatide
pSYK response to sulfatide liposomes was abolished by addition and PS liposomes: KD = 20 mM, RU = 191, and KD = 14 mM,
of recombinant hTREM2 but not hTREM1 ECD, confirming RU = 267, respectively, suggesting ligand specificity (Figures
TREM2 binding and signaling specificity (Figure 6C). 6D and 6E). Lower binding response was due to a faster off-
Certain TREM2 LOAD variants, including R47H, show rate of the interaction, leading to shorter residency of mutant
reduced affinity to lipids (Kober et al., 2016; Ulland and TREM2 on the lipid surface. Decreased affinity and response
Colonna, 2018; Wang et al., 2015). We characterized the bind- values observed with the R47H variant were accounted for
ing affinity and kinetics of sulfatide and PS to the ECD of wild- by 5-fold (sulfatide) and 2.4-fold (PS) faster off-rates and rela-
type and mutant TREM2 R47H (Figure S6C) with surface plas- tively similar on-rates (Figures S6D–S6G). These data show
mon resonance. hTREM2 exhibited similar binding affinity and that sulfatide binds and signals via TREM2 and that the
response at the highest analyte concentration to sulfatide- R47H LOAD variant is significantly impaired in its interaction
and PS-containing liposomes: KD = 6.8 mM, response units with this lipid.

846 Neuron 105, 837–854, March 4, 2020

A Liposome stimulation B Liposome titration
8 TREM2/DAP12 HEK293 cells 6
human macrophage
TREM2/DAP12
Sulfatide
6 DAP12 PS
pSyk FOB

4 PI

pSyk FOB
GalCer
4 IgG3
TREM2 agonist
2
2

0 0
-10 -5 0

t
is
PC

ol

SM

PE

PS

PI

er

3
tid

on
er

C
Log [mg/mL]

Ig
al

lfa
st

ag
G
le

Su

2
ho

EM
C

TR
C TREM2 ECD competition D Sulfatide liposome binding E PS liposome binding
8 human macrophage
* 800 800
* hTREM2 hTREM2
Binding response (RU)

Binding response (RU)

6 Stimulated hTREM2 R47H
600 KD 7uM 600 hTREM2 R47H
pSyk FOB

TREM2-ECD KD 6uM
4 400
TREM1-ECD 400

2 200 200 KD 14uM

KD 20uM
0 0 0
0 5 10 15 0 5 10 15
e

PS

PC

r
ffe
tid

TREM2 concentration (uM) TREM2 concentration (uM)

Bu
lfa
Su

5ng/mL M-CSF
F G
Vehicle pHrodo-myelin pHrodo-myelin phagocytosis
1.5
5ng/mL M-CSF
(normalized to WT 180 min)
Trem2 +/+

pHrodo intensity / cell

Trem2 +/+
1.0 Trem2 -/-

0.5
*
Trem2 -/-

0.0
0 50 100 150
Time (minutes)

pHrodo-myelin/cell mask/Hoechst

50ng/mL M-CSF
H I
Vehicle pHrodo-myelin pHrodo-myelin phagocytosis
50ng/mL M-CSF
1.5
(normalized to WT 180 min)
Trem2 +/+

pHrodo intensity / cell

Trem2 +/+
1.0 Trem2 -/-

0.5
Trem2 -/-

0.0
0 50 100 150
Time (minutes)

pHrodo-myelin/cell mask/Hoechst

(legend on next page)

Neuron 105, 837–854, March 4, 2020 847

High M-CSF Rescues Myelin Phagocytosis Defects in
Trem2–/– BMDMs
Because TREM2 binds to myelin lipids such as sulfatide, Trem2–/–
cells may be impaired in myelin uptake, although our in vivo CPZ
data suggest that myelin lipid uptake is unchanged in Trem2–/– mi-
croglia. To evaluate acute myelin uptake in vitro, we treated
Trem2+/+ and Trem2–/– BMDMs with pHrodo-conjugated myelin,
after verifying loss of TREM2 protein in null cells (Figures S6H
and S6I). At low concentrations of pro-survival factor M-CSF
(5 ng/mL), pHrodo-myelin phagocytosis was reduced in Trem2–/–
BMDMs compared with Trem2+/+, consistent with previous
studies (Poliani et al., 2015) (Figures 6F and 6G). However, at
high concentrations of M-CSF (50 ng/mL), pHrodo-myelin phago-
cytosis was comparable for both genotypes (Figures 6H and 6I).
Thus, high levels of M-CSF may provide compensatory upregula-
tion of phagocytic pathways in Trem2–/– BMDMs and reveal that
the phagocytosis defects in TREM2-deficient cells can be context
dependent. Importantly, although they are obtained from BMDMs,
these data do not support a mechanism by which increased myelin
binding or phagocytosis causes lipid accumulation in Trem2–/– mi-
croglia in vivo, prompting further mechanistic analyses.
CE Accumulation in Trem2–/– Cells Is Rescued by ACAT1
Inhibitor and LXR Agonist In Vitro
Because Trem2–/– BMDMs can normally phagocytose myelin in
high M-CSF, we tested if accumulation of neutral lipids, including
CE, occurs downstream of myelin phagocytosis in null cells.
BMDMs were treated with 25 mg/mL myelin over 48 h, then
stained with Nile red to assess neutral lipid storage with fluores-
cence microscopy. Trem2–/– BMDMs exhibited neutral lipid
accumulation upon myelin treatment, as shown by an 3-fold in-
crease in Nile red stain (Figures 7A and 7B). Lipidomics also re-
vealed prominent genotype-specific accumulation of CE18:2,
CE20:4, and CE20:5 (Figure 7C; Data S5; p < 0.01, two-way
ANOVA). Cholesterol and various species of TG, diacylglycerol
(DG), and myelin-derived HexCer also accumulated in Trem2–/–
BMDMs (Figure 7C). These changes were reminiscent of those
observed in Trem2–/– microglia in vivo with CPZ diet (Figure 4).
Next, we tested if CE accumulation in Trem2–/– BMDMs is due
to excess storage of myelin cholesterol by ACAT1, which con-
verts free cholesterol to CE in the endoplasmic reticulum.
Because myelin debris contain high levels of cholesterol but
low levels of CE, CE can accumulate in lipid droplets only if
myelin-derived cholesterol escapes from lysosomes post-
phagocytosis. BMDMs were treated for 48 h with myelin and
co-dosed with ACAT1 inhibitor (500 nM K604) (Ikenoya et al.,
2007). LCMS analysis showed that K604 selectively rescues
the accumulation of most CE species in response to myelin in
both genotypes, suggesting myelin cholesterol is indeed stored
as an esterified form in lipid droplets (Figure 7C; Data S5).
We subsequently asked if CE accumulation in Trem2–/–
BMDMs is specific to myelin phagocytosis or if other physiolog-
ically relevant uptake mechanisms, such as endocytosis of
CE-containing lipoproteins, may cause a similar CE increase.
First, we tested if oxidized low-density lipoprotein (oxLDL) binds
and stimulates TREM2, using low-density lipoprotein (LDL) as a
control. Using our hTREM2/hDAP12 HEK293 line, we found that
oxLDL elevates pSYK levels (Figure S7A). A dose-dependent
increase in pSYK levels was found in oxLDL-treated human mac-
rophages (Figure S7B). The increase trended toward attenuation
by pre-incubating oxLDL with high concentrations of recombi-
nant hTREM2 ECD at 9 mM (p = 0.1, two-way ANOVA, Tukey
test) but not at lower concentrations such as those used in lipo-
some/hTREM2 competition experiments (3 mM) (Figures S7C
and 6C), consistent with the fact that oxLDL is known to bind
and stimulate multiple immune receptors (Choi et al., 2017).
When treated chronically with 50 mg/mL oxLDL, Trem2–/–
BMDM exhibited an increase in Nile red stain (Figures S7D and
S7E), which did not stem from increased oxLDL uptake by
Trem2–/– BMDMs, as indicated by comparable internalization
of DiI-labeled oxLDL (Figure S7F). By LCMS, certain species of
CE, TG, HexCer, and cholesterol displayed genotype-specific in-
creases in both vehicle and oxLDL-treated conditions (Fig-
ure S7G; Data S5). K604 reduced levels of CE20:5 and CE22:6
in Trem2–/– BMDMs upon oxLDL exposure (Figure S7G; p <
0.05, Student’s t test), albeit less substantially than seen in

Figure 6. Myelin Sulfatide Binds TREM2 and Promotes Downstream Signaling

(A) Phospho-SYK (pSYK) fold change in TREM2/DAP12-HEK293 cells stimulated with indicated liposomes, normalized to buffer control (dotted line) and
compared with TREM2 agonist antibody and isotype control. N R 2 experimental replicates from two or more averaged technical replicates. PE, phosphati-
dylethanolamine. Data represent mean ± SEM.
(B) Liposome titration curve of pSYK fold changes in human macrophage cells from two to four donors upon stimulation with indicated test lipid, normalized to
buffer control, and compared with TREM2 antibody and isotype control.
(C) Liposome stimulation of indicated test lipid in human macrophage cells from four or five donors with liposomes only (stimulated) or liposomes with 3 mM
recombinant TREM2- or TREM1-extracellular domain (ECD) protein normalized to buffer control (dotted line). *p < 0.05, two-way ANOVA, Tukey test. Data
represent median ± max/min.
(D and E) Surface plasmon resonance binding response of increasing concentrations of wild-type (gray) and mutant R47H (orange) hTREM2 protein to (D) 30%
sulfatide/70%PC or (E) 30% PS/70% PC 100 nm liposomes.
(F) Trem2+/+ and Trem2–/– BMDM vehicle or pHrodo-myelin (red) phagocytosis with 5 ng/mL M-CSF. Green, cell membrane stain; blue, Hoechst. Scale
bar: 20 mm.
(G) Intensity of pHrodo-myelin (5 mg/mL) phagocytosis in Trem2+/+ and Trem2–/– BMDMs with 5 ng/mL M-CSF. Data represent mean ± SEM (n = 3 biological
replicates); *p < 0.05 comparing Trem2+/+ and Trem2–/– AUC, two-tailed t test.
(H) Trem2+/+ and Trem2–/– BMDM vehicle or pHrodo-myelin (red) phagocytosis with 50 ng/mL M-CSF. Green, cell membrane stain; blue, Hoechst. Scale
bar: 20 mm.
(I) Intensity of pHrodo-myelin (5 mg/mL) phagocytosis in Trem2+/+ and Trem2–/– BMDM with 50 ng/mL M-CSF. Data represent mean ± SEM (n = 3 biological
replicates).
See also Figure S6.

848 Neuron 105, 837–854, March 4, 2020

 A BMDM Nile Red/DAPI B BMDM
Vehicle Myelin

*
Trem2 +/+
2500

Nile Red Spot Area

2000 Vehicle
Myelin
1500

1000
Trem2 -/-

500

0
Trem2 +/+ Trem2 -/-

C BMDM Sterols
CE 18:1 CE 18:2 CE 20:4 CE 20:5
## ## ##

10000 1000 100000 10000

** ** *
Adjusted Mean Abundance

Adjusted Mean Abundance

Adjusted Mean Abundance

Adjusted Mean Abundance
** 10000 1000 Vehicle
1000 100
**
1000 100 Myelin
100 10
100 10
Myelin + K604

10 1 10 1
+/+ -/- +/+ -/- +/+ -/- +/+ -/-

Triacyglycerols
CE 22:6 Cholesterol TG (52:3/18:1) TG (54:2/18:0)
### ###
###
10 10
Adjusted Mean Abundance

Adjusted Mean Abundance

100000 10000
**
Adjusted Mean Abundance

Adjusted Mean Abundance

10000
**
1000
1000 1 1
100

10
Interaction p-value: 0.017
1 100 0.1 0.1
+/+ -/- +/+ -/- +/+ -/- +/+ -/-

Diacyglycerols
DG (16:0/18:1) DG (18:0/18:1) DG (18:0/20:4) DG (18:1/18:1)
# ## ### #

10 10 100 10
Adjusted Mean Abundance

Adjusted Mean Abundance

Adjusted Mean Abundance

Adjusted Mean Abundance

10 *
1 1 1

0.1 0.1 0.1 0.1

+/+ -/- +/+ -/- +/+ -/- +/+ -/-

Hexosylceramides
HexCer (d18:1/16:0) HexCer (d18:1/18:0) HexCer (d18:1/24:0) HexCer (d18:1/24:1)
# ### ### ###

10 10 10 100
Adjusted Mean Abundance

Adjusted Mean Abundance

Adjusted Mean Abundance

Adjusted Mean Abundance

1 10

1 1

0.1 1

Interaction p-value: 0.029

0.1 0.01 0.1 0.1
+/+ -/- +/+ -/- +/+ -/- +/+ -/-
D iMG Sterols
CE 18:1 CE 20:4 CE 22:6 Cholesterol
# # ##

10000 10000
*
Adjusted Mean Abundance

* * *
Adjusted Mean Abundance
Adjusted Mean Abundance

Adjusted Mean Abundance

* * * * Vehicle
** * 10000 ** 10000 **
1000 Myelin
1000 1000 Myelin + K604
1000

100
100 100
Myelin + GW

10 10 10 100
+/+ -/- +/+ -/- +/+ -/- +/+ -/-

Figure 7. TREM2 Deficiency-Associated CE Accumulation Is Rescued by ACAT1 Inhibitor and LXR Agonist In Vitro
(A) Nile red stain of neutral lipids in BMDMs cultured from Trem2+/+ and Trem2–/– mice, treated with vehicle or 25 mg/mL purified myelin for 48 h. Scale bar: 50 mm.
(B) Quantification of total spot area of Nile red stain in vehicle or 25 mg/mL myelin-treated Trem2+/+ and Trem2–/– BMDMs. Data represent mean + SEM (n = 5
biological replicates); *p < 0.05; one-tailed t test for comparison between Trem2+/+ with myelin versus Trem2–/– with myelin.
(C and D) Quantification of Trem2+/+ and Trem2–/– (C) sterols, TG, diacylglycerols (DG), and HexCer from cultured BMDMs and (D) sterols from cultured iPSC-
derived microglia (iMG) treated with vehicle, myelin, or myelin with ACAT1 inhibitor K604 (500 nM) or LXR agonist GW3965 (10 mM) for 48 h. Batch-adjusted mean

(legend continued on next page)

Neuron 105, 837–854, March 4, 2020 849

myelin uptake experiments (Figure 7C), without altering choles-
terol or other lipid levels (Figure S7G; Data S5). In contrast to
the myelin uptake paradigm, ACAT1 is responsible for only a
fraction of CE accumulation in oxLDL-treated Trem2–/– BMDMs,
suggesting that a pool of CE accumulates in organelles other
than lipid droplets, likely lysosomes.
Next, we sought a relevant in vitro model to extend our findings
to human cells. We differentiated human iPSCs into microglia
and established that the transcriptome is similar to that of pri-
mary human microglia (Figure S7H). We then genetically ablated
TREM2 using CRISPR (Figure S7I) and found that TREM2–/–
iPSC-derived microglia (iMG) have a 20% decrease in
pHrodo-myelin uptake relative to wild-type iMG after a 4 h incu-
bation (n = 4 technical replicates). Despite reduced myelin
phagocytosis, there was a genotype-specific increase in
CE20:4, CE22:6, and cholesterol (Figure 7D; Data S5; CE, p <
0.05; cholesterol, p < 0.01 [two-way ANOVA]). As in BMDMs,
the increase in CE, but not cholesterol, was abolished by the
ACAT1 inhibitor (Figure 7D; p < 0.05, Student’s t test).
Drugs such as LXR agonists can reduce CE stores by upregu-
lating the expression of ABCA1 and ABCG1 transporters (Moore
and Tabas, 2011). We thus treated iMG with the LXR agonist
GW3695 which enhanced the mRNA expression of ABCA1 (by
10.3 ± 0.2 and 14.8 ± 2.1 fold in wild-type and knockout iMG,
respectively) and ABCG1 (by 46.6 ± 5.0 and 53.8 ± 2.9 fold in
wild-type and knockout iMG, respectively; n = 2 biological repli-
cates). GW3695 rescued CE accumulation in myelin-treated
TREM2+/+ and TREM2–/– iMG (Figure 7D; Data S5; TREM2+/+
iMG, p < 0.01; TREM2–/– iMG, p < 0.05 [Student’s t test]). Our
data suggest that TREM2 deficiency causes cholesterol efflux
defects, leading to accumulation of an ACAT1 inhibitor-sensitive
pool of CE in human iMG.
DISCUSSION
Growing evidence indicates that TREM2 signaling is necessary
for normal microglial functions, including control of proliferation,
survival, energy metabolism, and gene expression (Deczkowska
et al., 2018; Ulland and Colonna, 2018; Yeh et al., 2017).
Although a link has been established between TREM2 and lipid
metabolism, previous studies have focused primarily on lipids
as candidate TREM2 ligands, in the form of either lipoprotein
particles or cell surface-exposed signals (Atagi et al., 2015;
Bailey et al., 2015; Sudom et al., 2018; Ulland and Colonna,
2018; Wang et al., 2016; Yeh et al., 2016). Further clues that
TREM2 mediates microglial lipid metabolism stemmed from mi-
croarray studies (Poliani et al., 2015). Recently, TREM2 was
shown to control blood cholesterol metabolism in obese mice
by modulating the macrophage transcriptome in adipose tissue,
further linking TREM2 to lipid metabolism (Jaitin et al., 2019).
Here, we combined FACS-based CNS cell type isolation with
RNA-seq, scRNA-seq, and lipidomics to establish a key role of
TREM2 signaling in the control of microglial cholesterol transport
and 95% confidence interval for each group, n = 3 biological replicates. Significant genotype effect (two-way ANOVA, comparing vehicle and myelin treatments in
Trem2+/+ and Trem2–/– cells): #p < 0.05, ##p < 0.01, and ###p < 0.001. Significant drug treatment effects within each genotype (Student’s t test): *p < 0.05
and **p < 0.01.
See also Figure S7 and Data S5.
850 Neuron 105, 837–854, March 4, 2020
and metabolism. TREM2 LOF caused robust intracellular accu-
mulation of a storage form of cholesterol, CE, which is also
observed in glial cells lacking APOE, the main cholesterol trans-
porter in the CNS. This function of TREM2 was revealed by a
chronic CPZ challenge that exposed microglia to myelin debris,
which are enriched for cholesterol and signaling lipids, such as
sulfatides. Lipid ligands are recognized by the TREM2 ECD
and mediate downstream signaling, which we hypothesize may
enable resolution of disease pathology within the damaged
CNS parenchyma.
As shown by scRNA-seq analysis of sorted microglia, CPZ
leads to TREM2-dependent gene expression changes in a sub-
set of microglia as part of a transition to DAM-like states related
to those found in other disease models (Keren-Shaul et al.,
2017). This transcriptional program encompasses a gene
‘‘toolbox’’ that may enable microglia to enhance not only their
chemotactic and phagocytic properties but also their lysosomal
degradative capacity and ability to facilitate intracellular and
extracellular cholesterol transport, including efflux from micro-
glia. Failure to induce this transcriptional program in Trem2–/–
microglia results in lipid accumulation rather than phagocytic de-
fects per se, suggesting that impaired lipid metabolism may
represent a critical aspect of TREM2 LOF in chronic diseases,
including Nasu-Hakola disease and LOAD. This maladaptive
state leads to neuritic dystrophy and Nf-L release, a phenome-
non generally reflective of neurodegeneration (Blennow and Zet-
terberg, 2018). Whether LOAD-associated LOF TREM2 variants
(e.g., R47H) phenocopy the complete LOF remains to be
established.
Trem2–/– macrophages and microglia exhibit a severe CE
storage disorder reminiscent of macrophage foam cells in
atherosclerotic lesions (Moore and Tabas, 2011). Cholesterol
dyshomeostasis in foam cells is generally associated with pro-
inflammatory responses (Choi et al., 2017; Heneka et al., 2018;
Moore and Tabas, 2011). Familial cases of atherosclerosis linked
to mutations in the LIPA gene (encoding the lysosomal acid
lipase) lead to a CE storage disorder, suggesting that accumula-
tion of this lipid is pathogenic, at least in lysosomes (Du and
Grabowski, 2004). CE accumulation may mediate toxicity via
enzymatic or non-enzymatic generation of oxidized metabolites
and lipid peroxidation products from the mono- or polyunsatu-
rated fatty acyl chains it harbors (Choi et al., 2017; Hutchins
et al., 2011). Most of the brain cholesterol is found in myelin,
and thus excessive release of myelin debris in disease situations
may cause cholesterol- or CE-induced toxicity, because choles-
terol cannot be efficiently catabolized and is deleterious to cells
at high concentrations. Our in vivo studies indicate that the
cholesterol burden imposed upon microglia by chronic demye-
lination leads to severe CE storage in Trem2–/– microglia, which
likely reflects impairment in the metabolic flux of cholesterol in
these cells. By failing to upregulate multiple genes involved in
cholesterol transport (Apoe, Apoc1, Npc2) and metabolism
(Lipa, Nceh1, Ch25h) upon chronic phagocytic challenge,
Trem2–/– cells accumulate intracellular CE to the detriment of mi-
croglial function. Reduced metabolic flux of CE may expose it to
oxidation, resulting in secondary accumulation of oxCE species,
as observed in Trem2–/– mouse brain. Whether oxCE species are
neurotoxic or contribute to inflammation as seen in atheroscle-
rotic lesions remains to be determined (Choi et al., 2017).
Our in vitro studies showed that post-phagocytic conversion
of myelin-derived cholesterol to CE by ACAT1, followed by its
storage in lipid droplets, underlies the molecular basis for CE
accumulation in Trem2–/– cells. Further mechanistic dissection
indicated that TREM2–/– iMG are defective in cholesterol efflux,
on the basis of the ability of an LXR agonist to fully alleviate
CE accumulation. Because TREM2 regulates microglial expres-
sion of APOE, an apolipoprotein that can function both in lipid
intake and lipid efflux (Shi and Holtzman, 2018; Tall, 2008), a
likely mechanism accounting for CE accumulation in Trem2–/–
cells is reduced efflux of cholesterol onto APOE-containing
lipoproteins. This is consistent with our Apoe–/– CPZ studies,
showing CE accumulation in mutant brain, sorted astrocytes,
and microglia. Trem2–/– and Apoe–/– mice thus share phenotypic
similarities, although APOE, unlike TREM2, is expressed in astro-
cytes and thus controls CNS cholesterol metabolism more
broadly. APOE isoform-specific differences in the handling of
lipids by glial cells may also account for the lower and higher
genetic risk of APOE2 and APOE4 (relative to APOE3) to
LOAD, respectively. The fact that human iPSC-derived APOE4/
E4 astrocytes were recently reported to accumulate endolysoso-
mal cholesterol suggests that cholesterol metabolism in glial
cells could be implicated in LOAD (Lin et al., 2018). As several
other LOAD genes expressed in microglia mediate lipid meta-
bolism (e.g., CLU, ABCA7, SORL1), alteration of cholesterol
and CE metabolism in microglia may be a unifying mechanism
accounting for disease risk, perhaps via impacts on Ab, tau,
inflammation, or other AD-relevant pathogenic processes.
The initial characterization of AD pathology by Dr. Alzheimer
noted accumulation of ‘‘adipose inclusions,’’ likely neutral lipids,
in glial cells from postmortem brain samples of patients with de-
mentia (Foley, 2010). More recent studies have indicated the rele-
vance of CE in AD, as it accumulates in postmortem brain tissue
from patients and in the brain of PSEN1/APP mutant mice (Astarita
et al., 2011; Chan et al., 2012; Morel et al., 2013) and other neuro-
degenerative models (Miranda et al., 2018). Importantly, genetic
ablation and pharmacological inhibition of ACAT1, which lowers
CE levels, is protective in amyloid and tauopathy models through
various mechanisms, including autophagy activation (Bhattachar-
yya and Kovacs, 2010; Bryleva et al., 2010; Di Paolo and Kim, 2011;
Hutter-Paier et al., 2004; Puglielli et al., 2001; Shibuya et al., 2015).
However, with the exception of Shibuya et al. (2014), these studies
primarily examined the role of ACAT1 in neurons and did not test if
ACAT1 inhibitors decrease CE burden in microglia. Brain-pene-
trant LXR agonists have also been considered as therapeutics
on the basis of anti-inflammatory and amyloid-lowering properties,
resulting from their ability to enhance cholesterol efflux. However,
they may present safety liabilities from broad impact on all cell
types and peripheral lipid metabolism (Moutinho and Landreth,
2017). Our study raises the possibility that enhancing TREM2
function may selectively alleviate cholesterol and CE burden in
myeloid cells, including microglia, potentially providing therapeutic
benefits in LOAD and other diseases associated with lipid
dysregulation.
STAR+METHODS
Detailed methods are provided in the online version of this paper
and include the following:
d KEY RESOURCES TABLE
d LEAD CONTACT AND MATERIALS AVAILABILITY
d EXPERIMENTAL MODEL AND SUBJECT DETAILS
B Animals
B Cell Lines
B Primary Cell Cultures
d METHOD DETAILS
B Immunofluorescence
B Neurofilament light detection
B Fluorescence activated cell sorting (FACS)
B RNA isolation, qPCR, QuantSeq library preparation,
and analysis
B Single cell RNAseq library preparation
B Single cell RNAseq data processing
B Single cell RNAseq cluster and expression analysis
B Brain sample collection and preparation for LCMS
analysis
B FACS lipid extraction
B CSF sample collection and preparation for LCMS
analysis
B LCMS analysis of lipids
B Liposome preparation
B pSYK AlphaLISA
B Recombinant purification of His tagged hTREM2 and
hTREM2 R47H ECD
B Lipid binding analysis using surface plasmon reso-
nance (SPR)
B Myelin purification and phagocytosis
B Lipid challenge experiments in BMDMs and human
iPSC-derived microglia
d QUANTIFICATION AND STATISTICAL ANALYSIS
d DATA AND CODE AVAILABILITY
SUPPLEMENTAL INFORMATION
Supplemental Information can be found online at https://doi.org/10.1016/j.
neuron.2019.12.007.
ACKNOWLEDGMENTS
We thank Rishi Rahkit and Nathan Moerke for help developing the pSYK assay
and Connie Ha for help with RNA-seq. We thank Kimberly Scearce-Levie,
Zachary Sweeney, Nga Bien-Ly, and Jeff Kessler for critical reading of the
manuscript. This study was funded by Denali Therapeutics.
AUTHOR CONTRIBUTIONS
K.M.M. and G.D.P. conceived the study. A.A.N., K.L., B.V.L., S.L., D.J.K., D.X.,
H.C., P.C.G.H., P.E.S., T.S., G.A., J.W.L., K.M.M., and G.D.P. designed exper-
iments. A.A.N., K.L., B.V.L., J.W., L.P., S.S.D., C.L., D.J.K., D.X., A.L., S.B.,
T.K.E., P.C.G.H., M.L., J.S., B.J.A., T.L., H.O.S., A.S., and G.A. performed ex-
periments. T.S. performed the QuantSeq analyses. S.L. performed the scRNA-
seq analyses. A.A.N., K.L., B.V.L., S.L., J.W., L.P., S.S.D., D.J.K., J.C.D.,
Neuron 105, 837–854, March 4, 2020 851
B.J.A., T.S., and G.A. analyzed data. L.P., T.L., J.W.L., P.E.S., S.B.P., and
R.J.W. edited the manuscript. A.A.N., K.L., B.V.L., S.L., T.S., G.A., K.M.M.,
and G.D.P. wrote the manuscript.
DECLARATION OF INTERESTS
All authors are paid employees and shareholders of Denali Therapeutics.
R.J.W. is a founder and member of the Board of Directors of Denali. Denali
has filed patent applications related to the subject matter of this paper.
Received: December 24, 2018
Revised: August 7, 2019
Accepted: December 4, 2019
Published: January 2, 2020
REFERENCES
Astarita, G., Jung, K.M., Vasilevko, V., Dipatrizio, N.V., Martin, S.K., Cribbs,
D.H., Head, E., Cotman, C.W., and Piomelli, D. (2011). Elevated stearoyl-
CoA desaturase in brains of patients with Alzheimer’s disease. PLoS ONE 6,
e24777.
Atagi, Y., Liu, C.C., Painter, M.M., Chen, X.F., Verbeeck, C., Zheng, H., Li, X.,
Rademakers, R., Kang, S.S., Xu, H., et al. (2015). Apolipoprotein E is a ligand
for triggering receptor expressed on myeloid cells 2 (TREM2). J. Biol. Chem.
290, 26043–26050.
Bailey, C.C., DeVaux, L.B., and Farzan, M. (2015). The triggering receptor ex-
pressed on myeloid cells 2 binds apolipoprotein E. J. Biol. Chem. 290,
26033–26042.
Bemiller, S.M., McCray, T.J., Allan, K., Formica, S.V., Xu, G., Wilson, G.,
Kokiko-Cochran, O.N., Crish, S.D., Lasagna-Reeves, C.A., Ransohoff, R.M.,
et al. (2017). TREM2 deficiency exacerbates tau pathology through dysregu-
lated kinase signaling in a mouse model of tauopathy. Mol. Neurodegener
12, 74.
Benjamini, Y., and Hochberg, Y. (1995). Controlling the false discovery rate: a
practical and powerful approach to multiple testing. J. R. Stat. Soc. Ser. A Stat.
Soc. 57, 289–300.
Bhattacharyya, R., and Kovacs, D.M. (2010). ACAT inhibition and amyloid beta
reduction. Biochim. Biophys. Acta 1801, 960–965.
Bissig, C., and Gruenberg, J. (2013). Lipid sorting and multivesicular endo-
some biogenesis. Cold Spring Harb. Perspect. Biol. 5, a016816.
Blennow, K., and Zetterberg, H. (2018). Biomarkers for Alzheimer’s disease:
current status and prospects for the future. J. Intern. Med. 284, 643–663.
Blondel, V.D., Guillaume, J.L., Lambiotte, R., and Lefebvre, E. (2008). Fast un-
folding of communities in large networks. J. Stat. Mech. 2008, P1008.
Bryleva, E.Y., Rogers, M.A., Chang, C.C., Buen, F., Harris, B.T., Rousselet, E.,
Seidah, N.G., Oddo, S., LaFerla, F.M., Spencer, T.A., et al. (2010). ACAT1 gene
ablation increases 24(S)-hydroxycholesterol content in the brain and amelio-
rates amyloid pathology in mice with AD. Proc. Natl. Acad. Sci. U S A 107,
3081–3086.
Cantoni, C., Bollman, B., Licastro, D., Xie, M., Mikesell, R., Schmidt, R., Yuede,
C.M., Galimberti, D., Olivecrona, G., Klein, R.S., et al. (2015). TREM2 regulates
microglial cell activation in response to demyelination in vivo. Acta
Neuropathol. 129, 429–447.
Carmona, S., Hardy, J., and Guerreiro, R. (2018). The genetic landscape of
Alzheimer disease. Handb. Clin. Neurol. 148, 395–408.
Chan, R.B., Oliveira, T.G., Cortes, E.P., Honig, L.S., Duff, K.E., Small, S.A.,
Wenk, M.R., Shui, G., and Di Paolo, G. (2012). Comparative lipidomic analysis
of mouse and human brain with Alzheimer disease. J. Biol. Chem. 287,
2678–2688.
Choi, S.H., Sviridov, D., and Miller, Y.I. (2017). Oxidized cholesteryl esters and
inflammation. Biochim. Biophys. Acta Mol. Cell. Biol. Lipids 1862, 393–397.
Chong, J., Soufan, O., Li, C., Caraus, I., Li, S., Bourque, G., Wishart, D.S., and
Xia, J. (2018). MetaboAnalyst 4.0: towards more transparent and integrative
metabolomics analysis. Nucl. Acids. Res. 46, 486–494.
852 Neuron 105, 837–854, March 4, 2020
Deczkowska, A., Amit, I., and Schwartz, M. (2018). Microglial immune check-
point mechanisms. Nat. Neurosci. 21, 779–786.
Di Paolo, G., and Kim, T.W. (2011). Linking lipids to Alzheimer’s disease:
cholesterol and beyond. Nat. Rev. Neurosci. 12, 284–296.
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, 15–21.
Doench, J.G., Fusi, N., Sullender, M., Hegde, M., Vaimberg, E.W., Donovan,
K.F., Smith, I., Tothova, Z., Wilen, C., Orchard, R., et al. (2016). Optimized
sgRNA design to maximize activity and minimize off-target effects of
CRISPR-Cas9. Nat. Biotechnol. 34, 184–191.
Du, H., and Grabowski, G.A. (2004). Lysosomal acid lipase and atheroscle-
rosis. Curr. Opin. Lipidol. 15, 539–544.
Durinck, S., Moreau, Y., Kasprzyk, A., Davis, S., De Moor, B., Brazma, A., and
Huber, W. (2005). BioMart and Bioconductor: a powerful link between biolog-
ical databases and microarray data analysis. Bioinformatics 21, 3439–3440.
Efthymiou, A.G., and Goate, A.M. (2017). Late onset Alzheimer’s disease ge-
netics implicates microglial pathways in disease risk. Mol. Neurodegener.
12, 43.
Fabregat, A., Jupe, S., Matthews, L., Sidiropoulos, K., Gillespie, M., Garapati,
P., Haw, R., Jassal, B., Korninger, F., May, B., et al. (2018). The Reactome
Pathway Knowledgebase. Nucleic Acids Res. 46 (D1), D649–D655.
Foley, P. (2010). Lipids in Alzheimer’s disease: A century-old story.
Biochimica. et. Biophysica. Acta. 1801, 750–753.
Götzl, J.K., Brendel, M., Werner, G., Parhizkar, S., Sebastian Monasor, L.,
Kleinberger, G., Colombo, A.V., Deussing, M., Wagner, M., Winkelmann, J.,
et al. (2019). Opposite microglial activation stages upon loss of PGRN or
TREM2 result in reduced cerebral glucose metabolism. EMBO Mol. Med.
11, e9711.
Guerreiro, R., Wojtas, A., Bras, J., Carrasquillo, M., Rogaeva, E., Majounie, E.,
Cruchaga, C., Sassi, C., Kauwe, J.S., Younkin, S., et al.; Alzheimer Genetic
Analysis Group (2013). TREM2 variants in Alzheimer’s disease. N. Engl. J.
Med. 368, 117–127.
Hamerman, J.A., Jarjoura, J.R., Humphrey, M.B., Nakamura, M.C., Seaman,
W.E., and Lanier, L.L. (2006). Cutting edge: inhibition of TLR and FcR re-
sponses in macrophages by triggering receptor expressed on myeloid cells
(TREM)-2 and DAP12. J. Immunol. 177, 2051–2055.
Hammond, T.R., Robinton, D., and Stevens, B. (2018). Microglia and the brain:
complementary partners in development and disease. Annu. Rev. Cell Dev.
Biol. 34, 523–544.
Heneka, M.T., McManus, R.M., and Latz, E. (2018). Inflammasome signalling in
brain function and neurodegenerative disease. Nat. Rev. Neurosci. 19,
610–621.
Hutchins, P.M., Moore, E.E., and Murphy, R.C. (2011). Electrospray MS/MS
reveals extensive and nonspecific oxidation of cholesterol esters in human pe-
ripheral vascular lesions. J. Lipid Res. 52, 2070–2083.
Hutter-Paier, B., Huttunen, H.J., Puglielli, L., Eckman, C.B., Kim, D.Y.,
Hofmeister, A., Moir, R.D., Domnitz, S.B., Frosch, M.P., Windisch, M., and
Kovacs, D.M. (2004). The ACAT inhibitor CP-113,818 markedly reduces amy-
loid pathology in a mouse model of Alzheimer’s disease. Neuron 44, 227–238.
Ikenoya, M., Yoshinaka, Y., Kobayashi, H., Kawamine, K., Shibuya, K., Sato,
F., Sawanobori, K., Watanabe, T., and Miyazaki, A. (2007). A selective
ACAT-1 inhibitor, K-604, suppresses fatty streak lesions in fat-fed hamsters
without affecting plasma cholesterol levels. Atherosclerosis 191, 290–297.
Ilicic, T., Kim, J.K., Kolodziejczyk, A.A., Bagger, F.O., McCarthy, D.J., Marioni,
J.C., and Teichmann, S.A. (2016). Classification of low quality cells from single-
cell RNA-seq data. Genome Biol. 17, 29.
Jaitin, D.A., Adlung, L., Thaiss, C.A., Weiner, A., Li, B., Descamps, H.,
Lundgren, P., Bleriot, C., Liu, Z., Deczkowska, A., et al. (2019). Lipid-associ-
ated macrophages control metabolic homeostasis in a Trem2-dependent
manner. Cell 178, 686–698.e14.
Jay, T.R., Hirsch, A.M., Broihier, M.L., Miller, C.M., Neilson, L.E., Ransohoff,
R.M., Lamb, B.T., and Landreth, G.E. (2017). Disease progression-dependent
effects of TREM2 deficiency in a mouse model of Alzheimer’s disease.
J. Neurosci. 37, 637–647.
Jiang, H., Lei, R., Ding, S.W., and Zhu, S. (2014). Skewer: a fast and accurate
adapter trimmer for next-generation sequencing paired-end reads. BMC
Bioinformatics 15, 182.
Jonsson, T., Stefansson, H., Steinberg, S., Jonsdottir, I., Jonsson, P.V.,
Snaedal, J., Bjornsson, S., Huttenlocher, J., Levey, A.I., Lah, J.J., et al.
(2013). Variant of TREM2 associated with the risk of Alzheimer’s disease.
N. Engl. J. Med. 368, 107–116.
Keren-Shaul, H., Spinrad, A., Weiner, A., Matcovitch-Natan, O., Dvir-
Szternfeld, R., Ulland, T.K., David, E., Baruch, K., Lara-Astaiso, D., Toth, B.,
et al. (2017). A unique microglia type associated with restricting development
of Alzheimer’s disease. Cell 169, 1276–1290.e17.
Kleinberger, G., Yamanishi, Y., Suárez-Calvet, M., Czirr, E., Lohmann, E.,
Cuyvers, E., Struyfs, H., Pettkus, N., Wenninger-Weinzierl, A., Mazaheri, F.,
et al. (2014). TREM2 mutations implicated in neurodegeneration impair cell
surface transport and phagocytosis. Sci. Transl. Med. 6, 243ra86.
Kober, D.L., Alexander-Brett, J.M., Karch, C.M., Cruchaga, C., Colonna, M.,
Holtzman, M.J., and Brett, T.J. (2016). Neurodegenerative disease mutations
in TREM2 reveal a functional surface and distinct loss-of-function mecha-
nisms. eLife 5, e20391.
Krasemann, S., Madore, C., Cialic, R., Baufeld, C., Calcagno, N., El Fatimy, R.,
Beckers, L., O’Loughlin, E., Xu, Y., Fanek, Z., et al. (2017). The TREM2-APOE
pathway drives the transcriptional phenotype of dysfunctional microglia in
neurodegenerative diseases. Immunity 47, 566–581.e9.
Kucukural, A., Yukselen, O., Ozata, D.M., Moore, M.J., and Garber, M. (2019).
DEBrowser: interactive differential expression analysis and visualization tool
for count data. BMC Genomics 20, 6.
Lee, C.Y.D., Daggett, A., Gu, X., Jiang, L.L., Langfelder, P., Li, X., Wang, N.,
Zhao, Y., Park, C.S., Cooper, Y., et al. (2018). Elevated TREM2 gene dosage
reprograms microglia responsivity and ameliorates pathological phenotypes
in Alzheimer’s disease models. Neuron 97, 1032–1048.e5.
Leyns, C.E.G., Gratuze, M., Narasimhan, S., Jain, N., Koscal, L.J., Jiang, H.,
Manis, M., Colonna, M., Lee, V.M.Y., Ulrich, J.D., and Holtzman, D.M.
(2019). TREM2 function impedes tau seeding in neuritic plaques. Nat.
Neurosci. 22, 1217–1222.
Leyns, C.E.G., Ulrich, J.D., Finn, M.B., Stewart, F.R., Koscal, L.J., Serrano,
J.R., Robinson, G.O., Anderson, E., Colonna, M., and Holtzman, D.M.
(2017). TREM2 deficiency attenuates neuroinflammation and protects against
neurodegeneration in a mouse model of tauopathy. PNAS 114, 11524–11529.
Liao, Y., Smyth, G.K., and Shi, W. (2014). featureCounts: an efficient general
purpose program for assigning sequence reads to genomic features.
Bioinformatics 30, 923–930.
Lin, Y.T., Seo, J., Gao, F., Feldman, H.M., Wen, H.L., Penney, J., Cam, H.P.,
Gjoneska, E., Raja, W.K., Cheng, J., et al. (2018). APOE4 causes widespread
molecular and cellular alterations associated with Alzheimer’s disease pheno-
types in human iPSC-derived brain cell types. Neuron 98, 1141–1154.e7.
Liu, L., Zhang, K., Sandoval, H., Yamamoto, S., Jaiswal, M., Sanz, E., Li, Z.,
Hui, J., Graham, B.H., Quintana, A., and Bellen, H.J. (2015). Glial lipid droplets
and ROS induced by mitochondrial defects promote neurodegeneration. Cell
160, 177–190.
Lun, A.T., Bach, K., and Marioni, J.C. (2016a). Pooling across cells to normalize
single-cell RNA sequencing data with many zero counts. Genome Biol. 17, 75.
Lun, A.T., McCarthy, D.J., and Marioni, J.C. (2016b). A step-by-step workflow
for low-level analysis of single-cell RNA-seq data with Bioconductor.
F1000Res. 5, 2122.
Martı́n, M.G., Pfrieger, F., and Dotti, C.G. (2014). Cholesterol in brain disease:
sometimes determinant and frequently implicated. EMBO Rep. 15,
1036–1052.
Masuda, T., Sankowski, R., Staszewski, O., Böttcher, C., Amann, L., Sagar,
Scheiwe, C., Nessler, S., Kunz, P., van Loo, G., et al. (2019). Author correction:
spatial and temporal heterogeneity of mouse and human microglia at single-
cell resolution. Nature 568, E4.
McCarthy, D.J., Campbell, K.R., Lun, A.T., and Wills, Q.F. (2017). Scater: pre-
processing, quality control, normalization and visualization of single-cell RNA-
seq data in R. Bioinformatics 33, 1179–1186.
Mecca, C., Giambanco, I., Donato, R., and Arcuri, C. (2018). Microglia and ag-
ing: the role of the TREM2-DAP12 and CX3CL1-CX3CR1 axes. Int. J. Mol. Sci.
19, E318.
Miranda, A.M., Lasiecka, Z.M., Xu, Y., Neufeld, J., Shahriar, S., Simoes, S.,
Chan, R.B., Oliveira, T.G., Small, S.A., and Di Paolo, G. (2018). Neuronal lyso-
somal dysfunction releases exosomes harboring APP C-terminal fragments
and unique lipid signatures. Nat. Commun. 9, 291.
Moore, K.J., and Tabas, I. (2011). Macrophages in the pathogenesis of athero-
sclerosis. Cell 145, 341–355.
Morel, E., Chamoun, Z., Lasiecka, Z.M., Chan, R.B., Williamson, R.L.,
Vetanovetz, C., Dall’Armi, C., Simoes, S., Point Du Jour, K.S., McCabe,
B.D., et al. (2013). Phosphatidylinositol-3-phosphate regulates sorting and
processing of amyloid precursor protein through the endosomal system.
Nat. Commun. 4, 2250.
Moutinho, M., and Landreth, G.E. (2017). Therapeutic potential of nuclear re-
ceptor agonists in Alzheimer’s disease. J. Lipid Res. 58, 1937–1949.
Muffat, J., Li, Y., Yuan, B., Mitalipova, M., Omer, A., Corcoran, S., Bakiasi, G.,
Tsai, L.H., Aubourg, P., Ransohoff, R.M., and Jaenisch, R. (2016). Efficient
derivation of microglia-like cells from human pluripotent stem cells. Nat.
Med. 22, 1358–1367.
Paloneva, J., Manninen, T., Christman, G., Hovanes, K., Mandelin, J.,
Adolfsson, R., Bianchin, M., Bird, T., Miranda, R., Salmaggi, A., et al. (2002).
Mutations in two genes encoding different subunits of a receptor signaling
complex result in an identical disease phenotype. Am. J. Hum. Genet. 71,
656–662.
Pandya, H., Shen, M.J., Ichikawa, D.M., Sedlock, A.B., Choi, Y., Johnson,
K.R., Kim, G., Brown, M.A., Elkahloun, A.G., Maric, D., et al. (2017).
Differentiation of human and murine induced pluripotent stem cells to micro-
glia-like cells. Nat. Neurosci. 20, 753–759.
Parhizkar, S., Arzberger, T., Brendel, M., Kleinberger, G., Deussing, M., Focke,
C., Nuscher, B., Xiong, M., Ghasemigharagoz, A., Katzmarski, N., et al. (2019).
Loss of TREM2 function increases amyloid seeding but reduces plaque-asso-
ciated ApoE. Nat. Neurosci. 22, 191–204.
Podbielska, M., Levery, S.B., and Hogan, E.L. (2011). The structural and func-
tional role of myelin fast-migrating cerebrosides: pathological importance in
multiple sclerosis. Clin. Lipidol. 6, 159–179.
Poliani, P.L., Wang, Y., Fontana, E., Robinette, M.L., Yamanishi, Y., Gilfillan, S.,
and Colonna, M. (2015). TREM2 sustains microglial expansion during aging
and response to demyelination. J. Clin. Invest. 125, 2161–2170.
Praet, J., Guglielmetti, C., Berneman, Z., Van der Linden, A., and Ponsaerts, P.
(2014). Cellular and molecular neuropathology of the cuprizone mouse model:
clinical relevance for multiple sclerosis. Neurosci. Biobehav. Rev. 47, 485–505.
Puglielli, L., Konopka, G., Pack-Chung, E., Ingano, L.A., Berezovska, O.,
Hyman, B.T., Chang, T.Y., Tanzi, R.E., and Kovacs, D.M. (2001). Acyl-coen-
zyme A: cholesterol acyltransferase modulates the generation of the amyloid
beta-peptide. Nat. Cell Biol. 3, 905–912.
Ransohoff, R.M. (2016). How neuroinflammation contributes to neurodegener-
ation. Science 353, 777–783.
Robinson, M.D., and Oshlack, A. (2010). A scaling normalization method for
differential expression analysis of RNA-seq data. Genome Biol. 11, R25.
Safaiyan, S., Kannaiyan, N., Snaidero, N., Brioschi, S., Biber, K., Yona, S.,
Edinger, A.L., Jung, S., Rossner, M.J., and Simons, M. (2016). Age-related
myelin degradation burdens the clearance function of microglia during aging.
Nat. Neurosci. 19, 995–998.
Sala Frigerio, C., Wolfs, L., Fattorelli, N., Thrupp, N., Voytyuk, I., Schmidt, I.,
Mancuso, R., Chen, W.T., Woodbury, M.E., Srivastava, G., et al. (2019). The
major risk factors for Alzheimer’s disease: age, sex, and genes modulate the
microglia response to Abeta plaques. Cell Rep. 27, 1293–1306.e6.
Sayed, F.A., Telpoukhovskaia, M., Kodama, L., Li, Y., Zhou, Y., Le, D., Hauduc,
A., Ludwig, C., Gao, F., Clelland, C., et al. (2018). Differential effects of partial
Neuron 105, 837–854, March 4, 2020 853
and complete loss of TREM2 on microglial injury response and tauopathy.
PNAS 115, 10172–10177.
Shi, Y., and Holtzman, D.M. (2018). Interplay between innate immunity and
Alzheimer disease: APOE and TREM2 in the spotlight. Nat. Rev. Immunol.
18, 759–772.
Shibuya, Y., Chang, C.C., Huang, L.H., Bryleva, E.Y., and Chang, T.Y. (2014).
Inhibiting ACAT1/SOAT1 in microglia stimulates autophagy-mediated lyso-
somal proteolysis and increases Ab1-42 clearance. J. Neurosci. 34,
14484–14501.
Shibuya, Y., Chang, C.C., and Chang, T.Y. (2015). ACAT1/SOAT1 as a thera-
peutic target for Alzheimer’s disease. Future Med. Chem. 7, 2451–2467.
Simes, R.J. (1986). An improved Bonferroni procedure for multiple tests of sig-
nificance. Biometrika 73, 751–754.
Sudom, A., Talreja, S., Danao, J., Bragg, E., Kegel, R., Min, X., Richardson, J.,
Zhang, Z., Sharkov, N., Marcora, E., et al. (2018). Molecular basis for the loss-
of-function effects of the Alzheimer’s disease-associated R47H variant of the
immune receptor TREM2. J. Biol. Chem. 293, 12634–12646.
Tall, A.R. (2008). Cholesterol efflux pathways and other potential mechanisms
involved in the athero-protective effect of high density lipoproteins. J. Intern.
Med. 263, 256–273.
Ulland, T.K., and Colonna, M. (2018). TREM2—a key player in microglial
biology and Alzheimer disease. Nat. Rev. Neurol. 14, 667–675.
Ulland, T.K., Song, W.M., Huang, S.C., Ulrich, J.D., Sergushichev, A., Beatty,
W.L., Loboda, A.A., Zhou, Y., Cairns, N.J., Kambal, A., et al. (2017). TREM2
854 Neuron 105, 837–854, March 4, 2020
maintains microglial metabolic fitness in Alzheimer’s disease. Cell 170, 649–
663.e13.
Wang, Y., Cella, M., Mallinson, K., Ulrich, J.D., Young, K.L., Robinette, M.L.,
Gilfillan, S., Krishnan, G.M., Sudhakar, S., Zinselmeyer, B.H., et al. (2015).
TREM2 lipid sensing sustains the microglial response in an Alzheimer’s dis-
ease model. Cell 160, 1061–1071.
Wang, Y., Ulland, T.K., Ulrich, J.D., Song, W., Tzaferis, J.A., Hole, J.T., Yuan,
P., Mahan, T.E., Shi, Y., Gilfillan, S., et al. (2016). TREM2-mediated early micro-
glial response limits diffusion and toxicity of amyloid plaques. J. Exp. Med.
213, 667–675.
Wickham, H. (2016). ggplot2: elegant graphics for data analysis (New York:
Springer).
Wu, D., and Smyth, G.K. (2012). Camera: a competitive gene set test account-
ing for inter-gene correlation. Nucleic Acids Res. 40, e133.
Xu, C., and Su, Z. (2015). Identification of cell types from single-cell transcrip-
tomes using a novel clustering method. Bioinformatics 31, 1974–1980.
Yeh, F.L., Wang, Y., Tom, I., Gonzalez, L.C., and Sheng, M. (2016). TREM2
binds to apolipoproteins, including APOE and CLU/APOJ, and thereby facili-
tates uptake of amyloid-beta by microglia. Neuron 91, 328–340.
Yeh, F.L., Hansen, D.V., and Sheng, M. (2017). TREM2, microglia, and neuro-
degenerative diseases. Trends Mol. Med. 23, 512–533.
Yuan, P., Condello, C., Keene, C.D., Wang, Y., Bird, T.D., Paul, S.M., Luo, W.,
Colonna, M., Baddeley, D., and Grutzendler, J. (2016). TREM2 haplodeficiency
in mice and humans impairs the microglia barrier function leading to decreased
amyloid compaction and severe axonal dystrophy. Neuron 90, 724–739.
KEEP
YOUR
FINGER
ON
THE
PULSE

With Cell Press Reviews

Looking for authoritative reviews on the
forefront of science? Turn to Cell Press.
Our editors stay abreast of the latest
developments so they can commission
expert, cutting-edge reviews.
To propel your research forward, faster, turn to
Cell Press Reviews. Our insightful, authoritative
reviews—published across the life sciences in
our primary research and Trends journals—go
beyond synthesis and offer a point of view.

Find your way

with Cell Press Reviews
www.cell.com/reviews
 ll

Article
Oxidative Metabolism Drives Immortalization
of Neural Stem Cells during Tumorigenesis
François Bonnay,1 Ana Veloso,2 Victoria Steinmann,1 Thomas Köcher,3 Merve Deniz Abdusselamoglu,1,5 Sunanjay Bajaj,1
Elisa Rivelles,4,6 Lisa Landskron,1,7 Harald Esterbauer,4 Robert P. Zinzen,2 and Juergen A. Knoblich1,8,*
1Instituteof Molecular Biotechnology of the Austrian Academy of Sciences (IMBA), 1030 Vienna, Austria
2Systems Biology of Neurogenesis, Berlin Institute for Medical Systems Biology (BIMSB), Max Delbrück Center for Molecular Medicine in the
Helmholtz Association (MDC), 13125 Berlin, Germany
3Vienna Biocenter Core Facilities (VBCF), 1030 Vienna, Austria
4Department of Laboratory Medicine, Medical University of Vienna, 1090 Vienna, Austria
5Present address: Robin Chemers Neustein Laboratory of Mammalian Cell Biology and Development, Howard Hughes Medical Institute, The

Rockefeller University, New York, NY, USA

6Present address: IHR Labor, Medical Diagnostic Laboratory, 1220 Vienna, Austria
7Present address: Netherlands Cancer Institute (NKI), 1066 CX Amsterdam, the Netherlands
8Lead Contact

*Correspondence: juergen.knoblich@imba.oeaw.ac.at
https://doi.org/10.1016/j.cell.2020.07.039

SUMMARY

Metabolic reprogramming is a key feature of many cancers, but how and when it contributes to tumorigenesis
remains unclear. Here we demonstrate that metabolic reprogramming induced by mitochondrial fusion can
be rate-limiting for immortalization of tumor-initiating cells (TICs) and trigger their irreversible dedication to
tumorigenesis. Using single-cell transcriptomics, we find that Drosophila brain tumors contain a rapidly
dividing stem cell population defined by upregulation of oxidative phosphorylation (OxPhos). We combine
targeted metabolomics and in vivo genetic screening to demonstrate that OxPhos is required for tumor
cell immortalization but dispensable in neural stem cells (NSCs) giving rise to tumors. Employing an in vivo
NADH/NAD+ sensor, we show that NSCs precisely increase OxPhos during immortalization. Blocking Ox-
Phos or mitochondrial fusion stalls TICs in quiescence and prevents tumorigenesis through impaired
NAD+ regeneration. Our work establishes a unique connection between cellular metabolism and immortali-
zation of tumor-initiating cells.

INTRODUCTION demonstrate defects in the tricarboxyic acid (TCA) cycle and

Tumor cells are immortalized and display a number of character-
istic features that distinguish them from normal cells. Among
those, reprogramming of metabolic pathways is indisputable
but only partially understood. Since Otto Warburg demonstrated
that tumor cells preferentially ferment glucose into lactate in lieu
of oxygen-enabled mitochondrial oxidative phosphorylation (the
Warburg effect) (Warburg, 1956; Warburg et al., 1927), energy
metabolism in tumor cells has been under growing scrutiny.
Although widely accepted for multiple tumor types and animal
models (Flaveny et al., 2015; Patra et al., 2013), the Warburg ef-
fect is not as universal as assumed previously. In fact, carcinoma
subtypes derived from the brain (Janiszewska et al., 2012), leu-
kocytes (Kuntz et al., 2017), pancreas (Franco et al., 2016),
skin (Haq et al., 2013; Vazquez et al., 2013), lungs (Rao et al.,
2019; Ye et al., 2011), and ovaries (Pastò et al., 2014) rely on
oxidative phosphorylation (OxPhos) for retaining stemness of
their cancerous cells. The complexity of metabolic reprogram-
ming is illustrated by a recent report using isotope tracing to
increased glycolytic metabolism in clear-cell renal cell carci-
noma (ccRCC) (Courtney et al., 2018). However, the same study
also confirmed that, unlike ccRCC, brain- and lung-derived tu-
mors retain fully functional and efficient glucose oxidation
(Courtney et al., 2018), suggesting that mitochondrion-depen-
dent bioenergetics can also be a key contributor of tumorigen-
esis. It is now also well established that oxidative metabolism
can contribute to tumor cell proliferation by promoting biosyn-
thesis of aspartate and subsequent metabolism of nucleotides
(Birsoy et al., 2015; Sullivan et al., 2015). However, when these
metabolic changes are required during tumorigenesis and how
they can contribute to the immortalization process is unknown.
Because the complex microenvironment and heterogeneity of
mammalian tumors make it difficult to address the precise timing
and function of metabolic changes during the tumorigenesis pro-
cess, we used a genetically defined Drosophila neural stem cell
(NSC)-derived tumor model to overcome these caveats.
Drosophila NSCs and neuroblasts (NBs) have proven to be an
important model for understanding key developmental

1490 Cell 182, 1490–1507, September 17, 2020 ª 2020 Elsevier Inc.

Article
processes like asymmetric cell division and temporal patterning
(Homem and Knoblich, 2012). During the larval stages of devel-
opment, NBs form neurons and glia cells following highly repro-
ducible lineages. NBs and their progeny do not migrate and are
surrounded by a glial sheath, so they can be easily recognized
and tracked in clonal analysis. Multiple cellular markers allow un-
ambiguous tracking of maturation in progenitors and differentia-
tion in the daughter cells so NBs are particularly suitable for
studying aberrant cell behavior such as tumorigenesis (Jiang
and Reichert, 2014). Changes in metabolism profoundly influ-
ence NB behavior (Homem et al., 2014), suggesting that bioen-
ergetics and cell fate are tightly connected.
Based on their lineages, NBs are subdivided into type 0, type I,
and type II. Type 0 NBs divide asymmetrically into one NB and
one post-mitotic neuron (Karcavich and Doe, 2005; Ulvklo et al.,
2012). Type I NBs (NBIs) produce a ganglion mother cell (GMC)
that again divides into two post-mitotic neurons or glia cells
(Buescher et al., 1998). Type II NBs (NBIIs) generate transit-ampli-
fying intermediate neural progenitors (INPs) (Bello et al., 2008;
Boone and Doe, 2008; Bowman et al., 2008) that divide asymmet-
rically 5–6 times, each time generating a GMC that ultimately
gives rise to two neurons or glia cells. NBIIs can be uniquely iden-
tified based on their expression of the Ets transcription factor
Pointed (PntP1) (Zhu et al., 2011) and by the absence of Asense
(Ase), a neural precursor marker present in all other NB subtypes
(Brand et al., 1993; Jarman et al., 1993). Right after division, each
INP undergoes a stereotypical maturation process that can be
tracked by sequential expression of unique markers. In newly
born INPs, the asymmetrically segregating RNA binding protein
Brain tumor (Brat) post-transcriptionally inhibits the general NB
marker Deadpan (Dpn) (Reichardt et al., 2018). In a stereotypical
fashion, the INP then turns on Ase, followed by re-expressing
Dpn, before it re-enters the cell cycle. The presence of a transit-
amplifying population greatly increases the number of neurons
generated by each NBII, and this might explain why NBIIs are
particularly susceptible to tumorigenesis when the asymmetric
cell division machinery or differentiation programs are altered
(Kelsom and Lu, 2012; Knoblich, 2010).
NBs mutant for brat give rise to abnormal INPs that constitu-
tively express the Brat targets dpn and zelda (Reichardt et al.,
2018). Instead of initiating Ase expression and re-entering
mitosis, tumor-initiating cells enter a long transient cell cycle ar-
rest (Bowman et al., 2008). Only after 24–48 h do they undergo
mitosis and irrevocably convert into immortal tumor NBs (tNBs)
(Caussinus and Gonzalez, 2005) that will form huge lethal and
transplantable brain tumors. Although all brain NBs become
quiescent and differentiate or die during pupariation, tNBs can
sustain proliferation until adulthood (Betschinger et al., 2006;
Landskron et al., 2018), indicating that they lack major prolifera-
tion control mechanisms. How these tumor cells become
immortal and why they resist metamorphosis-induced differenti-
ation remains a mystery.
Because our previous results indicated a strong connection
between metabolism and growth control in the Drosophila brain
(Homem et al., 2014), we used the NBII lineage to investigate
metabolic reprogramming during tumorigenesis. Our results
reveal that brat NSC-derived tumors exhibit increased TCA inter-
mediates and higher oxygen consumption, whereas lactate pro-
ll
duction and acidity rates, hallmarks of the Warburg effect, are
only minimally changed. By single-cell RNA sequencing
(scRNA-seq) profiling of brat tumors, we identified a distinct sub-
population of tumor cells with higher expression of respiratory
enzymes. Our genetic data suggest that the oxidative population
drives tumor formation, whereas the glycolytic population is sec-
ondary. We generated an in vivo metabolic nicotinamide adenine
dinucleotide (NAD+) sensor to demonstrate that uncoupling of
transcriptional and metabolic changes in the brat mutant type
II NB population is rate limiting for their irreversible commitment
to tumorigenesis. Finally, we show that uncontrolled mitochon-
drial fusion causes the rate-limiting metabolic changes observed
in the tumor population. Our data reveal an unexpected connec-
tion between mitochondrial metabolism and cellular immortali-
zation and show that precise coordination between cell fate
and metabolic changes is essential to avoid tumorigenesis.
RESULTS
brat tNBs Undergo Oxidative Metabolism
Metabolic regulation is critical for the onset and termination of
proliferation in Drosophila NSCs (Homem et al., 2014; Otsuki
and Brand, 2018). To determine whether immortalization of
Drosophila NSCs is associated with alterations of metabolic
pathways, we re-analyzed the transcriptome of NBIIs and their
tumorigenic bratRNAi counterparts (tNBs) (Landskron et al.,
2018). We determined mRNA expression changes between
tNBs and NBIIs for a wide range of enzymes catalyzing the reac-
tions of gluconeogenesis, glycolysis, fermentation, the pentose
phosphate pathway (PPP), the TCA cycle, OxPhos, and nucleo-
tide and amino acid biosynthesis (Figure S1A). Interestingly, four
glycolytic enzymes (Hex-A, Pfk, Ald, and Pgm) and the two
fermentation-promoting enzymes lactate dehydrogenase (LDH)
and pyruvate dehydrogenase kinase (PDK) were increased
more than 4-fold in tNBs. In contrast, most OxPhos-related en-
zymes were unchanged or downregulated. These gene expres-
sion data could be interpreted to mean that bratRNAi brain tumors
show enhanced glycolytic activity at the expense of OxPhos, as
already suggested for epithelial tumors in imaginal discs (Eichen-
laub et al., 2018; de la Cova et al., 2014; Wang et al., 2016).
We challenged our observations by directly measuring meta-
bolic changes in brain tumors (Figure 1A). We measured the ox-
ygen consumption rate (OCR) and extracellular acidification rate
(ECAR) of bratRNAi and control brain cells (Figures 1B–1D; Fig-
ures S1C and S1D). Although the OCR and ECAR were increased
in brain tumor cells, surprisingly, the OCR/ECAR ratio was also
strongly increased, which is indicative of increased oxidative
metabolism (Figure 1D). To confirm these initial results, we per-
formed a targeted metabolomics analysis by mass spectrom-
etry. Among 22 measured intermediates of glycolysis, the TCA
cycle, and PPP or amino acids (Figure 1E), we found six signifi-
cantly increased metabolites in brat tumors. Three were interme-
diates of the TCA cycle (citrate, alpha-ketoglutarate, and ma-
late), whereas none of the eight glycolytic intermediates were
increased significantly. Lactate was increased slightly, but pyru-
vate, the pivotal final glycolytic product, was the most upregu-
lated metabolite we observed. Although bratRNAi tumors mostly
consist of tNBs, whereas control brains contain a majority of
Cell 182, 1490–1507, September 17, 2020 1491
 ll
Article

A B C D E

F
H

G
I

K L
J

M N O

(legend on next page)

1492 Cell 182, 1490–1507, September 17, 2020

 ll
Article

neurons, the observed metabolic changes are most likely due to
alterations in the NB population. In mammals and Drosophila
(Hall et al., 2012; Volkenhoff et al., 2015), neurons predominantly
rely on OxPhos; therefore, observing higher TCA cycle interme-
diates and oxygen consumption in stem-cell-enriched tumor
brains is particularly intriguing.
We next investigated the origin of TCA cycle intermediates in
brat tumors. Normally, glucose is metabolized into TCA cycle in-
termediates via glycolysis and subsequent import of pyruvate
into mitochondria and its transformation into acetyl-coenzyme
A (CoA). Tumor cells, however, can use other carbon sources,
such as glutamine (DeBerardinis et al., 2007) or acetate (Ma-
shimo et al., 2014), via generation of alpha-ketoglutarate and
acetyl-CoA, respectively.
We independently performed [U-13C]glucose, [U-13C]L-Gln,
and [U-13C]acetate labeling in control and tumor brains and
analyzed the subsequent integration of 13C into glycolysis,
fermentation, and TCA cycle metabolites (Figures 1F–1I; Figures
S1E and S1F). Interestingly, incorporation of glucose into the
TCA cycle (citrate [M2]) was lower in tumor brains than in control
brains, whereas incorporation of glucose into lactate was higher
(Figures 1F and 1G). This is consistent with the higher LDH and
PDK levels found in tNBs in bulk RNA sequencing (RNA-seq)
data (Figure S1A). Acetate was incorporated into the TCA cycle
only to a minor extent in control and brat tumor brains (citrate
[M2]) and therefore does not seem to be a main contributor to
OxPhos in Drosophila brains (Figures S1E and S1F). Strikingly
however, Gln was predominantly incorporated into the TCA cy-
cle in control and tumor brains (malate [M4]) (Figures 1H and
1I). In control brains, 13C from Gln was rapidly diluted during a
second TCA cycle (citrate [M4] = 25%, citrate [M2] = 4%), indi-
cating that these TCA intermediates were diverted toward other
metabolic paths. In brat tumors, however, labeled TCA interme-
diates originating from 13C-Gln were recycled to a higher degree
(citrate [M4] = 46%, citrate [M2] = 10%), suggesting that, in
brains tumors, Gln-derived TCA cycle intermediates play a
more active role in OxPhos.
Altogether, these results indicate that brat tumors rely on Gln
rather than glucose to fuel OxPhos and that Gln-generated
TCA cycle intermediates are better recycled in the TCA cycle
in brat tumors than in control brains, further suggesting their pref-
erential use of OxPhos.
To determine bioenergetics in brat tNBs at cellular resolution,
we generated transgenic flies expressing the sensor for the co-
enzymes NAD+ and NADH (reduced form of NAD+), sensor of
NAD(H) redox (SoNar) (Zhao et al., 2016). SoNar can bind to
either NAD+ or NADH with specific conformations and excitatory
wavelengths, allowing us to follow the NAD+/NADH ratio by live
imaging. Because NADH consumption and NAD+ production are
not specific for fermentation or OxPhos, we first determined their
relative contribution to the SoNar readout. We used oxamate to
inhibit LDH, koningic acid to inhibit glyceraldehyde 3-phosphate
dehydrogenase (GAPDH), and rotenone and antimycin A to
inhibit ETC complex I and III on dissociated tNBs from bratRNAi
SoNar-expressing tNBs (Videos S1, S2, and S3; Figures 1J–
1O). Although OxPhos inhibitors induced a clear drop in NAD+/
NADH ratio, oxamate and koningic acid had no visible effect (Fig-
ures 1J–1M, 1O). Interestingly, however, not all tNBs decreased
their NAD+/NADH ratio upon rotenone or antimycin A treatment
(Figure 1J, white arrow; Videos S2 and S3), indicating metabolic
heterogeneity among tumor cells. Our measurements were
confirmed by fluorescence-activated cell sorting (FACS) analysis
of SoNar activity after treating cells with LDH and OxPhos inhib-
itors (Figure S1B; Figures 1K, 1L, and 1O). Finally, we used oxa-
mate, koningic acid, and rotenone on SoNar-expressing NBIs to
measure the tumor specificity of these responses (Figures 1N
and 1O); as expected, normal NBs decreased their NAD+/
NADH ratio upon oxamate but not rotenone treatment. Alto-
gether, these results indicate that most brat tNBs shift to OxPhos
and, unlike their non-tumorigenic counterparts, do not rely on
fermentation for their bioenergetic needs.
Dividing tNBs Display an Enhanced Oxidative
Metabolism Transcriptomic Signature
Mammalian tumors are known to be dynamic and increase their
heterogeneity with growth (Lawson et al., 2018). In particular,
bioenergetic heterogeneity is considered to be essential for tu-
mor progression and adaptability (Nakajima and Van Houten,
2013). To test the heterogeneity of brat tumors, we performed
single-cell Switch Mechanism At the 5’ end of RNA Templates
2 (Smart-Seq2) RNA-seq on 157 bratRNAi tumor and 93 control
central brain NBs sorted by FACS (STAR Methods; Figure S2A).
Although the sorting gate of control NBs was stringent enough to
only collect NBIs and NBIIs (the largest and brightest cells), the
sorting gate for tNBs was enlarged to account for the range of
tNB cell sizes within the tumor (Figure S2B). Subsequently, 29
non-tumorigenic cells originating from NBI lineages were
sequenced but excluded from tumor samples based on NBI/
GMC markers (asehigh) or absence of the proliferative marker
(dpnlow) (Figure S2A).

Figure 1. brat tNBs Exhibit an Oxidative Metabolism

(A) Processing of control and tumor third-instar larva whole brains (WB) for metabolic profiling.
(B–D) Oxygen consumption rate (OCR; B), extracellular acidification rate (ECAR; C), and OCR/ECAR ratio of control and tumor brains (D).
(E) Fold changes of metabolites from glycolysis, the TCA cycle, and PPP or amino acids (AA) metabolic pathways between tumor and control brains, normalized
by all analyzed metabolites as measured by liquid chromatography (LC)/mass spectrometry (MS). Metabolites highlighted in boldface are significantly higher in
tumor versus control samples (unpaired [two-tailed] t test, p < 0.05).
(F and H) Schematic view of [U-13C]glucose-originated (F) and [U-13C]glutamine-originated (H) 13C incorporation into the TCA cycle.
(G and I) 13C incorporation of [U-13C]glucose (G) and [U-13C]glutamine (I) into glycolytic (G) and TCA cycle intermediates (G and I) in control (gray bars) and tumor
(blue bars) third-instar larva WBs.
(J–N) Live-imaging of type II (J and M) or type I (N) NB-expressed SoNar and RFPnuclear and (K and L) FACS analysis of dissociated tNBs from third-instar larva
tumor (J–M) or control (N) brains before and after treatment with oxamate, rotenone, or koningic acid.
(O) Quantification of the NAD+/NADH ratio from individual control and tNBs 10 min after rotenone or oxamate treatment.
Scale bars, 20 mm. See also Figure S1, Table S2, and Videos S1, S2, and S3.

Cell 182, 1490–1507, September 17, 2020 1493

 ll
Article

A B

(legend on next page)

1494 Cell 182, 1490–1507, September 17, 2020

 ll
Article

Initially, tNBs were analyzed separately using the Pseudotime
analytical tool (Monocle R package) and separated by unsuper-
vised clustering into three clusters of 14 (cluster 1 [c1]), 49 (c2),
and 94 (c3) cells (Figure 2A). Interestingly, c1 was distinguished
from all other tNBs by its proliferative signature: the G2 > M cell
cycle inducer string (stg) was among the top 10 clustering genes.
Furthermore, c1 tNBs also expressed higher levels of several key
OxPhos components (CG5214 [a-ketoglutarate dehydrogenase
E2], Mdh2, and the mitochondrial phosphate carrier [Mpcp])
and two enzymes catalyzing formation of a-ketoglutarate (Fig-
ure 2B; Figure S2C) from lysine (Sccpdh1) and glutamate
(Gdh). This analysis suggests that c1 tNBs are the most prolifer-
ative and associated with higher levels of enzymes involved in
OxPhos (Figures 2D, 2E, and 2G). Additionally, the inducers of
G1 > S transition cyclin-E and E2f1 were less differentially ex-
pressed between the three clusters (Figures 2G–2I), suggesting
that increased OxPhos is specifically correlated with the G2 >
M phase transition in tNBs.
c3, the cluster most transcriptionally distinct from c1, was
separated from other tNBs by its higher expression of gluta-
thione-producing enzymes (GstD9 and GstE1) and unfolded pro-
tein response chaperones (Hsp26 and Hsp22) (Figure 2B), sug-
gesting higher stress levels in these cells. Furthermore, LDH
was higher in c3 compared with c1 and c2, probably indicating
a distinct bioenergetic profile (Figure 2F). Finally, c2 expressed
intermediate levels of stg and a-KGDH-E2, indicating an inter-
mediate or transient state between c1 and c3 (Figures 2D, 2E,
and 2G). Overall, these results reveal a striking heterogeneity
among bratRNAi tumor cells and identify a minor population of
highly proliferative cells with higher OxPhos among a majority
of cells that appear to be less proliferative and high in reactive
oxygen species (ROS) detoxification and endoplasmic reticulum
(ER)/mitochondrial stress.
To compare the three tNB clusters with untransformed control
NBs, we analyzed tNBs together with normal control NBs using
Monocle (Figure 2C). Intriguingly, cells in c1, which express high
levels of cell cycle regulators and OxPhos enzymes, were tran-
scriptionally closest to the majority of the control NBs, whereas
c3 was furthest away. A small minority of control NBs did not
follow this general trend and clustered in close proximity to c3
tNBs (Figure 2C). This subpopulation is not identified when
NBs are clustered without tNBs and shows significant expres-
sion differences in only three genes (bru3, troll, and Hr4), sug-
gesting that this is a clustering artifact generated by the pres-
ence of residual non-tNBs in the bratRNAi population.
Taken together, our data indicate that proliferative and oxida-
tive tNBs in c1 are generated first during tumorigenesis, whereas
the glycolytic cells in c2 and c3 represent later cells that have
already undergone changes to adapt to the hypoxic tumor
environment.
To understand how OxPhos was maintained in c1 cells, we
analyzed the expression pattern of all genes annotated as ‘‘tran-
scription factors’’ (TFs) or ‘‘DNA_binding’’ in pseudotime across
all tNBs and displayed candidates with cluster-specific expres-
sion changes (Figure S2D). We tested the role of 26 candidate
TFs with a clear association with c1 in bratRNAi tumorigenesis
(Figure S2E) and found that RNAi of the TFs E2F transcription
factor 1 (E2f1) and PAR-domain protein 1 (pdp1) led to partial
rescue of the tumor, as monitored by the tumor-specific RFP
signal in adult flies. Although E2f1 is a known cell cycle inducer
(Duronio et al., 1995), pdp1 is mostly known to regulate tran-
scriptional activity in muscle and circadian clock neurons (Cyran
et al., 2003), and its RNAi did not affect normal NBII proliferation
(Figure S2F). These data suggest that pdp1 acts as a tumor-spe-
cific transcriptional regulator of the most oxidative and prolifera-
tive subpopulation of brat tumors.
brat Tumorigenesis Requires Aerobic Metabolism
To test whether the observed metabolic changes are biologically
relevant for tumor growth, we designed a genetic screen for
genes that are essential for tumor-induced adult lethality (Fig-
ure 3A). Compared with control flies, bratRNAi tumor-bearing flies
have a dramatically shorter lifespan (15 days after hatching).
Depleting genes essential for NSC proliferation, such as dpn,
within the bratRNAi tumor restores the lifespan almost completely
to that of control flies. Therefore, we used the time when 50% of
the flies had died (LT50) as a readout for the degree of tumor
rescue that can be achieved by depletion of individual metabolic
enzymes (Figure 3B).
Using this assay, we performed an F2 genetic screen (Fig-
ure S3A) to test RNAi constructs targeting nearly all known en-
zymes involved in gluconeogenesis, glycolysis, fermentation,
the TCA cycle, ETC, and PPP (Figures 3C, 3D, and 3F; Fig-
ure S3B) for their ability to rescue brat tumor-induced lethality.
Strikingly, no RNAi construct targeting gluconeogenic, glyco-
lytic, or fermenting enzymes led to significant lifespan rescue.
In stark contrast, depletion of 10 TCA enzymes (Idh, Pdhb,
Acon, aKGDH-E1, E2, E3, Mdh2, Scs-a, Scs-b, and Fum1) and
four of the five electron transport chain (ETC) complexes (com-
plex I, III, IV, and V) (Figures S4C–S4H) each led to significant
rescue of the tumor-induced lethality.
To further show that a rescued lifespan was indeed correlated
with tumor prevention, we assessed the tumor burden by visual-
izing the tumor-specific RFP signal of adult flies. As expected,
tumors depleted of glycolytic and fermentation enzymes had
largely similar or even increased tumor burdens compared with

Figure 2. Dividing tNBs Display an Enhanced Oxidative Metabolism Transcriptomic Signature

(A) t-distributed Stochastic Neighbor Embedding (tSNE) plot of the 157 confirmed tNBs.
(A and B) Unsupervised clustering of tNBs using Monocle v2. Three clusters could be defined: c1 (14 cells, red), c2 (49 cells, blue), and c3 (94 cells, green). Shown
are the top 10 differentially expressed genes as determined by contrasting each cluster with all other cells in the sample.
(C) tSNE plot of the 157 confirmed tNBs (red, blue, and green dots) and the 93 confirmed control NBs (purple) together. The 4 NBIIs were determined by high
expression levels of pnt.
(D) Relative expression of lactate LDH, stg, and CG5214 among tumor and control NBs.
(E–I) Relative expression of CG5214, LDH, stg, cyclin-E, and E2f1 along pseudotime, as defined by the Monocle function plot_genes_in_pseudotime.
See also Figure S2.

Cell 182, 1490–1507, September 17, 2020 1495

 ll
Article

A B C

D E

G
F

H I

J K

(legend on next page)

1496 Cell 182, 1490–1507, September 17, 2020

 ll
Article

control tumors (with the notable exception of AldolaseRNAi
tumors, which showed partial rescue) (Figure 3E; Figures S4A
and S4B). In contrast, dpn- and OxPhos-deficient tumors
showed no or very low tumor-specific signal, similar to control
conditions, where no tumorigenesis was induced (Figures 3E
and 3G; Figures S4A and S4B).
These results could be explained if OxPhos enzymes were
generally required for all NBII fitness and/or proliferation and,
therefore, depleting them from the tumor would prevent its
growth. To rule out this possibility, we tested the same RNAi con-
structs for their effect on normal, non-transformed NBII lineages
(Figures S3C and S3D). As expected (Zacharioudaki et al., 2012),
dpn RNAi strongly reduced NBII progeny numbers. Interestingly,
AldolaseRNAi led to the same phenotype, indicating that glycol-
ysis is required for normal NBII lineage proliferation and explain-
ing its partial rescue of tumor formation. In striking contrast, RNAi
of the three TCA cycle enzymes aKGDH-E2, Scs-a, and Fum1
led to a non-significant slight increase in NBII progeny, and
RNAi of the ETC complex I, III, and IV top tumor-rescuing
enzymes led to stable NBII progeny numbers, indicating that
OxPhos is specifically required for tNB but not normal NBII pro-
liferation. These results are in line with our previous work
showing that OxPhos is not required for non-tumorigenic NB
proliferation (Homem et al., 2014). Altogether, these results indi-
cate that tNBs, unlike healthy NBs, rely on oxygen-dependent
bioenergetics to proliferate .
To further demonstrate that brat tumors rely on oxygen to pro-
liferate, we raised bratRNAi tumor-forming larvae under hypoxic
conditions (5% O2) for 24, 48, or 72 h after 3 days of develop-
ment. Strikingly, hypoxic incubation for 2 and 3 days led to
54% and 70% reduced tumor brain volume, respectively
(Figures 3H and 3I). This oxygen dependence was tumor specific
because control NBIIs grown under the same conditions did not
alter their progeny numbers. Thus, reduced oxygen levels do not
impair NBII proliferation but can significantly impair brain tumor
growth (Figures 3J and 3K).
OxPhos can contribute to tumor cell proliferation by promoting
biosynthesis of aspartate and subsequent metabolism of nucle-
otides (Birsoy et al., 2015; Sullivan et al., 2015) by the enzymes
glutamate oxaloacetate transaminase 1 and 2 (Got1 and Got2).
RNAi of both enzymes, however, did not rescue brat tumor for-
mation (Figure S3E), suggesting that brat tumor OxPhos is not
required for this purpose.
tNBs Undergo and Depend on Mitochondrial Fusion
Our data indicate a metabolic shift from glycolysis to OxPhos
when NBs become tumorigenic. Because the transcript levels
of key TCA cycle and ETC enzymes were unchanged or even
downregulated in tNBs (Figure S1A), we again analyzed our tran-
scriptome data for other causes of metabolic reprogramming
(Landskron et al., 2018) and focused on markers of key mito-
chondrial functions (Figure 4A). Strikingly, the only mitochondrial
markers upregulated in tNBs were the Dynamin-like GTPases
marf and opa1. Marf and Opa1 are required for mitochondrial
fusion and regulate fusion of the inner and outer mitochondrial
membranes, respectively (Westermann, 2010). Because mito-
chondrial fusion is well known to increase OxPhos (Mishra and
Chan, 2016) and causes cells to rely more heavily on OxPhos
for energy production (Rossignol et al., 2004), we wanted to
find out whether deregulation of these enzymes could explain
the observed metabolic transformations.
We investigated mitochondrial morphology in control NBIIs
and tNBs at the L3 stage using a mitochondrial GFP reporter
(GFPCOX8A) (Figure 4B). Mitochondria appeared to be numerous,
small, and round in control NBIIs and their immediate progeny,
indicative of fragmented mitochondria. However, in most tNBs,
mitochondria were thicker and more elongated, indicating sus-
tained mitochondrial fusion events. A few tNBs, however, also
showed fragmented mitochondria, even within the tumor tissue.
It is well known that mitochondria are fragmented during mitosis
(Yamano and Youle, 2011), and because those cells often occur
in pairs, they seem to be early post-mitotic cells. To confirm the
presence of fused mitochondria in tNBs, we performed transmis-
sion electron microscopy (TEM) on dorsal sections of control and
bratRNAi central brains (Figure 4C). bratRNAi tNBs contained
longer mitochondria (2.3-fold in length and 2.8-fold in area)
compared with NBIIs (Figures 4D and 4E). Interestingly, fused
mitochondria were also observed in numbRNAi larval brain tu-
mors (Figure 4B, rightmost panels). Like bratRNAi, numbRNAi re-
sults in formation of large, lethal, transplantable brain tumors
(Bowman et al., 2008; Caussinus and Gonzalez, 2005). However,
unlike brat tumors, numb tumors arise because of unrestricted,
enhanced Notch pathway activation (Bowman et al., 2008).
Thus, formation of enlarged, elongated mitochondria is a general
feature of Drosophila NSC-derived tumors that is independent of
the tumor-inducing signaling pathway.
Next, to determine whether bratRNAi tNBs mitochondrial fusion
resulted in increased OxPhos, we labeled dissociated control
and tNBs with tetramethylrhodamine (TMRM), whose fluores-
cence intensity correlates with mitochondrial membrane poten-
tial (Figures S6D–S6F). Expectedly, TMRM staining showed that
tNBs have significantly increased mitochondrial potential
compared with control NBs, indicating that bratRNAi tNB-fused
mitochondria display increased ETC activity.

Figure 3. The TCA Cycle, OxPhos, and Normoxia Are Necessary for brat Tumorigenesis
(A and B) Lifespan and (B) 50% lethality time (LT50) of flies with no tumor (mCherryRNAi) and tumor-bearing (bratRNAi, mCherryRNAi) and rescued tumor-bearing
(bratRNAi, dpnRNAi) flies.
(C, D, and F) Lifespan and (D and F) LT50 of metabolic pathway-deficient tumor-bearing flies. Genes depletions highlighted in boldface significantly rescued
tumor-induced lethality versus control tumors (unpaired [two-tailed] t test, p < 0.05).
(E and G) Tumor-specific RFP signal of flies bearing glycolysis pathway-, PPP-, glutaminolysis-, TCA-, or ETC-deficient tumors.
(H and I) RFPnuclear, Miranda, and PH3 immunostaining and (I) tumor lobe volume estimation of third-instar larva tumor brains incubated for 24, 48, or 72 h in 5%
O2. Scale bars, 50 mm.
(J) RFPnuclear, Miranda, and PH3 immunostaining and (K) NBII progeny count (RFP+ Miranda+) of third-instar larva control brains incubated for 24, 48, or 72 h in 5%
O2. Scale bars, 10 mm.
Each data point corresponds to one independent brain lobe in (I) and (K). See also Figures S3 and S4 and Table S1.

Cell 182, 1490–1507, September 17, 2020 1497

 ll
Article

A B

C
D

F
G

H I

(legend on next page)

1498 Cell 182, 1490–1507, September 17, 2020

 ll
Article

To analyze the role of marf, opa1, and other mitochondrial reg-
ulators in tumorigenesis, we analyzed their effect on brat tumor
lethality (Figures 4F and 4G) and tumor burden (Figures 4H and
4I; Figures S5D and S5E). Strikingly, marfRNAi was sufficient to
fragment tNB mitochondria (Figure S5A), completely abrogate
the tumor, and rescue tumor-induced lethality, whereas overex-
pressing marf slightly increased the tumor burden. Depleting
opa1 abrogated the tumor but did not rescue tumor-induced
lethality. Because adult escapers did not harbor a detectable tu-
mor, this early lethality is likely due to effects of opa1 RNAi
outside of the NBII lineage. In contrast, depleting drp1, a regu-
lator of mitochondrial fission, did not affect tumor growth, indi-
cating that mitochondrial fragmentation is not essential for the
tumorigenesis process. Interestingly, we also found that ses-B,
an ATP:ADP antiporter, and tko, a component of the mitochon-
drial small ribosomal subunit are essential for tumor growth (Fig-
ures 4F and 4G; Figures S5D and S5E).
We also analyzed the role of marf, opa1, and ses-B in normal
NBIIs (Figures S5B and S5C). RNAi of any of these genes led
to a similar or higher number of INPs compared with control
NBII lineages. Altogether, these results indicate that mitochon-
drial fusion is an essential component of tumorigenesis in the
Drosophila brain and the key mechanism underlying metabolic
reprograming in tNBs.
Metabolic Reprogramming Coincides with
Immortalization in tNBs
To determine the precise role of metabolic reprogramming dur-
ing tumor cell immortalization, we used SoNar to determine the
metabolic state of each cell in the NBII lineage during tumorigen-
esis. For this, we first determined the point of no return for tumor
formation using temperature-sensitive brat RNAi. We used the
Temporal And Regional Gene Expression Targeting (TARGET)
system (McGuire et al., 2003), where expression of the upstream
activating sequence (UAS)/Gal4 system is prevented by a ther-
mosensitive Gal80 (Gal80TS) inhibitor. This allowed us to induce
bratRNAi by a temperature shift and then re-express the Brat pro-
tein by shifting back to the permissive temperature.
After 24 h of brat RNAi (induced by inactivating Gal80TS at
29 C), NBIIs had generated several abnormal immature INPs
(imINPs) that continued to express Dpn because dpn translation
is not repressed by Brat (Figure 5A). However, at this stage,
these dysfunctional imINPs are cell cycle arrested in G2
(Bowman et al., 2008) and will not form a tumor when brat
expression is restored (by restoring Gal80TS by shifting the larvae
back to 18 C) (Figures S5F and S5G). However, 48 h and 72 h
after brat depletion, imINPs already began to regrow and reiniti-
ate cell division (Bowman et al., 2008). At this point, restoring brat
expression can no longer prevent tumor formation, indicating
that cell cycle re-entry irreversibly commits the abnormal INPs
to tumorigenesis.
To correlate metabolic changes with tumor commitment, we
reversibly depleted Brat by RNAi for 24 h in flies also expressing
SoNar (Figure 5A). For this, we first used the detection wave-
length for SoNar-NAD+ because the SoNar-NADH signal was
very sensitive to photobleaching and harder to detect (Figures
5A, 5B, and 5D, left panels). However, SoNar-NAD+ levels
were robustly detected (second left panels) and allowed us to
distinguish two cell populations with a remarkably sharp bound-
ary in between. Although NBs, imINPs (white arrowheads), and
the first mature INPs (mINPs; yellow arrowheads) show relatively
low NAD+ levels, later mINPs and all more differentiated progeny
have higher NAD+ levels (Figure 5A, top panels). Interestingly,
24 h bratRNAi NBII lineages showed the exact same profile,
with a sharp boundary between young and older tumor-initiating
cells, despite the fact that all of them mis-expressed dpn (red
arrowheads). However, after 48 h and 72 h of bratRNAi, a strong
increase in NAD+ levels occurred in the oldest tumor-initiating
cells (TICs) furthest away from the NB (Figures 5B, bottom
panels, blue arrowheads, and 5D, 5E, and 5G, blue arrowheads).
Importantly, NAD+ and NADH levels increased in 48 h and
72 h bratRNAi TICs, as revealed by careful scanning of both
SoNar excitation lengths (Figure 5C). Furthermore, these bio-
energetically abnormal (NAD+very high) cells contained higher
levels of marf compared with the other TICs (Figure 5F; Fig-
ure S6A) and were the first TICs harboring fused mitochondria
(Figure 5G, blue arrowheads). Importantly, marfRNAi almost
completely abrogated formation of these bioenergetically
abnormal TICs (Figure S6B), suggesting that mitochondrial
fusion generated this metabolic state. Finally, pdp1 was also
significantly enriched in NAD+very high cells (Figure S6C), indi-
cating that this TF may already be required for increasing
OxPhos during tumor initiation. Altogether, these results suggest
that mitochondrial fusion and a bioenergetic switch precisely
coincide with tumor cell immortalization in brat-deficient TICs.
OxPhos and Mitochondrial Fusion Are Required for
Tumor Cell Immortalization
To test whether the shift to OxPhos and the underlying fusion of
mitochondria are indeed required for TIC immortalization, we
measured how a-KGDH-E2, UQCR-Q, and marf RNAi affected
the different stages of tumor initiation. After 24 h of bratRNAi,

Figure 4. tNBs Undergo and Depend on Mitochondrial Fusion

(A) Transcriptional changes between tumor and control NBIIs (extracted from Landskron et al., 2018). Gene transcripts highlighted in boldface and blue are
significantly higher in tumor versus control NBIIs (unpaired [two-tailed] t test, p < 0.05).
(B) Immunostaining of NBII-expressed mitochondrion-addressed GFP (GFPCOX8A), RFPnuclear, and PH3 in control NBIIs and brat and numb tNBs. Yellow and
white arrows label tNBs with fused and fragmented mitochondria, respectively. Scale bars, 10 mm.
(C) TEM of control NBIIs and tNBs. Mitochondria are highlighted in blue (NBs) or in green (INPs). Scale bars, 2 mm.
(D and E) Mitochondrial surface and length from one control and one tNBII.
(F) Lifespan and (G) LT50 of mitochondrial biology gene-deficient tumor-bearing flies, representative of three independent experiments. Gene depletions
highlighted in boldface significantly rescued tumor-induced lethality versus control tumors (unpaired [two-tailed] t test, p < 0.05).
(H and I) Tumor-specific RFP signal of flies bearing mitochondrial fusion-deficient or -enhanced tumors.
(I) Head tumor burden quantification. Each dot corresponds to the average of 5–10 heads. n = 3 except for opa1RNAi (n = 1) because of developmental lethality.
See also Figures S5A–S5E and Table S1.

Cell 182, 1490–1507, September 17, 2020 1499

 ll
Article

A C

E F

Figure 5. tNB Start Changing Their Bioenergetics When They Immortalize

(A–C and G) Confocal images of NBII-expressed SoNar showing Dpn and ATP5A immunostaining in the control and bratRNAi NBII lineage after (A) 24 h, (B) 48 h,
and (D) 72 h of RNAi. Yellow stars, NBIIs; white and yellow arrowheads, imINPs and mINPs; red and blue arrowheads, young and older tumor-initiating cells (TICs);
white arrows, NAD+- and NADH-high cells. Scale bars, 10 mm.

(legend continued on next page)

1500 Cell 182, 1490–1507, September 17, 2020


Article
a-KGDH-E2RNAi, UQCR-QRNAi, and marfRNAi, TICs behaved
similarly as control TICs and failed to repress dpn (Figure 6A). Af-
ter 48 h of bratRNAi, control TICs immortalized and started prolif-
erating (Figure 6B, second column, yellow arrowheads). In sharp
contrast, at the same time point, OxPhos- and marf-deficient
TICs stacked but failed to start proliferation as efficiently (Figures
6B–6D). After 72 h of bratRNAi, control tumor cells were largely
proliferative and had expanded beyond their original lineage in
the brain (Figure 6C). At the same time point, a-KGDH-E2-,
UQCR-Q-, and marf-deficient TICs kept stacking and mostly
failed to re-initiate cell division (Figures 6C and 6D).
Finally, unrestrained bratRNAi-mediated tumorigenesis led to
highly proliferative (PH3+) gigantic brain expansion (Figure 6E).
In contrast, a-KGDH-E2-, UQCR-Q-, marf-, and sesB-deficient
bratRNAi NBII lineages kept producing intermediate progenitors
that mostly failed to re-initiate cell division (PH3 ) and did not
affect overall brain size. Unlike healthy NBII lineages, however,
these NBII lineages failed to generate fully differentiated
(Miranda-negative) smaller-sized progeny. Instead, nonprolifer-
ative TICs kept stacking in these NBII lineages that were no
longer clearly separated.
Finally, to understand how increased OxPhos promotes tumor
cell immortalization, we tested the possibility that release of ROS
generated by OxPhos could be required to initiate tumorigen-
esis. Indeed, brat tumors exhibit significantly more ROS than
normal NBIIs (Figures S7A and S7B). Importantly however,
depleting ROS by treating larvae initiating brat tumors with the
ROS scavenger N-acetylcysteine for 72 h did not alter or pro-
mote tumor initiation (Figures S7C and S7D). Furthermore, over-
expressing the Superoxide dismutase 1 and 2 similarly did not
affect bratRNAi tumor formation (Figures S7E and S7F). To further
decipher OxPhos function during tumorigenesis, we investigated
whether OxPhos-dependent NAD+ metabolism was sufficient to
sustain tumorigenesis independent of ATP synthesis. Strikingly,
overexpressing the proton-pump-independent yeast mitochon-
drial NADH dehydrogenase Ndi1 (Sanz et al., 2010) in ATP syn-
thase-reduced brat tumors was sufficient to fully restore tumor-
igenesis (Figures 7A–7D). These results suggest that NAD+
metabolism, rather than ATP or ROS production from this
increased oxidative metabolism, is primarily required by bratRNAi
TICs to immortalize. Altogether, these results indicate that a
metabolic switch to OxPhos and higher NAD+ metabolism
induced by uncontrolled mitochondrial fusion is the rate-limiting
step for irreversible immortalization of NSCs during tumor forma-
tion (Figure 7E).
DISCUSSION
The potential to divide indefinitely and to ignore organismal pro-
liferation control signals is a key hallmark of tumor cells. Several
cellular processes have been shown to play a crucial role in the
transition: telomere stabilization by telomerase (Shay and
Wright, 2011), sustained and deregulated proliferation control
(D) SoNar-NAD+ intensity quantification of successive INPs or TICs relative to their originating NBIIs.
(E) Schematic representation of (B).
(F) qPCR analysis of dpn, marf, stg, alpha-KGDH-E2, and LDH in control INPs and in NAD+-low and NAD+-high TICs.
See also Figures S5F and S5G.
ll
signaling pathways (Feitelson et al., 2015), inactivated cell cycle
checkpoints (Bertoli et al., 2013; Burkhart and Sage, 2008; Ho
and Dowdy, 2002), and evasion of apoptosis by expressing
anti-apoptotic factors (Fernald and Kurokawa, 2013). Metabolic
reprogramming has been widely considered an adaptation to the
environment when the tumor has been formed or during metas-
tasis but has not been described as a primary feature of tumor
cell immortalization (DeBerardinis and Chandel, 2016). Our re-
sults indicate that a metabolic switch toward OxPhos and
increased NAD+ biogenesis driven by glutamine incorporation
is rate limiting for the immortalization of stem cells during tumor
initiation in Drosophila brain tumors. Moreover, extensive fusion
of mitochondria induces a change in OxPhos in TICs to trigger
transition into tumor cells. Our data support the idea that, like
other changes of fate (Tatapudy et al., 2017), immortalization is
a dramatic and irreversible event in a cell that requires abrupt
metabolic changes.
Unlike other described Drosophila tumor models, brat NSC-
derived tumors originate from a glycolytic lineage and adapt
into oxidative cells strictly relying on oxygen and mitochondrial
NAD+ production. There is strong evidence supporting a key
role of OxPhos in various mammalian tumor types (Franco
et al., 2016; Haq et al., 2013; Kuntz et al., 2017; Pastò et al.,
2014; Rao et al., 2019; Vazquez et al., 2013; Ye et al., 2011),
including brain-derived tumors (Janiszewska et al., 2012), indi-
cating that this tumorigenic feature is evolutionarily conserved
from arthropods to humans. The key role of OxPhos in tumor
cell proliferation in Drosophila NSC-derived tumors is also sup-
ported by a recent report (van den Ameele and Brand, 2019).
Although this work did not specifically address the role of meta-
bolic reprogramming during tumor cell immortalization, together
our conclusions indicate a fundamental role of OxPhos in dereg-
ulation of cell proliferation programs.
Intriguingly, a molecularly distinct switch to oxidative meta-
bolism has been described in Drosophila NSCs but in a very
different context: when NSCs exit proliferation (Homem et al.,
2014). It has been described previously that NSCs transcription-
ally upregulate TCA and ETC enzymes during pupal stages and
that this is required for timely exit from the cell cycle. Although
superficially similar, the metabolic switch during pupal stages
is not accompanied by altered mitochondrial morphology and
is instead triggered by increased levels of TCA or ETC enzymes.
Furthermore, it has requirements for metabolic pathways that
are different from the one we describe here; although complex
II and flavoproteins fueling the ETC with reduced forms of flavin
adenine dinucleotide (FADH) are key to prevent NB regrowth and
to end proliferation in normal NBs, they are fully dispensable for
tNB proliferation. Overall, these combined observations suggest
that two distinct pathways can increase OxPhos in NSCs and
that their relative role is context dependent. When NSCs exit
proliferation, they do so by increasing the levels of respiratory
enzymes and utilizing the complete ETC to increase ATP produc-
tion. When immature TICs are immortalized, however, they do so
Cell 182, 1490–1507, September 17, 2020 1501
 ll
Article

A NB24h > CD8::GFP,

mCherryRNAi bratRNAi, bratRNAi, bratRNAi, bratRNAi,
mCherryRNAi KGDH-E2RNAi marfRNAi UQCR-QRNAi
PH3 GFP Dpn PH3 GFP Dpn PH3 GFP Dpn PH3 GFP Dpn PH3 GFP Dpn

*
*

B NB48h > CD8::GFP,

mCherryRNAi bratRNAi, bratRNAi, bratRNAi, bratRNAi,
mCherryRNAi KGDH-E2RNAi marfRNAi UQCR-QRNAi
PH3 GFP Dpn PH3 GFP Dpn PH3 GFP Dpn PH3 GFP Dpn

C NB72h > CD8::GFP,

bratRNAi, bratRNAi, bratRNAi, bratRNAi,
mCherryRNAi KGDH-E2RNAi marfRNAi UQCR-QRNAi
PH3 GFP Dpn PH3 GFP Dpn PH3 GFP Dpn PH3 GFP Dpn D
15 p=0.0075
48h of RNAi
Dpn+ PH3+ per lineage

12 72h of RNAi

9
p=0.0029
6
p<0.0001 p=0.022
p=0.021
3 p=0.008 p<0.0001

0
mCherry KGDH-E2 marf UQCR-Q
NB II > RNAi-brat + RNAi-

E
NBII > RFPnuclear,
mCherryRNAi bratRNAi, bratRNAi, bratRNAi, bratRNAi, bratRNAi,
mCherryRNAi KGDH-E2RNAi marfRNAi UQCR-QRNAi sesBRNAi
PH3 RFP Mira

(legend on next page)

1502 Cell 182, 1490–1507, September 17, 2020


Article
by enhancing mitochondrial membrane fusion and shortcutting
ETC complex II to critically increase mitochondrial NAD+ synthe-
sis. Interestingly, complex II has already been described to be
non-essential for respirasome formation in other contexts
(Jang and Javadov, 2018).
Our discovery that mitochondrial fusion and increased mito-
chondrial NAD+ production are key for triggering tumor cell
immortalization and inducing mitosis is fundamentally unex-
pected and surprising. In fact, multiple previous reports have
suggested the opposite (Lee et al., 2013; Peiris-Pagès et al.,
2018; Senft and Ronai, 2016). However, mitochondrial dynamics
(fragmentation and fusion) have already been shown in a very
different system to be instructive and critical for cell fate decision
between effector (fragmented mitochondria) and memory (fused
mitochondria) T cells (Buck et al., 2016). In this case, however, in
memory T cells as well as in other biological contexts, such as in
neurons (Beckervordersandforth, 2017) or oxidative muscle fi-
bers (Mishra and Chan, 2016), the combination of mitochondrial
fusion and increased OxPhos has been associated with longer-
lived and non-dividing cells. Only a recent report in fibroblasts
(Yao et al., 2019) associated mitochondrial fusion and increased
OxPhos with actively dividing cells, which, in this case, provides
enhanced aspartate production. Therefore, our study provides
the first strong evidence showing that OxPhos and mitochondrial
fusion can be potent proliferative mechanisms in multiple cellular
contexts.
How could this increased oxidative and NAD+ metabolism
contribute to tumor cell immortalization? First, our results indi-
cate that Got1 and Got2 are not required for tumorigenesis.
This argues against a key role of TICs’ increased OxPhos and
NAD+ pool to promote biosynthesis of aspartate and subsequent
anabolism of nucleotides. More generally, it seems unlikely that
the main role of increased OxPhos in TICs is to promote anabo-
lism because brat tumor cells divide at a similar or even slower
pace than normal type II NBs (Homem et al., 2014).
Instead, it is tempting to speculate that the increased NAD+ pool
observed in TICs is required for signaling during the long quiescent
period enabling TIC immortalization. Indeed, NAD+ is a well-
described co-factor regulating poly(ADP-ribose) polymerase
(PARP)-mediated poly-ADP ribosylation and Sirtuin-mediated de-
acetylation, which are known to influence chromatin plasticity,
genome stability, and cell fate (Hottiger, 2015). It will therefore
be of utmost interest to investigate the potential role of NAD+ in
these pathways specifically during tumor cell immortalization.
Importantly, our data based on RNAi cannot completely
exclude a role of glycolysis and fermentation in the tumor pro-
cess. However, the higher tumor burden observed upon
GAPDH, PGK, or LDH depletion argues that they would be detri-
mental to tumor development despite LDH being one of the most
Figure 6. OxPhos and Mitochondrial Fusion Are Required For Tumor Cell Immortalization
(A–C) Immunostaining of GFP, Dpn, and PH3 in mCherryRNAi, bratRNAi and TCA cycle-deficient, marf-deficient and ETC-deficient bratRNAi NBII lineages after (A) 24
h, (B) 48 h, and (C) 72 h of RNAi. Red arrowheads, Dpn misrepression; yellow arrowheads, PH3+ TICs.
(D) Quantification of PH3+ cells per NBII lineage after 48 h (orange) and 72 h (red) of TCA cycle-deficient, marf-deficient, and ETC-deficient bratRNAi. Scale
bars, 10 mm.
(E) WB lobe imaging of RFPnuclear, Miranda, and PH3 in the no-tumor control, TCA cycle-deficient tumors, and marf-deficient, ETC-deficient, and sesB-deficient
tumors. Yellow dashed line, RFP+ tumor area.
Scale bars, 50 mm. See also Figure S6.
ll
transcriptionally upregulated genes in tNBs and their preferential
use of pyruvate for fermentation. Interestingly, the dispensability
of high LDH levels in tumors has also been observed recently in
squamous cell carcinoma (Flores et al., 2019). This raises the
question of why such high levels of tumor-expressed glycolytic
and fermenting enzymes (HexA and LDH) transcript levels do
not result in lactate accumulation or increased acidity. One pos-
sibility suggested by other tumor models (Brisson et al., 2016;
Faubert et al., 2017; Hui et al., 2017) is that a fraction of the
tNB LDH levels would catalyze the opposite reaction to create
pyruvate out of lactate and fuel the TCA cycle. Furthermore,
the elevated levels of the PPP intermediates D-erythrose-4-
phosphate and phosphoribosyl pyrophosphate observed in
brat tumors could reveal an enhanced nucleotide biosynthesis
strategy to which high levels of glycolytic enzymes and PDK
may also contribute.
Our results shed light on uncharacterized events during the
long transition of a TIC converting into an ectopic tNB. Other
NB factors have been implicated in the earliest stages that
initiate the tumorigenesis process (such as dpn and zld in a
brat context; Reichardt et al., 2018). Although these factors
can undoubtedly affect tumorigenesis, they also have important
functions in normal NSC proliferation and development and,
therefore, cannot fully and unambiguously explain the tumori-
genic process. In contrast, our results show a strictly tumor-spe-
cific process that is independent of the tumor-triggering muta-
tion and that regulates the later step of tumor initiation.
OxPhos is not required for dpn mis-expression in brat-deficient
TICs; therefore, inhibiting OxPhos did not rescue the accumula-
tion of TICs. Instead, TICs switched to OxPhos only later, when it
was required for their immortalization and conversion into tNBs.
Is OxPhos only required during tumor initiation or can it also
play a broader role during tumorigenesis? Our results indicate
that established tumors are oxidative even at later stages and
that reducing oxygen concentration can prevent the growth of
already established tumors. Furthermore, scRNA-seq clearly
associated the most proliferative cell population with higher tran-
scriptional levels of OxPhos-associated genes. This strongly ar-
gues that this bioenergetic switch has to be maintained long after
tumor initiation to sustain growth of the tumor.
Finally, what are the molecular mechanisms driving mitochon-
drial fusion and the oxidative metabolic switch of the immortal-
izing tumor cells? It seems unlikely that Brat or Notch pathway
direct transcriptional targets are responsible for these changes;
most of them are widely expressed and required in NBs, which
undertake very distinct metabolic routes. Instead, it seems
more likely that TICs develop their own transcriptional program
during their long quiescent phase of cellular transformation to
induce this metabolic change. Identifying these regulatory
Cell 182, 1490–1507, September 17, 2020 1503
 ll
Article

B C

(legend on next page)

1504 Cell 182, 1490–1507, September 17, 2020

 ll
Article

mechanisms and analyzing their significance for mammalian tu- in the J.A. Knoblich laboratory is supported by the Austrian Federal Ministry of
mor cell immortalization will be of utmost interest. Education, Science and Research; the Austrian Academy of Sciences, the
Austrian Science Fund (Z_153_B09), and the City of Vienna and The Austrian
Science Fund (grants SFB F778 and DOC 72-B27, respectively). This project
STAR+METHODS has received funding from the European Research Council under the European
Union Horizon 2020 Research and Innovation Program (grants 695642,
Detailed methods are provided in the online version of this paper 693184, and 874769) and under the FP7 Program (grant 250342). F.B. was
and include the following: supported by a European Molecular Biology Organization long-term fellowship
(LTF 1280-2014). The VBCF Metabolomics Facility is funded by the City of
d KEY RESOURCES TABLE Vienna through the Vienna Business Agency.
d RESOURCE AVAILABILITY
B Lead Contact AUTHOR CONTRIBUTIONS

B Materials Availability
B Data and Code availability
d EXPERIMENTAL MODEL AND SUBJECT DETAILS
B Fly lines
B Fly husbandry
d METHOD DETAILS
B Cloning
B Seahorse metabolic analysis
B Mass-spectrometry targeted metabolomics
B Ex vivo live imaging of SoNar
B Single-cell RNA-sequencing
B RNA extraction, reverse-transcription and qPCR
analysis
B Lifespan and in vivo tumor burden analysis
B Hypoxic incubations
B N-acetylcysteine treatments
B Immunofluorescence
B ROS detection (Cell-ROX)
B Mitochondrial membrane potential assay
B Transmitted electron microscopy
d QUANTIFICATIONS AND STATISTICAL ANALYSIS
SUPPLEMENTAL INFORMATION
Supplemental Information can be found online at https://doi.org/10.1016/j.
cell.2020.07.039.
ACKNOWLEDGMENTS
We would like to dedicate this paper to Suzanne Eaton, one of the greatest
Drosophila biologists, who inspired many of the described experiments in fruit-
ful discussions during the early stages of this work. We thank all Knoblich lab
members for support and discussions; Joshua A. Bagley, Christopher Esk,
and Chong Li for comments on the manuscript; Peter Duchek, Joseph Gokce-
zade, Elke Kleiner, the IMP/IMBA Biooptics Facility, the Next Generation
Sequencing Unit, and the Electron Microscopy Unit of the Vienna Biocenter
Core Facilities (VBCF) for assistance; and the Harvard TRiP collection, the
Bloomington Drosophila Stock Center, and the Vienna Drosophila Resource
Center (VDRC) from the Vienna Biocenter Core Facilities (VBCF) for reagents.
We thank the BIMSB Genomics Core Facility, especially Bora Uyar and Vedran
Franke, for supplying the PiGx pipeline. We also thank Dr. Yi Yang (East China
University of Science and Technology) for providing the SoNar construct. Work
Conceptualization, F.B., A.V., R.P.Z., and J.A.K.; Methodology, F.B., A.V.,
V.S., L.L., T.K., R.P.Z., and J.A.K.; Investigation, F.B., A.V., V.S., M.D.A.,
S.B., E.R., T.K., and R.P.Z.; Resources, T.K., R.P.Z., and H.E.; Writing – Orig-
inal Draft, F.B. and J.A.K.; Writing – Review & Editing, F.B., R.P.Z., and J.A.K.;
Supervision, R.P.Z. and J.A.K.
DECLARATION OF INTERESTS
The authors declare no competing interests.
Received: June 17, 2020
Revised: June 24, 2020
Accepted: July 28, 2020
Published: September 10, 2020
REFERENCES
Beckervordersandforth, R. (2017). Mitochondrial Metabolism-Mediated Regu-
lation of Adult Neurogenesis. Brain Plast. 3, 73–87.
Bello, B.C., Izergina, N., Caussinus, E., and Reichert, H. (2008). Amplification
of neural stem cell proliferation by intermediate progenitor cells in Drosophila
brain development. Neural Dev. 3, 5.
Bertoli, C., Skotheim, J.M., and de Bruin, R.A.M. (2013). Control of cell cycle
transcription during G1 and S phases. Nat. Rev. Mol. Cell Biol. 14, 518–528.
Betschinger, J., Mechtler, K., and Knoblich, J.A. (2006). Asymmetric segrega-
tion of the tumor suppressor brat regulates self-renewal in Drosophila neural
stem cells. Cell 124, 1241–1253.
Birsoy, K., Wang, T., Chen, W.W., Freinkman, E., Abu-Remaileh, M., and Sa-
batini, D.M. (2015). An Essential Role of the Mitochondrial Electron Transport
Chain in Cell Proliferation Is to Enable Aspartate Synthesis. Cell 162, 540–551.
Boone, J.Q., and Doe, C.Q. (2008). Identification of Drosophila type II neuro-
blast lineages containing transit amplifying ganglion mother cells. Dev. Neuro-
biol. 68, 1185–1195.
Bowman, S.K., Rolland, V., Betschinger, J., Kinsey, K.A., Emery, G., and Kno-
blich, J.A. (2008). The tumor suppressors Brat and Numb regulate transit-
amplifying neuroblast lineages in Drosophila. Dev. Cell 14, 535–546.
Brand, M., Jarman, A.P., Jan, L.Y., and Jan, Y.N. (1993). asense is a Drosophila
neural precursor gene and is capable of initiating sense organ formation.
Development 119, 1–17.
Brisson, L., Banski, P., Sboarina, M., Dethier, C., Danhier, P., Fontenille, M.-J.,
Van Hée, V.F., Vazeille, T., Tardy, M., Falces, J., et al. (2016). Lactate Dehydro-
genase B Controls Lysosome Activity and Autophagy in Cancer. Cancer Cell
30, 418–431.

Figure 7. NAD+ Regeneration Is the Key Output of OxPhos during brat Tumorigenesis
(A) RFP-specific tumor burden analysis of flies bearing control tumors, ATP synthase-deficient tumors, and ATP synthase-deficient tumors rescued by over-
expression of Ndi1 bratRNAi.
(B) Head tumor burden quantification from (A). Each dot corresponds to one head.
(C) WB lobe imaging of RFPnuclear, Miranda, and PH3 in control, ATPsynd-deficient, and Ndi1-rescued ATPsynd-deficient tumors.
(D) Schematic of the experiments in (A)–(C) and their outcomes.
(E) Proposed mechanism of tumor cell immortalization during bratRNAi tumorigenesis. Scale bars, 50 mm. See also Figure S7.

Cell 182, 1490–1507, September 17, 2020 1505

 ll
Buck, M.D., O’Sullivan, D., Klein Geltink, R.I., Curtis, J.D., Chang, C.-H., Sanin,
D.E., Qiu, J., Kretz, O., Braas, D., van der Windt, G.J.W., et al. (2016). Mito-
chondrial Dynamics Controls T Cell Fate through Metabolic Programming.
Cell 166, 63–76.
Buescher, M., Yeo, S.L., Udolph, G., Zavortink, M., Yang, X., Tear, G., and
Chia, W. (1998). Binary sibling neuronal cell fate decisions in the Drosophila
embryonic central nervous system are nonstochastic and require inscute-
able-mediated asymmetry of ganglion mother cells. Genes Dev. 12,
1858–1870.
Burkhart, D.L., and Sage, J. (2008). Cellular mechanisms of tumour suppres-
sion by the retinoblastoma gene. Nat. Rev. Cancer 8, 671–682.
Butler, A., Hoffman, P., Smibert, P., Papalexi, E., and Satija, R. (2018). Inte-
grating single-cell transcriptomic data across different conditions, technolo-
gies, and species. Nat. Biotechnol. 36, 411–420.
Caussinus, E., and Gonzalez, C. (2005). Induction of tumor growth by altered
stem-cell asymmetric division in Drosophila melanogaster. Nat. Genet. 37,
1125–1129.
Courtney, K.D., Bezwada, D., Mashimo, T., Pichumani, K., Vemireddy, V.,
Funk, A.M., Wimberly, J., McNeil, S.S., Kapur, P., Lotan, Y., et al. (2018).
Isotope Tracing of Human Clear Cell Renal Cell Carcinomas Demonstrates
Suppressed Glucose Oxidation In Vivo. Cell Metab. 28, 793–800.e2.
Cyran, S.A., Buchsbaum, A.M., Reddy, K.L., Lin, M.-C., Glossop, N.R.J., Har-
din, P.E., Young, M.W., Storti, R.V., and Blau, J. (2003). vrille, Pdp1, and
dClock form a second feedback loop in the Drosophila circadian clock. Cell
112, 329–341.
de la Cova, C., Senoo-Matsuda, N., Ziosi, M., Wu, D.C., Bellosta, P., Quinzii,
C.M., and Johnston, L.A. (2014). Supercompetitor status of Drosophila Myc
cells requires p53 as a fitness sensor to reprogram metabolism and promote
viability. Cell Metab. 19, 470–483.
DeBerardinis, R.J., and Chandel, N.S. (2016). Fundamentals of cancer meta-
bolism. Sci. Adv. 2, e1600200.
DeBerardinis, R.J., Mancuso, A., Daikhin, E., Nissim, I., Yudkoff, M., Wehrli, S.,
and Thompson, C.B. (2007). Beyond aerobic glycolysis: transformed cells can
engage in glutamine metabolism that exceeds the requirement for protein and
nucleotide synthesis. Proc. Natl. Acad. Sci. USA 104, 19345–19350.
Duronio, R.J., O’Farrell, P.H., Xie, J.E., Brook, A., and Dyson, N. (1995). The
transcription factor E2F is required for S phase during Drosophila embryogen-
esis. Genes Dev. 9, 1445–1455.
Eichenlaub, T., Villadsen, R., Freitas, F.C.P., Andrejeva, D., Aldana, B.I.,
Nguyen, H.T., Petersen, O.W., Gorodkin, J., Herranz, H., and Cohen, S.M.
(2018). Warburg Effect Metabolism Drives Neoplasia in a Drosophila Genetic
Model of Epithelial Cancer. Curr. Biol. 28, 3220–3228.e6.
Eroglu, E., Burkard, T.R., Jiang, Y., Saini, N., Homem, C.C.F., Reichert, H., and
Knoblich, J.A. (2014). SWI/SNF complex prevents lineage reversion and in-
duces temporal patterning in neural stem cells. Cell 156, 1259–1273.
Faubert, B., Li, K.Y., Cai, L., Hensley, C.T., Kim, J., Zacharias, L.G., Yang, C.,
Do, Q.N., Doucette, S., Burguete, D., et al. (2017). Lactate Metabolism in Hu-
man Lung Tumors. Cell 171, 358–371.e9.
Feitelson, M.A., Arzumanyan, A., Kulathinal, R.J., Blain, S.W., Holcombe, R.F.,
Mahajna, J., Marino, M., Martinez-Chantar, M.L., Nawroth, R., Sanchez-Gar-
cia, I., et al. (2015). Sustained proliferation in cancer: Mechanisms and novel
therapeutic targets. Semin. Cancer Biol. 35 (Suppl ), S25–S54.
Fernald, K., and Kurokawa, M. (2013). Evading apoptosis in cancer. Trends
Cell Biol. 23, 620–633.
Flaveny, C.A., Griffett, K., El-Gendy, Bel.-D., Kazantzis, M., Sengupta, M.,
Amelio, A.L., Chatterjee, A., Walker, J., Solt, L.A., Kamenecka, T.M., and Bur-
ris, T.P. (2015). Broad Anti-tumor Activity of a Small Molecule that Selectively
Targets the Warburg Effect and Lipogenesis. Cancer Cell 28, 42–56.
Flores, A., Sandoval-Gonzalez, S., Takahashi, R., Krall, A., Sathe, L., Wei, L.,
Radu, C., Joly, J.H., Graham, N.A., Christofk, H.R., and Lowry, W.E. (2019).
Increased lactate dehydrogenase activity is dispensable in squamous carci-
noma cells of origin. Nat. Commun. 10, 91.
1506 Cell 182, 1490–1507, September 17, 2020
Article
Franco, J., Balaji, U., Freinkman, E., Witkiewicz, A.K., and Knudsen, E.S.
(2016). Metabolic Reprogramming of Pancreatic Cancer Mediated by CDK4/
6 Inhibition Elicits Unique Vulnerabilities. Cell Rep. 14, 979–990.
Hall, C.N., Klein-Flügge, M.C., Howarth, C., and Attwell, D. (2012). Oxidative
phosphorylation, not glycolysis, powers presynaptic and postsynaptic mech-
anisms underlying brain information processing. J. Neurosci. 32, 8940–8951.
Haq, R., Shoag, J., Andreu-Perez, P., Yokoyama, S., Edelman, H., Rowe, G.C.,
Frederick, D.T., Hurley, A.D., Nellore, A., Kung, A.L., et al. (2013). Oncogenic
BRAF regulates oxidative metabolism via PGC1a and MITF. Cancer Cell 23,
302–315.
Ho, A., and Dowdy, S.F. (2002). Regulation of G(1) cell-cycle progression by
oncogenes and tumor suppressor genes. Curr. Opin. Genet. Dev. 12, 47–52.
Homem, C.C.F., and Knoblich, J.A. (2012). Drosophila neuroblasts: a model for
stem cell biology. Development 139, 4297–4310.
Homem, C.C.F., Steinmann, V., Burkard, T.R., Jais, A., Esterbauer, H., and
Knoblich, J.A. (2014). Ecdysone and mediator change energy metabolism to
terminate proliferation in Drosophila neural stem cells. Cell 158, 874–888.
Hottiger, M.O. (2015). Nuclear ADP-Ribosylation and Its Role in Chromatin
Plasticity, Cell Differentiation, and Epigenetics. Annu. Rev. Biochem. 84,
227–263.
Hui, S., Ghergurovich, J.M., Morscher, R.J., Jang, C., Teng, X., Lu, W., Es-
parza, L.A., Reya, T., Le Zhan, Yanxiang Guo, J., et al. (2017). Glucose feeds
the TCA cycle via circulating lactate. Nature 551, 115–118.
Jang, S., and Javadov, S. (2018). Elucidating the contribution of ETC com-
plexes I and II to the respirasome formation in cardiac mitochondria. Sci.
Rep. 8, 17732.
Janiszewska, M., Suvà, M.L., Riggi, N., Houtkooper, R.H., Auwerx, J.,
Clément-Schatlo, V., Radovanovic, I., Rheinbay, E., Provero, P., and Stamen-
kovic, I. (2012). Imp2 controls oxidative phosphorylation and is crucial for pre-
serving glioblastoma cancer stem cells. Genes Dev. 26, 1926–1944.
Jarman, A.P., Brand, M., Jan, L.Y., and Jan, Y.N. (1993). The regulation and
function of the helix-loop-helix gene, asense, in Drosophila neural precursors.
Development 119, 19–29.
Jiang, Y., and Reichert, H. (2014). Drosophila neural stem cells in brain devel-
opment and tumor formation. J. Neurogenet. 28, 181–189.
Karcavich, R., and Doe, C.Q. (2005). Drosophila neuroblast 7-3 cell lineage: a
model system for studying programmed cell death, Notch/Numb signaling,
and sequential specification of ganglion mother cell identity. J. Comp. Neurol.
481, 240–251.
Kelsom, C., and Lu, W. (2012). Uncovering the link between malfunctions in
Drosophila neuroblast asymmetric cell division and tumorigenesis. Cell Biosci.
2, 38.
Knoblich, J.A. (2010). Asymmetric cell division: recent developments and their
implications for tumour biology. Nat. Rev. Mol. Cell Biol. 11, 849–860.
Kuntz, E.M., Baquero, P., Michie, A.M., Dunn, K., Tardito, S., Holyoake, T.L.,
Helgason, G.V., and Gottlieb, E. (2017). Targeting mitochondrial oxidative
phosphorylation eradicates therapy-resistant chronic myeloid leukemia stem
cells. Nat. Med. 23, 1234–1240.
Landskron, L., Steinmann, V., Bonnay, F., Burkard, T.R., Steinmann, J., Reich-
ardt, I., Harzer, H., Laurenson, A.-S., Reichert, H., and Knoblich, J.A. (2018).
The asymmetrically segregating lncRNA cherub is required for transforming
stem cells into malignant cells. eLife 7, R106.
Lawson, D.A., Kessenbrock, K., Davis, R.T., Pervolarakis, N., and Werb, Z.
(2018). Tumour heterogeneity and metastasis at single-cell resolution. Nat.
Cell Biol. 20, 1349–1360.
Lee, K.S., Wu, Z., Song, Y., Mitra, S.S., Feroze, A.H., Cheshier, S.H., and Lu, B.
(2013). Roles of PINK1, mTORC2, and mitochondria in preserving brain tumor-
forming stem cells in a noncanonical Notch signaling pathway. Genes Dev. 27,
2642–2647.
Mashimo, T., Pichumani, K., Vemireddy, V., Hatanpaa, K.J., Singh, D.K., Sira-
sanagandla, S., Nannepaga, S., Piccirillo, S.G., Kovacs, Z., Foong, C., et al.
(2014). Acetate is a bioenergetic substrate for human glioblastoma and brain
metastases. Cell 159, 1603–1614.

Article
McGuire, S.E., Le, P.T., Osborn, A.J., Matsumoto, K., and Davis, R.L. (2003).
Spatiotemporal rescue of memory dysfunction in Drosophila. Science 302,
1765–1768.
Mishra, P., and Chan, D.C. (2016). Metabolic regulation of mitochondrial dy-
namics. J. Cell Biol. 212, 379–387.
Nakajima, E.C., and Van Houten, B. (2013). Metabolic symbiosis in cancer: re-
focusing the Warburg lens. Mol. Carcinog. 52, 329–337.
Neumüller, R.A., Richter, C., Fischer, A., Novatchkova, M., Neumüller, K.G.,
and Knoblich, J.A. (2011). Genome-wide analysis of self-renewal in Drosophila
neural stem cells by transgenic RNAi. Cell Stem Cell 8, 580–593.
Otsuki, L., and Brand, A.H. (2018). Cell cycle heterogeneity directs the timing of
neural stem cell activation from quiescence. Science 360, 99–102.
Pastò, A., Bellio, C., Pilotto, G., Ciminale, V., Silic-Benussi, M., Guzzo, G., Ra-
sola, A., Frasson, C., Nardo, G., Zulato, E., et al. (2014). Cancer stem cells from
epithelial ovarian cancer patients privilege oxidative phosphorylation, and
resist glucose deprivation. Oncotarget 5, 4305–4319.
Patra, K.C., Wang, Q., Bhaskar, P.T., Miller, L., Wang, Z., Wheaton, W., Chan-
del, N., Laakso, M., Muller, W.J., Allen, E.L., et al. (2013). Hexokinase 2 is
required for tumor initiation and maintenance and its systemic deletion is ther-
apeutic in mouse models of cancer. Cancer Cell 24, 213–228.
Peiris-Pagès, M., Bonuccelli, G., Sotgia, F., and Lisanti, M.P. (2018). Mito-
chondrial fission as a driver of stemness in tumor cells: mDIVI1 inhibits mito-
chondrial function, cell migration and cancer stem cell (CSC) signalling. Onco-
target 9, 13254–13275.
Piyankarage, S.C., Augustin, H., Grosjean, Y., Featherstone, D.E., and Shippy,
S.A. (2008). Hemolymph amino acid analysis of individual Drosophila larvae.
Anal. Chem. 80, 1201–1207.
Rao, S., Mondragón, L., Pranjic, B., Hanada, T., Stoll, G., Köcher, T., Zhang, P.,
Jais, A., Lercher, A., Bergthaler, A., et al. (2019). AIF-regulated oxidative phos-
phorylation supports lung cancer development. Cell Res. 29, 579–591.
Reichardt, I., Bonnay, F., Steinmann, V., Loedige, I., Burkard, T.R., Meister, G.,
and Knoblich, J.A. (2018). The tumor suppressor Brat controls neuronal stem
cell lineages by inhibiting Deadpan and Zelda. EMBO Rep. 19, 102–117.
Rossignol, R., Gilkerson, R., Aggeler, R., Yamagata, K., Remington, S.J., and
Capaldi, R.A. (2004). Energy substrate modulates mitochondrial structure and
oxidative capacity in cancer cells. Cancer Res. 64, 985–993.
Sanz, A., Soikkeli, M., Portero-Otı́n, M., Wilson, A., Kemppainen, E., McIlroy,
G., Ellilä, S., Kemppainen, K.K., Tuomela, T., Lakanmaa, M., et al. (2010).
Expression of the yeast NADH dehydrogenase Ndi1 in Drosophila confers
increased lifespan independently of dietary restriction. Proc. Natl. Acad. Sci.
USA 107, 9105–9110.
Senft, D., and Ronai, Z.A. (2016). Regulators of mitochondrial dynamics in can-
cer. Curr. Opin. Cell Biol. 39, 43–52.
Shay, J.W., and Wright, W.E. (2011). Role of telomeres and telomerase in can-
cer. Semin. Cancer Biol. 21, 349–353.
Sullivan, L.B., Gui, D.Y., Hosios, A.M., Bush, L.N., Freinkman, E., and Vander
Heiden, M.G. (2015). Supporting Aspartate Biosynthesis Is an Essential Func-
tion of Respiration in Proliferating Cells. Cell 162, 552–563.
Tatapudy, S., Aloisio, F., Barber, D., and Nystul, T. (2017). Cell fate decisions:
emerging roles for metabolic signals and cell morphology. EMBO Rep. 18,
2105–2118.
Trapnell, C., Cacchiarelli, D., Grimsby, J., Pokharel, P., Li, S., Morse, M., Len-
non, N.J., Livak, K.J., Mikkelsen, T.S., and Rinn, J.L. (2014). The dynamics and
regulators of cell fate decisions are revealed by pseudotemporal ordering of
single cells. Nat. Biotechnol. 32, 381–386.
ll
Tumanov, S., Bulusu, V., Gottlieb, E., and Kamphorst, J.J. (2016). A rapid
method for quantifying free and bound acetate based on alkylation and GC-
MS analysis. Cancer Metab. 4, 17.
Ugrankar, R., Theodoropoulos, P., Akdemir, F., Henne, W.M., and Graff, J.M.
(2018). Circulating glucose levels inversely correlate with Drosophila larval
feeding through insulin signaling and SLC5A11. Commun. Biol. 1, 110–115.
Ulvklo, C., MacDonald, R., Bivik, C., Baumgardt, M., Karlsson, D., and Thor, S.
(2012). Control of neuronal cell fate and number by integration of distinct
daughter cell proliferation modes with temporal progression. Development
139, 678–689.
van den Ameele, J., and Brand, A.H. (2019). Neural stem cell temporal
patterning and brain tumour growth rely on oxidative phosphorylation. eLife
8, 859.
Vazquez, F., Lim, J.-H., Chim, H., Bhalla, K., Girnun, G., Pierce, K., Clish, C.B.,
Granter, S.R., Widlund, H.R., Spiegelman, B.M., and Puigserver, P. (2013).
PGC1a expression defines a subset of human melanoma tumors with
increased mitochondrial capacity and resistance to oxidative stress. Cancer
Cell 23, 287–301.
Volkenhoff, A., Weiler, A., Letzel, M., Stehling, M., Klämbt, C., and Schirmeier,
S. (2015). Glial Glycolysis Is Essential for Neuronal Survival in Drosophila. Cell
Metab. 22, 437–447.
Wang, C.-W., Purkayastha, A., Jones, K.T., Thaker, S.K., and Banerjee, U.
(2016). In vivo genetic dissection of tumor growth and the Warburg effect. eLife
5, 11.
Warburg, O. (1956). On respiratory impairment in cancer cells. Science 124,
269–270.
Warburg, O., Wind, F., and Negelein, E. (1927). THE METABOLISM OF TU-
MORS IN THE BODY. J. Gen. Physiol. 8, 519–530.
Westermann, B. (2010). Mitochondrial fusion and fission in cell life and death.
Nat. Rev. Mol. Cell Biol. 11, 872–884.
Wurmus, R., Uyar, B., Osberg, B., Franke, V., Gosdschan, A., Wreczycka, K.,
Ronen, J., and Akalin, A. (2018). PiGx: reproducible genomics analysis pipe-
lines with GNU Guix. Gigascience 7, 1226.
Yamano, K., and Youle, R.J. (2011). Coupling mitochondrial and cell division.
Nat. Cell Biol. 13, 1026–1027.
Yao, C.-H., Wang, R., Wang, Y., Kung, C.-P., Weber, J.D., and Patti, G.J.
(2019). Mitochondrial fusion supports increased oxidative phosphorylation
during cell proliferation. eLife 8, 279.
Ye, X.-Q., Li, Q., Wang, G.-H., Sun, F.-F., Huang, G.-J., Bian, X.-W., Yu, S.-C.,
and Qian, G.-S. (2011). Mitochondrial and energy metabolism-related proper-
ties as novel indicators of lung cancer stem cells. Int. J. Cancer 129, 820–831.
Zacharioudaki, E., Magadi, S.S., and Delidakis, C. (2012). bHLH-O proteins are
crucial for Drosophila neuroblast self-renewal and mediate Notch-induced
overproliferation. Development 139, 1258–1269.
Zhao, Y., Wang, A., Zou, Y., Su, N., Loscalzo, J., and Yang, Y. (2016). In vivo
monitoring of cellular energy metabolism using SoNar, a highly responsive
sensor for NAD(+)/NADH redox state. Nat. Protoc. 11, 1345–1359.
Zhu, S., Lin, S., Kao, C.-F., Awasaki, T., Chiang, A.-S., and Lee, T. (2006). Gra-
dients of the Drosophila Chinmo BTB-zinc finger protein govern neuronal tem-
poral identity. Cell 127, 409–422.
Zhu, S., Barshow, S., Wildonger, J., Jan, L.Y., and Jan, Y.-N. (2011). Ets tran-
scription factor Pointed promotes the generation of intermediate neural pro-
genitors in Drosophila larval brains. Proc. Natl. Acad. Sci. USA 108,
20615–20620.
Cell 182, 1490–1507, September 17, 2020 1507
 Cell Metabolism

Resource

Single-Cell RNA Sequencing Maps Endothelial

Metabolic Plasticity in Pathological Angiogenesis
Katerina Rohlenova,1,19 Jermaine Goveia,1,19 Melissa Garcı́a-Caballero,1,19 Abhishek Subramanian,1,19
Joanna Kalucka,1,16 Lucas Treps,1 Kim D. Falkenberg,1 Laura P.M.H. de Rooij,1 Yingfeng Zheng,2 Lin Lin,3,4 Liliana Sokol,1
Laure-Anne Teuwen,1,5,6 Vincent Geldhof,1 Federico Taverna,1 Andreas Pircher,1,14 Lena-Christin Conradi,1,15
Shawez Khan,1 Steve Stegen,7 Dena Panovska,8 Frederik De Smet,8 Frank J.T. Staal,9 Rene J. Mclaughlin,9

(Author list continued on next page)

1Laboratory of Angiogenesis and Vascular Metabolism, Center for Cancer Biology, VIB, Department of Oncology, Leuven Cancer Institute, KU

Leuven, Leuven 3000, Belgium

2State Key Laboratory of Ophthalmology, Zhongshan Ophthalmic Center, Sun Yat-Sen University, Guangzhou 510060, Guangdong, China
3Department of Biomedicine, Aarhus University, Aarhus 8000, Denmark
4Lars Bolund Institute of Regenerative Medicine, BGI-Qingdao, Qingdao 266555, China
5Translational Cancer Research Unit, GZA Hospitals Sint-Augustinus, Antwerp 2610, Belgium
6Center for Oncological Research, University of Antwerp, Antwerp 2000, Belgium
7Laboratory of Clinical and Experimental Endocrinology, Department of Chronic Diseases, Metabolism and Aging, KU Leuven, Leuven 3000,

Belgium
8Laboratory for Precision Cancer Medicine, Translational Cell & Tissue Research, Department of Imaging & Pathology, KU Leuven, Leuven

3000, Belgium
9Department of Immunology and Blood Transfusion, Leiden University Medical Center, Leiden 2300 RC, the Netherlands
10OXURION NV, Leuven 3001, Belgium
11BGI-Shenzhen, Shenzhen 518083, China

(Affiliations continued on next page)

SUMMARY cycle progression and in quiescence. Hypothesizing

Endothelial cell (EC) metabolism is an emerging
target for anti-angiogenic therapy in tumor angiogen-
esis and choroidal neovascularization (CNV), but little
is known about individual EC metabolic transcrip-
tomes. By single-cell RNA sequencing 28,337 murine
choroidal ECs (CECs) and sprouting CNV-ECs, we
constructed a taxonomy to characterize their hetero-
geneity. Comparison with murine lung tumor ECs
(TECs) revealed congruent marker gene expression
by distinct EC phenotypes across tissues and dis-
eases, suggesting similar angiogenic mechanisms.
Trajectory inference predicted that differentiation of
venous to angiogenic ECs was accompanied by
metabolic transcriptome plasticity. ECs displayed
metabolic transcriptome heterogeneity during cell-
that conserved genes are important, we used an inte-
grated analysis, based on congruent transcriptome
analysis, CEC-tailored genome-scale metabolic
modeling, and gene expression meta-analysis in
cross-species datasets, followed by in vitro and
in vivo validation, to identify SQLE and ALDH18A1
as previously unknown metabolic angiogenic targets.
INTRODUCTION
Endothelial cell (EC) metabolism regulates angiogenesis, and is
an emerging target for anti-angiogenic therapy (AAT) in cancer
and wet age-related macular degeneration (AMD) (Eelen et al.,
2018). The design of new AATs by targeting EC metabolism
would benefit from a better understanding of individual EC

Context and Significance

Targeting the metabolism of endothelial cells (ECs) is a promising strategy to block pathological blood vessel growth, or
angiogenesis, for the treatment of diseases like cancer. Understanding the landscape of metabolic gene expression at
the single-cell level will aid in identifying novel angiogenic targets. Here, researchers in Belgium and their colleagues sur-
veyed thousands of ECs in pre-clinical models of age-related macular degeneration and lung cancer. Their comprehensive
investigation identified genes and metabolic pathways that are congruently upregulated across diseases and tissues during
angiogenesis. Using an integrated analysis, the researchers generated a list of prioritized metabolic candidates and vali-
dated the importance of two candidates, SQLE and ALDH18A1, in pathological angiogenesis, supporting their potential
as therapeutic targets.

862 Cell Metabolism 31, 862–877, April 7, 2020 ª 2020 Elsevier Inc.
Stefan Vinckier,1 Tine Van Bergen,10 Nadine Ectors,8 Patrik De Haes,10 Jian Wang,11,12 Lars Bolund,3,4 Luc Schoonjans,1,2
Tobias K. Karakach,1,17,18 Huanming Yang,11,12 Geert Carmeliet,7 Yizhi Liu,2 Bernard Thienpont,13 Mieke Dewerchin,1
Guy Eelen,1 Xuri Li,2,20,* Yonglun Luo,3,4,11,12,20,* and Peter Carmeliet1,2,20,21,*
12China National GeneBank, BGI-Shenzhen, Shenzhen 518120, China
13Laboratory for Functional Epigenetics, Department of Human Genetics, KU Leuven, Leuven 3000, Belgium
14Present address: Department of Hematology and Oncology, Internal Medicine V, Medical University Innsbruck, Innsbruck, Austria
15Present address: Clinic of General, Visceral and Pediatric Surgery, University Medical Center Göttingen, Göttingen, Germany
16Present address: Aarhus Institute of Advanced Studies (AIAS), Department of Biomedicine, Aarhus University, Aarhus DK-8000, Denmark
17Present address: Bioinformatics Core Laboratory, Children’s Hospital Research Institute of Manitoba, Winnipeg, MB, Canada
18Present address: Rady Faculty of Health Sciences, Department of Pediatrics and Child Health, University of Manitoba, Winnipeg, MB,

Canada
19These authors contributed equally
20Senior author
21Lead Contact

*Correspondence: lixr6@mail.sysu.edu.cn (X.L.), alun@biomed.au.dk (Y.L.), peter.carmeliet@kuleuven.vib.be (P.C.)

https://doi.org/10.1016/j.cmet.2020.03.009

metabolism, but it remains unknown if ECs express a heteroge-
neous metabolic gene signature and how single ECs reprogram
their metabolic transcriptome signature when forming new ves-
sels in disease. However, metabolomics (measuring metabolite
levels or metabolic fluxes) is insufficiently sensitive to determine
single EC metabolism. Since we documented that changes in
metabolic gene expression signatures at the bulk population
level can be predictive of changes in metabolism in ECs (Bruning
et al., 2018; Cantelmo et al., 2016; Kalucka et al., 2018; Vande-
keere et al., 2018), we analyzed the metabolic transcriptome of
ECs at the single-cell level.
During vessel sprouting, a navigating tip EC leads the way,
while proliferating stalk cells elongate the vessel sprout (Po-
tente et al., 2011); once newly formed vessels become
perfused, ECs adopt a quiescent phalanx phenotype (Welti
et al., 2013). ECs rely on metabolic reprogramming when
switching from quiescence to vessel sprouting (Eelen et al.,
2018; Li et al., 2019; Sawada and Arany, 2017; Yu et al.,
2018). In tumors, bulk metabolic gene expression profiling
identified metabolic targets in tumor ECs (Cantelmo et al.,
2016). AMD is a common blinding disease of elderly people,
characterized by ocular neovascularization. Laser-induced
choroid neovascularization (CNV) is a preclinical model of
AMD (Ambati and Fowler, 2012). Since angiogenic ECs in
AMD/CNV have not been studied at the single-cell level, we
used single-cell RNA sequencing (scRNA-seq) to profile their
(metabolic) transcriptome heterogeneity.
Anti-VEGF drugs are used for the treatment of cancer and
AMD, but resistance limits their efficacy (Jain, 2014; Yang
et al., 2016). Hence, there is an unmet clinical need to identify
novel angiogenic targets. scRNA-seq is a powerful technology
to identify such candidates, but an outstanding challenge is to
prioritize targets for further clinical translation. Here, we present
a strategy, starting from scRNA-seq and complemented with
orthogonal techniques, to prioritize metabolic targets that con-
trol angiogenesis.
RESULTS
Identification and Characterization of CNV-ECs by
scRNA-Seq
To model CNV in mice, we laser-induced 10 lesions per eye and
microdissected choroids 7 days later. We pooled choroids from
6 mice and repeated this procedure 3 times, using choroids from
healthy mice as controls (6 mice per sample, in triplicate) (Figures
1A and 1B). For comparative analysis, we generated a pooled
sample of two choroids from one healthy human donor (see
below). Single-cell suspensions were MACS-enriched for
CD45 /CD31+ ECs (Cantelmo et al., 2016) and subjected to
scRNA-seq. After quality filtering (Table S1), batch correction,
and in silico EC selection, graph-based clustering was per-
formed to group a total of 28,337 ECs according to their gene
expression profile. Clusters were annotated based on marker
genes (Tables S2 and S3) and results were visualized using
t-distributed stochastic neighbor embedding (t-SNE) (Figures
1C, 1D, S1A, and S1B).
CECs from control mice were indistinguishable from healthy
peripheral CECs from lasered mice and clustered together (Fig-
ures 1C–1F). We detected a new separate population in lasered
mice, not present in healthy CECs, representing CNV-ECs (Fig-
ure 1D). Compared to healthy CECs, CNV-ECs expressed acti-
vation markers associated with response to injury such as Sparc
(Bradshaw and Sage, 2001) and Col18a1, a source of the angio-
static endostatin, previously used for CNV treatment (Marneros
et al., 2007), a finding confirmed at the protein level by quantita-
tive mass cytometry (CyTOF) (Figures 1G, S1C, and S1D;
Table S2).
Taxonomy of CECs and CNV-ECs
In CECs, we identified previously unknown sublineages of the
classical arterial, capillary, and venous EC phenotypes (Figures
1E–1G; for more complete description of marker genes and puta-
tive inferred biological activity, see Table S3). For instance, we
identified large supply arteries (P1 on Figure 1D), smaller ramifying
arterioles (P2) and arterial ECs expressing the shear stress marker
(Pi16) (tentatively coined shear-stress induced arterial ECs [P3]),
and a laser-induced arterial subpopulation (activated arterial
CEC) that upregulated activation markers and matricellular pro-
teins (P4) (Figure 1H). Activated arterial CECs clustered together
with other arterial phenotypes (Figure 1D), suggesting a relatively
normal transcriptome. Capillary CEC phenotypes expressed sig-
natures of the outer (P5) and inner choriocapillaries (P6), charac-
terized by the differential expression of genes involved in fenestra-
tion and VEGF signaling (Blaauwgeers et al., 1999; McLeod et al.,
1995) (Table S2). Venous subclusters included cells expressing
markers of large caliber vessels (P7), venules (P8), shear stress

Cell Metabolism 31, 862–877, April 7, 2020 863

 A C D

G H I J

Figure 1. Heterogeneity of Choroid NEC and CNV-EC Phenotypes

(A) Schematic overview of the choroidal vasculature.
(B) Schematic overview of the study design.
(C) t-SNE plot, color-coded for the sample type. ECs isolated from healthy choroid are depicted in gray (top); ECs isolated from laser-injured choroids in red
(bottom). Red arrowhead and dotted circles indicate an emerging population of laser-injured specific CNV-EC phenotypes.
(D) t-SNE visualization of EC subpopulations in healthy and laser-injured choroids. CNV-EC phenotypes (red arrowhead in C) are boxed in the global t-SNE plot
and shown in a separate t-SNE plot on the lower left. Subclusters of peripheral (p) and intralesional CNV-ECs (c) are numbered.
(legend continued on next page)

864 Cell Metabolism 31, 862–877, April 7, 2020

(P9), post-capillary venules (pcvs) (P10) that upregulated a previ-
ously identified CEC signature of resident endothelial stem cells
(ESCs) (Naito et al., 2012; Wakabayashi et al., 2013), and an acti-
vated pcv CEC phenotype (p11) (Figures 1I and S1E). We
observed two putative lymphatic EC phenotypes (LEC [P12] and
LEC-like [P13]) that differentially expressed Lyve-1 (Figure 1D; Ta-
ble S2). The existence of lymphatics in the choroid remains
debated (Heindl et al., 2015; Koina et al., 2015).
Angiogenic CNV-ECs were distinct from normal CECs and
included proliferating ECs (C1 in Figure 1D) and tip ECs (C2),
but also 3 previously unknown phenotypes that expressed sig-
natures associated with transitioning from pcv to angiogenic
EC phenotypes (transitioning CNV-ECs [C3]), and immature
(immature [C4]) and maturing (neophalanx [C5]) neovasculature
(Figure 1J). Tip cells upregulated transcripts of the disease-
restricted angiogenic factor Pgf (encoding placental growth fac-
tor, Plgf) (Figures 1J and S1F). Immature ECs were characterized
by the lack of specific marker gene expression, but expressed
activation markers and upregulated ribosomal gene expression
consistent with an activated intermediate phenotype. Neopha-
lanx ECs expressed markers of mature capillaries and arteries,
and were characterized by upregulation of a Notch signaling
gene signature (Figure S1G).
Interestingly, transcription factor activity analysis using single-
cell regulatory network inference and clustering (SCENIC) (Aibar
et al., 2017) indicated differential transcription factor activity in
EC subtypes (Figure 2A). Consistent with previous reports, Nr2f2
expression was induced in activated pcvs and transitioning ECs
(Jeong et al., 2017), while Sox17 expression was highest in arterial
ECs (Corada et al., 2013; You et al., 2005). SCENIC analysis of
CNV-ECs also identified transcription factors not previously impli-
cated in EC specification, such as in tip (Tgif1), immature (Smad1
and Sox4), and proliferating (Trp53) CNV-ECs (Figure 2A).
We validated the taxonomy using orthogonal in situ localization
techniques. Quantitative RNAscope to count transcript numbers,
combined with staining for the EC marker CD105, confirmed that
arterial (Gja4) and venous (Nr2f2) marker transcripts did not coloc-
alize in the same CNV-ECs (Figure S1H). We confirmed by immu-
nostaining of healthy choroids the expression of the following EC
markers: (1) artery ECs (ELN) and arteriole ECs (CXCL12) (Figures
S2A–S2C), (2) capillary ECs (VEGFR2) (Figure S2D), and (3)
venous ECs (VWF and SELP) (Figures S2E and S2F). Immuno-
staining of CNV lesions confirmed the expression of a marker of
immature ECs (APLNR), tip ECs (PlGF, LXN, and CXCR4), and
pcv ECs (SPARCL1) (Figures S2G–S2J).
Metabolic Transcriptome Reprogramming during
Pathological Vessel Sprouting
We explored if ECs underwent a differentiation trajectory during
vessel sprouting and if EC differentiation was associated with
metabolic transcriptome changes. Trajectory inference analysis
predicted that the hierarchy of angiogenic phenotypes resulted
(E and F) t-SNE plots, color-coded for the expression of the indicated marker genes (E) or gene sets (F). Red arrowheads indicate cells with high expression of the
indicated marker gene. Scale: white/gray is low expression; black (gene) or red (gene sets) is high gene expression.
(G–J) Heatmap of gene expression levels of the top 50 marker genes for broad categories of EC phenotypes (G), artery subclusters (H), vein subclusters (I), and
CNV-EC subclusters (J). In this and all further heatmaps depicting marker genes, colors represent row-wise scaled gene expression with a mean of 0 and SD of 1
(Z scores).
c.c., choriocapillaris; r.p.e., retinal pigment epithelium. See also Figures S1 and S2 and Tables S1, S2, and S3.
from differentiation of activated pcv CECs to transitioning CNV-
ECs, then to immature CNV-ECs, which thereafter differenti-
ated to tip cells and finally to more mature neophalanx CNV-
ECs (Figure 2B). This prediction extends previous morpholog-
ical evidence that neovessels may originate from pcvs (Folk-
man, 1982). Since pcv CECs expressed a previously validated
signature of resident ESCs, our analysis provides further sug-
gestion that ESCs might contribute to new vessel sprouting,
as previously established by lineage tracing (Corey et al.,
2016; Manavski et al., 2018; McDonald et al., 2018; Mondor
et al., 2016; Red-Horse et al., 2010; Wakabayashi et al.,
2013, 2018).
Interestingly, when focusing on metabolic genes and path-
ways, we noted that membrane transport, ATP synthase, and
glycolysis gene signatures were dynamically regulated during
differentiation from quiescent vein to angiogenic ECs (Figure 2B).
Maximal differences in metabolic gene expression of central car-
bon metabolism were observed in the most angiogenic EC phe-
notypes (immature and tip ECs), possibly suggesting that these
ECs had higher metabolic demands to execute their biological
functions (Figure 2B).
Congruency Analysis of Metabolic Transcriptome
Reprogramming
We explored whether metabolic transcriptome reprogramming
was specific to CNV-ECs or a more general hallmark of the
angiogenic switch in pathological angiogenesis (such as in tu-
mors), as this would address a fundamental question in vascular
biology of whether vessels in different tissues and diseases form
via similar or different mechanisms. We therefore explored to
which extent CNV and tumors contained similar EC phenotypes,
and whether they expressed congruent genes.
Similarity and Congruency Analysis
We analyzed a publicly available, previously in-house gener-
ated dataset of murine lung tumor ECs (TECs) (Goveia et al.,
2020), which comprised largely similar EC phenotypes as
CNV-ECs. However, in addition, murine lung TECs contained
breach and pre-breach ECs (that expressed both tip cell
and podosome rosette markers, presumably involved in
vessel sprouting initiation), and interferon ECs (displaying a
transcriptome response to interferon, possibly involved in im-
mune surveillance) (Goveia et al., 2020). Focusing on all de-
tected genes, we explored whether similar EC phenotypes
could be detected in these diseases, and whether they ex-
pressed congruent genes.
We performed differential gene expression and gene set
enrichment analysis to determine which processes were upregu-
lated in CNV-ECs and TECs versus CECs and NECs, respec-
tively (Figures 2C and 2D). Gene sets associated with prolifera-
tion, hypoxia signaling, and extracellular matrix formation were
commonly upregulated (Figure 2C). Consistently, many of the
175 commonly upregulated genes were involved in extracellular
Cell Metabolism 31, 862–877, April 7, 2020 865
 A B

E F

Figure 2. Gene Expression Signatures in CNV and TEC Subtypes

(A) Dot plot heatmap of the inferred activity of the indicated transcription factors. Scale: white/gray is low expression; black is high gene expression; dot size
corresponds to the fraction of cells in each cluster that have higher than average activity of the indicated transcription factor.
(B) Pseudotime trajectory of the indicated CNV-EC phenotypes (left) and Loess regression-smoothened gene expression of the indicated genes and metabolic
gene sets in pseudotime (right).
(C) Vascular gene sets upregulated in TECs (versus lung NECs) and in CNV-ECs (versus choroid CECs). Gene sets congruently upregulated in TECs and CNV-ECs
are summarized on the right.
(D) Genes upregulated in TECs versus NECs and choroid CECs versus CNV-ECs. A selection of genes upregulated in lung TECs versus NECs (left) and in CNV-
ECs versus choroid CECs (right) are listed. Genes congruently upregulated in TECs and CNV-ECs are listed on the right. Genes encoding ribosomal proteins, ATP
synthase subunits, and elongation factors are not listed individually; instead, their total number is displayed.

(legend continued on next page)

866 Cell Metabolism 31, 862–877, April 7, 2020

matrix remodeling, cytoskeleton, glycolysis, EC activation, and
others. Interestingly, Aplnr, an angiogenic and vasculoprotective
gene that regulates EC metabolism (Apostolidis et al., 2018;
Hwangbo et al., 2017), was identified as a congruent marker of
CNV-ECs and TECs (Figure 2D).
Since differential analysis of pooled populations may not
adequately discover genes with restricted expression in small EC
subpopulations such as tip and proliferating cells, we determined
whether the same EC subpopulations were present in CNV-ECs
and TECs. We used the Jaccard similarity index to score the simi-
larity of marker gene sets of all EC subpopulations, and observed
that marker gene sets across CNV-ECs and TECs were relatively
similar for several EC subpopulations (Figures 2E and S3A).
Further, TECs and CNV-ECs of the same phenotype expressed
congruent marker genes (Figure 2F). Similar to CNV-ECs, trajectory
inference analysis predicted that the hierarchy of TEC phenotypes
originated in veins that expressed resident ESC markers (Goveia
et al., 2020; Wakabayashi et al., 2018) differentiating to postcapil-
lary veins and further to an immature TEC phenotype, tip cells, neo-
phalanx TECs, and activated arteries (Figure S3B).
Metabolic Transcriptome Signatures
Focusing on metabolic genes, we observed that proliferating
ECs in both disease models upregulated the expression of meta-
bolic genes involved in one-carbon metabolism, nucleotide syn-
thesis, tricarboxylic acid (TCA) cycle, and oxidative phosphoryla-
tion (OXPHOS) (Figure 3A). In contrast, glycolytic gene
expression was upregulated in proliferating, tip, and immature
ECs in tumors, and was elevated in CNV in proliferating ECs,
but less in tip and immature ECs (Figure 3B). These observations
might suggest that the metabolic demands of proliferating ECs
are disease- or tissue-type independent, while metabolic adap-
tations of other subtypes may be more plastic.
The metabolic gene expression signatures between the
different TEC phenotypes were more outspoken, possibly re-
flecting the harsh nutrient-deprived micro-environment in tu-
mors and the fact that TECs grow in an uncontrolled, non-
resolving manner. Indeed, heatmap analysis revealed that
most TECs exhibited a different metabolic transcriptome signa-
ture (Figure S3C). Subsequent analysis at the gene level showed
that capillary TECs upregulated the expression of genes control-
ling lipid uptake (Figures 3C and S3C), raising the question of
whether they need lipids for internal use when switching to
quiescence (Kalucka et al., 2018) and/or for trans-EC transport
to cancer cells for energy production or lipogenesis (Santos
and Schulze, 2012). Venous TECs upregulated transcripts of
genes involved in prostaglandin metabolism (Figures 3C and
S3C), suggesting a role in vasoregulation, sprouting, or vascular
inflammation (Félétou et al., 2011; Iñiguez et al., 2003). Interferon
(IFN)-activated TECs upregulated genes involved in nucleotide
catabolism to salvage/lower nucleotide content (Figure 3C) (Bar-
ankiewicz et al., 1986). In turn, breach TECs upregulated genes
involved in extracellular matrix production the most, in line with
(E) Three-dimensional principal component analysis (PCA) on the pairwise Jaccard similarity coefficients of marker gene sets between subpopulations in TECs
and CNV-ECs. Squares denote CNV-EC phenotypes; circles denote TEC phenotypes. Note that equivalents of breach, pre-breach, and interferon TEC phe-
notypes were not present in CNV-ECs.
(F) Heatmap of expression levels of congruent genes in TEC and CNV-EC phenotypes (all genes analyzed). Note that the TEC and CNV-EC heatmaps show the
same set of congruent genes.
Comp, component; PC, principal component. See also Figure S3.
their presumed role in vessel sprouting initiation (Goveia et al.,
2020) (Figures 3C and S3C).
Correlation with Transcription Factor Expression
Consistent with literature reports (Kanda et al., 2009), correlating
the gene signatures with inferred transcription factor activity
scores revealed that Pparg correlated with the induction of a tri-
glyceride catabolism signature. Interestingly, hypoxia, as indicated
by the inferred activity of HIF-1a, correlated with glycolysis and OX-
PHOS gene expression in TECs (Figure S3D). While expected for
glycolysis (given that hypoxia upregulates glycolysis; Eales et al.,
2016), the upregulation of OXPHOS genes in hypoxic conditions
was surprising (given that hypoxia suppresses OXPHOS; Eales
et al., 2016; Semenza, 2011), but in line with findings that HIF-1a
activation can occur at oxygen levels that are not sufficiently low
to suppress mitochondrial respiration (Eales et al., 2016). Thus,
while proliferating CNV-ECs and TECs upregulate OXPHOS and
glycolysis gene signatures, the metabolic transcriptome profile of
other angiogenic ECs is rather diverse across diseases.
Analysis of Metabolic Transcriptome Reprogramming
during the Cell Cycle
Since our congruency analysis showed that proliferating CNV-ECs
and TECs have similar gene expression profiles, we analyzed their
(metabolic) signature and transcription factor regulation in more
detail. We therefore developed a prediction model to reconstruct
continuous cell-cycle pseudotime in proliferating murine CNV-
ECs and TECs based on periodically expressed genes identified
in cultured TECs (Figures S3E–S3I; STAR Methods). Comparative
analysis of the top 1% most periodic genes in proliferating TECs
and CNV-ECs revealed 97 genes that were periodically expressed
during the cell cycle in both models, 43 of which were not previ-
ously described in reference databases (Santos et al., 2015). Inter-
estingly, in both models, several metabolic genes ranked in the top
1% most periodic genes (Rrm2, Tk1, Tyms, Dut, Rrm1, and
Dctpp1) (Figure 3D; Table S4). Further, gene set variation analysis
revealed that various metabolic pathways involved in nucleotide
biosynthesis (e.g., one-carbon metabolism and pyrimidine nucle-
oside biosynthesis) ranked among the top 5% most periodically
expressed gene sets (Figure 3E; Table S4). Overall, our scRNA-
seq offers new possibilities to study metabolic gene expression ki-
netics during the cell cycle.
Metabolic Angiogenic Target Identification
Identification of Commonly Upregulated Metabolic Genes
Since OXPHOS and glycolysis are validated metabolic angiogenic
targets (Cantelmo et al., 2016; De Bock et al., 2013; Diebold et al.,
2019), we designed an integrated analysis to identify other previ-
ously unrecognized angiogenic candidates regulating EC meta-
bolism. We thus performed differential gene expression analysis
to determine which metabolic genes and gene sets were
commonly upregulated in TECs and CNV-ECs, versus normal
ECs. Pathway mapping of gene transcripts involved in central
Cell Metabolism 31, 862–877, April 7, 2020 867
A serine, one-carbon B glycolysis C lipid metabolism
& nucleotide synthesis

TEC CNV-EC TEC CNV-EC

serine Slc2a1 Lpl
1C
Hk1 Lipe
cycle biosynthesis
TCA nucleotide

Pfkl Mgll
Pfkm Fabp4
Pfkp Fabp5
2
Aldoa Pparg 1
Tpi1 Cd36 0
Gapdh Gpihbp1 -1

Pgk1
Pgam1 prostaglandin metabolism
OXPHOS

Eno1 Ptgs1
2 2
2 1
Eno1b 1 Ptgs2 1
Pkm Ptgis
0 0 0 0
Ldha 0
-1 Slco2a1
-2 Slc16a3 -1
-2
tip cell
proliferating

transitioning
tip cell
proliferating

breach cell
pre-breach cell

vein

interferon
large vein

immature
immature

neophalanx
postcapillary

activated artery

artery
neophalanx

TEC capillary

tip cell
proliferating

transitioning
tip cell

breach cell

interferon
proliferating

pre-breach cell

vein
large vein

immature
immature

neophalanx
postcapillary

activated artery

artery
neophalanx

TEC capillary
nucleotide breakdown
Cmpk2
2
Upp1
1
Nt5c3 0
Pnp -1

ECM biosynthesis
Chst7
D expression in cell cycle E expression in cell cycle Mgat5
Stt3a
metabolic genes metabolic gene sets Ganab
Colgalt1 2

100 0 100 0 Plod1

0
Plod3
M M P4ha1
88 88 -2
87 87

tip cell
proliferating

breach cell

interferon
pre-breach cell

vein
large vein
immature
postcapillary

activated artery

artery
neophalanx

TEC capillary
G2M

G2M
G1

G1
TECs

TECs

75 75
74 74
G2

G2

64 64
63 S 63 S
G1S 42 G1S 42
43 Tk1 43 G1/S specific transcription
53 52 53 52
Rrm2 kinesins
Tyms one-carbon pool by folate
100 0 100 0
Dut deoxyribonucleotide biosynthesis
88
M Dctpp1 88
M pyrimidine nucleoside biosynthesis
87 87
G2M

G2M
G1

G1
CNV-ECs

CNV-ECs

75 75
74 74
G2

G2

64 64
63 S 63 S
G1S 42 G1S 42
43 43
53 52 53 52

Figure 3. Metabolic Heterogeneity in TECs and CNV-ECs

(A) Heatmap of the gene expression levels of the indicated metabolic pathways in TEC and CNV-EC subpopulations. Genes were grouped according to metabolic
pathways and ordered so that the most discriminative genes for proliferating ECs are depicted first.
(B) Heatmap of gene expression levels of the indicated glycolytic genes in TEC and CNV-EC subpopulations.
(C) Heatmap of gene expression levels of the indicated genes in TECs.
(D and E) Circos plot representation of Loess regression-smoothened gene expression of the indicated genes (D) and gene sets (E) during the cell cycle in TECs
(top) and CNV-ECs (bottom). Cell-cycle phase assignment, indicated as color-coded sectors, was based on the known periodicity of cyclins. Cell-cycle
pseudotime is represented as a percentage and indicated outside of the circos plot, starting (t = 0%) from cytokinesis (start of G1). The thickness of the line
corresponds to the expression levels.
1C, one-carbon. See also Figure S3 and Table S4.

carbon metabolism confirmed that CNV-ECs upregulated tran- also involved in other metabolic/biological activities, such as redox
scripts of metabolic pathways supporting biomass synthesis, and energy homeostasis (Eelen et al., 2018). Compared to NECs,
including glycolysis, nucleotide synthesis, TCA cycle, OXPHOS, TECs showed a qualitatively largely similar upregulation of
and others (Figures 4A and 4B). Several of these pathways are anabolic pathways (Figures 4A and S4A).

868 Cell Metabolism 31, 862–877, April 7, 2020

A B
C D E analysis, angiogenic ECs have at least
The upregulation of OXPHOS and TCA cycle gene expression
signatures in proliferating ECs was noteworthy, when consid-
ering that ECs are glycolysis-addicted (Cantelmo et al., 2016;
De Bock et al., 2013), but is in line with reports showing the
importance of OXPHOS for EC proliferation (Diebold et al.,
2019; Ying et al., 2017). The upregulation of a glycolysis tran-
scriptome signature in TECs has been validated by functional ev-
idence that genetic deletion in ECs or pharmacological blockade
of the glycolytic activator PFKFB3 reduces TEC proliferation and
induces tumor vessel normalization (Cantelmo et al., 2016), and
may also explain why treatment with a PFKFB3 blocker inhibited
CNV in the mouse (Schoors et al., 2014).
An unbiased metabolic gene set enrichment analysis, combined
with a congruency analysis to identify commonly upregulated
metabolic pathways, revealed that genes involved in pathways in
central carbon metabolism were among the most upregulated in
TECs and CNV-ECs, including glycolysis, OXPHOS, nucleotide
Figure 4. Global Metabolic Reprogramming
in Pathological Angiogenesis
(A) Dot plot heatmap of the indicated glycolytic
genes. The dot size corresponds to the fraction of
cells that have higher than average activity of the
indicated genes. Scale: white/gray is low expres-
sion; black is high gene expression.
(B) Map of upregulated central carbon metabolic
pathways in CNV-ECs versus peripheral CECs.
Blue color indicates downregulated expression,
red upregulated expression, and gray unchanged
expression.
(C and D) Metabolic gene set enrichment analysis
in NEC versus TECs (C) and CEC versus CNV-ECs
(D) (q < 0.25 for all gene sets). Numbers between
parentheses indicate alternative gene sets per-
taining to the same biological function or signaling
pathway.
(E) Metabolic gene sets, upregulated in TECs
(versus lung NECs) and in CNV-ECs (versus
choroid CECs). Congruent upregulated metabolic
pathways are listed underneath.
See also Figure S4.
biosynthesis, and TCA cycle, compared
to NECs and peripheral ECs (Figures 4C–
4E). Notably, transcripts of genes involved
in collagen synthesis were also highly upre-
gulated in angiogenic ECs in both diseases
(Figure 4E). Thus, based on transcriptome
two prominent metabolic gene expression
signatures, i.e., that of biomass production
for proliferation and of collagen biosyn-
thesis for extracellular matrix remodeling.
Prediction of Candidates with
Functionally Relevant Role in
Metabolism by GEMs
Given that changes in transcript levels of
metabolic genes alone may not relate to
changes in metabolic fluxes, we used
genome-scale metabolic models (GEMs)
to in silico prioritize metabolic candidates.
GEMs are mathematical representations of a network of active
metabolic enzymes (pathways) (Kim and Lun, 2014; Ryu et al.,
2015) and represent computational tools to predict the impor-
tance of metabolic reactions for biological responses (Pagliarini
et al., 2016; Thiele and Palsson, 2010). We constructed a CEC-
tailored GEM based on the generic human metabolic recon-
struction RECON1 and scRNA-seq data from 1,670 freshly iso-
lated human CECs, which recapitulated murine CEC pheno-
types but with lower resolution (Figures 5A, S4B, and S4C),
and optimized it for biomass (as a proxy for proliferation) or
collagen production using two distinct EC-specific objective
functions.
Using the CEC-tailored GEM, we identified 288 essential
genes for biomass synthesis (Table S5), involved in glycolysis,
TCA cycle, pentose phosphate pathway, OXPHOS, fatty acid
oxidation, nucleotide synthesis and salvage, cholesterol biosyn-
thesis, sphingolipid metabolism, and amino acid metabolism
Cell Metabolism 31, 862–877, April 7, 2020 869
 A B

C D

Figure 5. Metabolic Target Prediction

(A) Schematic representation of GEM reconstruction.
(B) Venn diagrams indicating metabolic genes that encode rate-limiting enzymes and are predicted to be essential for biomass production (top) and collagen
biosynthesis (bottom) by four different methods (see STAR Methods for details).
(C) Upset plot visualization of the results of a differential gene set variation enrichment meta-analysis of nine bulk transcriptomics datasets, showing the number of
genes that were more highly expressed in TECs than NECs isolated from the indicated tumor type. The bar graph represents the number of gene sets detected in
the tumor type(s) indicated by the dot plot panel below. Five gene sets (displayed on the figure; involved in the displayed processes) were consistently higher
expressed in TECs than NECs (red bar and intersection). HCC, hepatocellular carcinoma; CRCLM, colorectal cancer liver metastasis; CRC, colorectal cancer;
medullo Wnt, Wnt-driven medulloblastoma; medullo Shh, sonic hedgehog driven medulloblastoma; RCC, renal cell carcinoma.
(D) Gene expression meta-analysis of the nine NEC versus TEC datasets shown in (C). The S-curve has 10,850 dots, representing genes that were detected in all
nine datasets. x axis, rank numbers from 1 to 10,850 (consistently overexpressed genes in TECs have a low rank number; consistently downregulated genes have
a high rank number); y axis, the scaled meta-analysis score (consistently overexpressed genes in TECs have a low meta-analysis score; consistently down-
regulated genes have a high meta-analysis score). SQLE and ALDH18A1 are shown as red dots and listed on the left.
See also Figure S5 and Table S5.

(Table S5). The roles of glycolysis (De Bock et al., 2013),
OXPHOS (Diebold et al., 2019), fatty acid oxidation (Schoors
et al., 2015), serine metabolism (Vandekeere et al., 2018), and
glutamine metabolism (Huang et al., 2017) in biomass synthesis
and EC proliferation have been established, thus validating the
predictive potential of the CEC-tailored GEM. However, a
possible role of cholesterol synthesis in EC growth has only mini-
mally been studied, without conclusive results. Further, consis-
tent with previous reports (Phang, 2019), the CEC-tailored
GEM predicted the essentiality of proline biosynthesis for
collagen production. In addition, glutamine and glutamate trans-
porters, and enzymes involved in alanine, glycine, and serine
metabolism, were also predicted to be essential, reflecting that
the molecular composition of collagen consists of >70% of (hy-
droxy)proline, glycine, glutamate, and alanine (Eastoe, 1955).
Integrated Analysis to Prioritize Metabolic Genes for
Functional Validation
To identify functionally relevant metabolic angiogenic targets in
CNV-ECs, we performed an integrated analysis (Figure S5A).
Specifically, we focused on the subset of rate-limiting single
gene encoded metabolic enzymes, predicted by GEM to be
essential for biomass production or collagen biosynthesis (Fig-
ure 5B; Table S5). We selected candidates that were more highly
expressed in CNV-ECs than CECs, yielding 30 genes for
biomass production and 4 for collagen synthesis (Table S5). Sec-
ond, we reasoned that genes that were conserved across
models, diseases, and species represent robust targets. We
thus filtered for genes consistently upregulated also in tumor
angiogenesis, resulting in a set of 17 candidate genes. Finally,
unbiased meta-analysis across 9 different murine and human

870 Cell Metabolism 31, 862–877, April 7, 2020

 A B C

E F G

Figure 6. Metabolic Target Validation

(A) 3H-thymidine incorporation in DNA assay upon SQLE (mean ± SEM, n = 16, *p < 0.05, unpaired two-tailed t test) or ALDH18A1 (mean ± SEM, n = 3, *p < 0.05,
unpaired two-tailed t test) silencing (KD denotes shRNA knockdown).
(B) Scratch wound migration assay with control and SQLE (mean ± SEM, n = 7, *p < 0.05, unpaired two-tailed t test) or ALDH18A1 (mean ± SEM, n = 3, *p < 0.05,
unpaired two-tailed t test) silenced ECs.
(legend continued on next page)
Cell Metabolism 31, 862–877, April 7, 2020 871
datasets revealed that Aldh18a1 and Sqle were among the
most consistently induced genes in TECs (Figures 5C and 5D;
Table S5). Aldh18a1 encodes pyrroline-5-carboxylate-synthase
(P5CS), the rate-controlling enzyme of proline and collagen
biosynthesis, a pathway that was consistently upregulated in
angiogenic ECs in tumors and CNV (Figures S5B and S5C).
Sqle encodes squalene monooxygenase, a rate-limiting enzyme
in cholesterol biosynthesis (Cerqueira et al., 2016).
Functional Validation of Selected Targets
To functionally validate the role of ALDH18A1 and SQLE in vessel
sprouting, we silenced these genes (Figures S5D–S5G), which
impaired EC proliferation and migration, vessel sprouting (Fig-
ures 6A–6D), and EC tip cell competitivity (Figures 6E and 6F).
To explore global expression changes induced by SQLE and
ALDH18A1 silencing, we performed multi-omics analysis. Tran-
scriptomics analysis confirmed decreased expression of genes
involved in cell proliferation and DNA replication in silenced cells
(Figures S6A and S6B). These results were corroborated at the
protein level by proteomics analysis (Figure S6C). Targeted me-
tabolomics showed that SQLE and ALDH18A1 silencing did not
affect the energy charge (Figure S6D), but confirmed that
ALDH18A1 knockdown lowered the levels of proline and hy-
droxyproline, a surrogate for collagen content (Stoilov et al.,
2018) (Figures S6E and S6F). Finally, in vivo intraocular treatment
with siRNAs against Sqle and Aldh18a1 decreased the expres-
sion levels of these targets in CNV-ECs (Figure S6G) and in-
hibited neovascularization of laser-induced choroid (Figure 6G).
To explore potential therapeutic relevance, we used NB-598 to
pharmacologically inhibit SQLE in vivo and observed a reduction
in both corneal angiogenesis and CNV (Figure 6H).
DISCUSSION
Objectives of this study were (1) to construct a taxonomy of
CECs and CNV-ECs, (2) to characterize the heterogeneity of
metabolic gene expression signatures of different ECs at the sin-
gle-cell level, (3) to explore if and how angiogenic ECs reprogram
their metabolic transcriptome signature during pathological
angiogenesis, and (4) to design an integrated analytical
approach to identify angiogenic targets regulating EC meta-
bolism, not previously known to co-determine pathological
angiogenesis.
We constructed a CEC/CNV-EC taxonomy and identified 18
choroidal EC phenotypes, of which 8 were previously not recog-
nized, and characterized transcription factors, predicted to be
involved in the differentiation of different choroidal ECs. The
availability of this choroidal CEC/CNV-EC taxonomy, as well as
a previously in-house constructed lung NEC/TEC taxonomy
(Goveia et al., 2020), which we now used to analyze metabolic
gene expression, enabled us to characterize metabolic gene
expression heterogeneity in sprouting ECs across two tissues
and diseases.
The metabolic transcriptome diversity was most outspoken for
TECs and, remarkably, TEC phenotypes expressed distinct
metabolic transcriptome signatures. Presumably, this complex
metabolic transcriptome heterogeneity is required for EC pheno-
types to execute their specialized functions in different vascular
compartments. However, given that the micro-environment can
influence cellular metabolism (Muir et al., 2018), part of the het-
erogeneity can possibly also be attributed to different environ-
mental signals in distinct vascular compartments. This may
partly explain why tip and immature ECs had a stronger glyco-
lytic gene signature in tumors than in CNV, possibly evoked by
the harsher nutrient-deprived environment and larger abun-
dance of metabolism-altering signals in the tumor as compared
to the CNV lesions (Lyssiotis and Kimmelman, 2017).
The EC taxonomies in the eye and lung offer integrated com-
parison of EC phenotypes across tissues and diseases. Even
though tissue-type-specific EC phenotypes were recognized,
angiogenic ECs in CNV and tumors seemed to form and maintain
neovessels by developing similar EC phenotypes, including
proliferating, tip, immature, and neophalanx ECs. Across dis-
eases and tissues, angiogenic ECs congruently upregulated
the expression of non-metabolic marker genes, as well as meta-
bolic genes involved in OXPHOS, glycolysis, TCA cycle, nucleo-
tide biosynthesis, and mRNA metabolism. Hence, blood vessels
seem to sprout by inducing the differentiation of largely similar
EC phenotypes across diseases and tissues. Tumors have
been named ‘‘non-healing wounds’’ and angiogenesis of tumors
may thus build on similar principles as a CNV wound (Dvorak,
1986). Nonetheless, tissue- and disease-specific differences in
metabolic transcriptome profiles were observed, likely in part
due to differences in micro-environmental conditions.
Trajectory inference analysis predicted that the hierarchy of
angiogenic phenotypes is derived from pcv ECs and that pheno-
typic differentiation is associated with plastic metabolic gene
expression reprogramming, suggesting that metabolic plasticity
supports phenotypic plasticity. Early morphological studies sug-
gested that vessel sprouting originates from this venous vascular
bed (Folkman, 1982), but using in silico lineage tracing, we iden-
tified a differentiation trajectory of venous ECs over intermediate
transitioning immature ECs to tip ECs and later to more mature
neophalanx ECs.
Another objective of this study was to explore whether meta-
bolic signatures of individual ECs could be utilized for the discov-
ery of metabolic genes, previously not recognized to fuel vessel

(C) Bright field photographs of control, ALDH18A1KD, and SQLEKD EC spheroids. Scale bar, 100 mm.
(D) Morphometric quantification of the number of sprouts, average, and cumulative sprout length for control, SQLEKD, and ALDH18A1KD spheroids with or without
MitoC treatment (mean ± SEM, n = 3 for all parameters; ns, p > 0.05; *p < 0.05, unpaired two-tailed t test).
(E and F) Mosaic EC spheroid competition of control (red), SQLEKD (E), or ALDH18A1KD (F) (green) ECs. Quantification of the fraction of tip cells with the indicated
genotype. Data are mean ± SEM, n = 3, *p < 0.05 by unpaired two-tailed t test.
(G) Quantification of CNV blood vessel area in mice treated with control siRNA (CTRL) or siRNA against murine Sqle or Aldh18a1. Representative images are
shown on the right. Scale bars, 75 mm. Data are mean ± SEM, n = 3 independent experiments each using six mice per group, *p < 0.05, unpaired two-tailed t test.
(H) Quantification of CNV (left) and corneal angiogenesis upon corneal cauterization-induced injury (right) in mice treated with vehicle (CTRL) or NB-598 (an SQLE
blocker). Representative micrographs are shown on the right. Scale bars, 75 mm (CNV) and 500 mm (cornea). Data are mean ± SEM, n = 3 independent ex-
periments each using six mice per group, *p < 0.05, unpaired two-tailed t test.
See also Figures S5 and S6.

872 Cell Metabolism 31, 862–877, April 7, 2020

sprouting. Hypothesizing that genes that were conservedly upre- B Mouse model of Lewis Lung Carcinoma
gulated across diseases and species represent robust candi- B Cell Lines and Primary Cell Culture
dates, we performed an integrated multi-layered approach d METHOD DETAILS
combining scRNA-seq of ECs from different tissues/diseases, B In Vitro Functional Assays
congruency transcriptome analysis, genome-scale metabolic B Knockdown Strategy
modeling, and cross-species meta-analysis to identify B Treatment with Inhibitor in the Murine Ocular Models
conserved angiogenic metabolic targets, i.e., Sqle for biomass B Murine Choroid Immunostaining
synthesis and Aldh18a1 for collagen synthesis. Functional vali- B RNA Isolation and Quantitative RT-PCR
dation revealed that silencing of these targets impaired vessel B Single-cell Droplet-based RNA Sequencing
sprouting in vitro and inhibited pathological ocular angiogenesis B Single-cell Transcriptomics Analysis
in vivo. While it was not the primary goal to develop new AAT B Quality Control and Data Normalization
strategies, but rather to provide proof of principle of the inte- B In Silico EC Selection
grated approach, the identified metabolic targets might none- B Batch Effect Correction
theless deserve further attention for AAT development, though B Feature Selection and Dimensionality Reduction
an EC-selective drug delivery approach would then be desirable. B EC Cluster Identification
B Marker Gene Analysis
Limitations of Study B Cluster Annotation
We acknowledge limitations of our study. First, the inferred biolog- B Gene Set Variation Analysis
ical role and topographical localization in the vasculature of each B Evaluation of Dissociation Artifacts
EC phenotype and the computational trajectory inference analysis B Heatmap Analysis
are putative/predicted and require additional marker gene identi- B Trajectory Inference
fication and functional validation to probe their biological roles/ B Jaccard Similarity Analysis
principles and anatomical topography. Second, the contribution B Gene Set Enrichment Analysis
of the resident ESCs to angiogenesis and the precise mechanisms B Metabolic Gene Expression Analysis and Pathway
of how ALDH18A1 and SQLE regulate EC migration/proliferation Mapping
and EC metabolism require further study. Third, transcript levels B Cell Cycle Pseudotime Analysis
and metabolic modeling by GEMs do not fully capture the B Development of a Multivariate Model to Predict Cell
complexity of metabolism (metabolic fluxes, enzyme activity, Cycle Pseudotime
and metabolite levels). However, gene signatures and GEM B Pathway and Transcription Factor Activity Analysis
modeling have been proven to be predictive of the metabolic B Bulk RNA-sequencing Analysis
flux changes in ECs (Bruning et al., 2018; Cantelmo et al., 2016; B Proteomics Sample Preparation
Kalucka et al., 2018; McGarrity et al., 2018; Patella et al., 2015; B Proteomics LC-MS/MS and Data Analysis
Vandekeere et al., 2018). Fourth, we expect that future exploration B Metabolomics and Data Analysis
of our dataset using newly developed pre-processing and analysis B Meta-analysis of Transcriptomics Data
algorithms might further fine-tune our analysis and interpretation. B RNAscope In Situ Hybridization and Quantification
Finally, the current data await future development of novel tech- B Cytometry by Time of Flight Mass Cytometry (CyTOF)
nology to quantify metabolism at the single-cell level in ECs, but B CyTOF Data Analysis
the potential of using an integrated analysis to overcome technical B Curation of a GEM Model
limitations in metabolic target prioritization is demonstrated by the B Development of an EC-tailored GEM
validation of the functional role of the selected candidates B Development of EC-Tailored Objective Functions
(ALDH18A1 and SQLE) in vessel sprouting in vivo. B Development of an Angiogenic CEC-GEM Model
To maximize resource value, we provide accompanying data B Prediction of Essential Genes in the Angiogenic
exploration web tools—available at https://www.vibcancer.be/ CEC Model
software-tools/Murine_ECTax and https://www.vibcancer.be/ d QUANTIFICATION AND STATISTICAL ANALYSIS
software-tools/scCycle, and via the publicly available added- d DATA AND CODE AVAILABILITY
value database EndoDB (Khan et al., 2019). B Data Resources
B Software

STAR+METHODS
Detailed methods are provided in the online version of this paper
and include the following:
d KEY RESOURCES TABLE
d LEAD CONTACT AND MATERIALS AVAILABILITY
d EXPERIMENTAL MODEL AND SUBJECT DETAILS
B Patient Material and Choroid EC Isolation
B Mice
B Mouse Model of Choroidal Neovascularization
B Murine Choroid Endothelial Cell Isolation
SUPPLEMENTAL INFORMATION
Supplemental Information can be found online at https://doi.org/10.1016/j.
cmet.2020.03.009.
ACKNOWLEDGMENTS
We acknowledge A. Bouché, T. Van Brussel, R. Schepers, M. Pellemans, L.
Roussel, and I. Van Hove for technical assistance. We also gratefully acknowl-
edge the expert advice and help of J. Qian, B. Boeckx, and D. Lambrechts.
Metabolomics analysis was done at the Metabolomics Expertise Center,
VIB, Leuven, Belgium. Proteomics analysis was done at the Proteomics

Cell Metabolism 31, 862–877, April 7, 2020 873

Expertise Center, VIB, Ghent, Belgium. Funding was provided by Fonds voor
Wetenschappelijk Onderzoek (FWO; to K.R., J.G., J.K., and L.T.); a Marie Cu-
rie-IEF Fellowship (to K.D.F.); Bettencourt Schueller Foundation (to L.T.); the
Lundbeck Foundation (R219–2016-1375) and DFF Sapere Aude Starting Grant
(8048-00072A) (to L.L.); University of Antwerp (to L.-A.T., in part); Strategisch
Basisonderzoek FWO-Vlaanderen (to V.G.); the Austrian Science Fund (FWF)
(J3730-B26; to A.P.); the Else Kröner-Fresenius-Stiftung and the Fritz Thyssen
Stiftung (10.16.2.017MN; to L.-C.C.); Kom op Tegen Kanker (to S.K.); the
Sanming Project of Medicine in Shenzhen (SZSM201612074), BGI-Research,
the Danish Research Council for Independent Research (DFF–1337–00128),
the Sapere Aude Young Research Talent Prize (DFF-1335–00763A), and Aar-
hus University Strategic Grant (AU-iCRISPR) (to L.B. and Y. Luo); the State Key
Laboratory of Ophthalmology, Zhongshan Ophthalmic Center at the Sun Yat-
Sen University, China, and the National Natural Science Foundation of China
(81670855), a Guangdong Province Leading Expert Program grant, and the
Key Program of Guangzhou Scientific Research Plan (3030901006074) (to
X.L.); and the VIB TechWatch Program, long-term structural Methusalem fund-
ing from the Flemish Government, FWO, Foundation against Cancer (2016-
078), Kom op Tegen Kanker, ERC Proof of Concept (ERC-713758), and ERC
Advanced Research Grant (EU-ERC743074) (to P.C.).
AUTHOR CONTRIBUTIONS
K.R. and J.G. designed and analyzed all experiments. M.G.-C. performed
in vivo experiments. A.S. developed GEMs. J.G., J.K., M.G.-C., L.T., A.P.,
and L.-C.C. set up endothelial cell isolation protocols. L.T. and K.D.F. per-
formed in vitro experiments. F.T. performed bioinformatics analysis and imple-
mented the online databases. N.E. provided human choroid samples. J.K.,
L.P.M.H.d.R., Y.Z., L. Sokol, V.G., D.P., F.D.S., F.J.T.S., R.J.M., and Y. Liu per-
formed CyTOF measurements. K.R., J.K., M.G.-C., L.T., L. Sokol, L.-A.T., V.G.,
and L.-C.C. prepared scRNA-seq samples. L.L., B.T., and Y. Luo performed
10x single-cell sequencing. S.K. processed scRNA-seq data. S.S. and G.C.
performed hydroxyproline measurements. F.D.S., S.V., T.V.B., G.E., P.D.H.,
M.D., L. Schoonjans, J.W., L.B., H.Y., B.T., X.L., and Y. Luo provided advice
and discussed results. K.R., J.G., and P.C. wrote the manuscript. P.C.
conceptualized the study. All authors discussed results and commented on
the manuscript.
DECLARATION OF INTERESTS
The authors declare no competing interests.
Received: July 11, 2019
Revised: December 20, 2019
Accepted: March 9, 2020
Published: April 7, 2020
SUPPORTING CITATIONS
The following references appear in the Supplemental Information: Atkins and
Jain (2007); Bjorklund et al. (2018); Carmeliet and Jain (2011); Chen (2009);
del Toro et al. (2010); dela Paz and D’Amore (2009); Epstein (2016); Hazell
et al. (2016); McCandless et al. (2008); McCormick et al. (2001); Naschberger
et al. (2016); Neve et al. (2014); Patel et al. (2017); Pusztaszeri et al. (2006);
Roca and Adams (2007); Sardana et al. (2017); Strasser et al. (2010); Takaba-
take et al. (2009); Thiriot et al. (2017); Vanlandewijck et al. (2018); Vartanian and
Weidner (1994); Wagenseil and Mecham (2012); Zanetta et al. (2000); Zhao
et al. (2018).
REFERENCES
Aibar, S., González-Blas, C.B., Moerman, T., Huynh-Thu, V.A., Imrichova, H.,
Hulselmans, G., Rambow, F., Marine, J.C., Geurts, P., Aerts, J., et al. (2017).
SCENIC: single-cell regulatory network inference and clustering. Nat.
Methods 14, 1083–1086.
Ambati, J., and Fowler, B.J. (2012). Mechanisms of age-related macular
degeneration. Neuron 75, 26–39.
874 Cell Metabolism 31, 862–877, April 7, 2020
Apostolidis, S.A., Stifano, G., Tabib, T., Rice, L.M., Morse, C.M., Kahaleh, B.,
and Lafyatis, R. (2018). Single cell RNA sequencing identifies HSPG2 and
APLNR as markers of endothelial cell injury in systemic sclerosis skin. Front.
Immunol. 9, 2191.
€ppel-like transcription factors in
Atkins, G.B., and Jain, M.K. (2007). Role of Kru
endothelial biology. Circ. Res. 100, 1686–1695.
Aurich, M.K., Fleming, R.M.T., and Thiele, I. (2016). MetaboTools: a compre-
hensive toolbox for analysis of genome-scale metabolic models. Front.
Physiol. 7, 327.
Barankiewicz, J., Kaplinsky, C., and Cohen, A. (1986). Modification of ribonu-
cleotide and deoxyribonucleotide metabolism in interferon-treated human B-
lymphoblastoid cells. J. Interferon Res. 6, 717–727.
Baydoun, A.R., Emery, P.W., Pearson, J.D., and Mann, G.E. (1990). Substrate-
dependent regulation of intracellular amino acid concentrations in cultured
bovine aortic endothelial cells. Biochem. Biophys. Res. Commun. 173,
940–948.
Becker, S.A., and Palsson, B.O. (2008). Context-specific metabolic networks
are consistent with experiments. PLoS Comput. Biol. 4, e1000082.
Bjorklund, G., Svanberg, E., Dadar, M., Card, D.J., Chirumbolo, S., Harrington,
D.J., and Aaseth, J. (2018). The role of matrix Gla protein (MGP) in vascular
calcification. Curr. Med. Chem. Published July 15, 2018. https://doi.org/10.
2174/0929867325666180716104159.
Blaauwgeers, H.G., Holtkamp, G.M., Rutten, H., Witmer, A.N., Koolwijk, P.,
Partanen, T.A., Alitalo, K., Kroon, M.E., Kijlstra, A., van Hinsbergh, V.W., and
Schlingemann, R.O. (1999). Polarized vascular endothelial growth factor
secretion by human retinal pigment epithelium and localization of vascular
endothelial growth factor receptors on the inner choriocapillaris. Evidence
for a trophic paracrine relation. Am. J. Pathol. 155, 421–428.
Bradshaw, A.D., and Sage, E.H. (2001). SPARC, a matricellular protein that
functions in cellular differentiation and tissue response to injury. J. Clin.
Invest. 107, 1049–1054.
Bruning, U., Morales-Rodriguez, F., Kalucka, J., Goveia, J., Taverna, F.,
Queiroz, K.C.S., Dubois, C., Cantelmo, A.R., Chen, R., Loroch, S., et al.
(2018). Impairment of angiogenesis by fatty acid synthase inhibition involves
mTOR malonylation. Cell Metab. 28, 866–880.e15.
Cannoodt, R., Saelens, W., Sichien, D., Tavernier, S., Janssens, S., Guilliams,
M., Lambrecht, B., Preter, K.D., and Saeys, Y. (2016). SCORPIUS improves
trajectory inference and identifies novel modules in dendritic cell development.
bioRxiv. https://doi.org/10.1101/079509.
Cantelmo, A.R., Conradi, L.C., Brajic, A., Goveia, J., Kalucka, J., Pircher, A.,
Chaturvedi, P., Hol, J., Thienpont, B., Teuwen, L.A., et al. (2016). Inhibition
of the glycolytic activator PFKFB3 in endothelium induces tumor vessel
normalization, impairs metastasis, and improves chemotherapy. Cancer Cell
30, 968–985.
Carmeliet, P., and Jain, R.K. (2011). Molecular mechanisms and clinical appli-
cations of angiogenesis. Nature 473, 298–307.
Cerqueira, N.M.F.S.A., Oliveira, E.F., Gesto, D.S., Santos-Martins, D., Moreira,
C., Moorthy, H.N., Ramos, M.J., and Fernandes, P.A. (2016). Cholesterol
biosynthesis: a mechanistic overview. Biochemistry 55, 5483–5506.
Chambers, M.C., Maclean, B., Burke, R., Amodei, D., Ruderman, D.L.,
Neumann, S., Gatto, L., Fischer, B., Pratt, B., Egertson, J., et al. (2012). A
cross-platform toolkit for mass spectrometry and proteomics. Nat.
Biotechnol. 30, 918–920.
Chen, L. (2009). Ocular lymphatics: state-of-the-art review. Lymphology
42, 66–76.
Corada, M., Orsenigo, F., Morini, M.F., Pitulescu, M.E., Bhat, G., Nyqvist, D.,
Breviario, F., Conti, V., Briot, A., Iruela-Arispe, M.L., et al. (2013). Sox17 is
indispensable for acquisition and maintenance of arterial identity. Nat.
Commun. 4, 2609.
Corey, D.M., Rinkevich, Y., and Weissman, I.L. (2016). Dynamic patterns of
clonal evolution in tumor vasculature underlie alterations in lymphocyte-endo-
thelial recognition to foster tumor immune escape. Cancer Res. 76,
1348–1353.
Cox, J., and Mann, M. (2008). MaxQuant enables high peptide identification
rates, individualized p.p.b.-range mass accuracies and proteome-wide pro-
tein quantification. Nat. Biotechnol. 26, 1367–1372.
Creemers, L.B., Jansen, D.C., van Veen-Reurings, A., van den Bos, T., and
Everts, V. (1997). Microassay for the assessment of low levels of hydroxypro-
line. Biotechniques 22, 656–658.
De Bock, K., Georgiadou, M., Schoors, S., Kuchnio, A., Wong, B.W.,
Cantelmo, A.R., Quaegebeur, A., Ghesquière, B., Cauwenberghs, S., Eelen,
G., et al. (2013). Role of PFKFB3-driven glycolysis in vessel sprouting. Cell
154, 651–663.
de Lichtenberg, U., Jensen, L.J., Fausbøll, A., Jensen, T.S., Bork, P., and
Brunak, S. (2005). Comparison of computational methods for the identification
of cell cycle-regulated genes. Bioinformatics 21, 1164–1171.
del Toro, R., Prahst, C., Mathivet, T., Siegfried, G., Kaminker, J.S., Larrivee, B.,
Breant, C., Duarte, A., Takakura, N., Fukamizu, A., et al. (2010). Identification
and functional analysis of endothelial tip cell-enriched genes. Blood 116,
4025–4033.
dela Paz, N.G., and D’Amore, P.A. (2009). Arterial versus venous endothelial
cells. Cell Tissue Res. 335, 5–16.
Desouki, A.A., Jarre, F., Gelius-Dietrich, G., and Lercher, M.J. (2015).
CycleFreeFlux: efficient removal of thermodynamically infeasible loops from
flux distributions. Bioinformatics 31, 2159–2165.
Diebold, L.P., Gil, H.J., Gao, P., Martinez, C.A., Weinberg, S.E., and Chandel,
N.S. (2019). Mitochondrial complex III is necessary for endothelial cell prolifer-
ation during angiogenesis. Nat Metab 1, 158–171.
Duarte, N.C., Becker, S.A., Jamshidi, N., Thiele, I., Mo, M.L., Vo, T.D., Srivas,
R., and Palsson, B.Ø. (2007). Global reconstruction of the human metabolic
network based on genomic and bibliomic data. Proc. Natl. Acad. Sci. USA
104, 1777–1782.
Dvorak, H.F. (1986). Tumors: wounds that do not heal. Similarities between tu-
mor stroma generation and wound healing. N. Engl. J. Med. 315, 1650–1659.
Eales, K.L., Hollinshead, K.E., and Tennant, D.A. (2016). Hypoxia and meta-
bolic adaptation of cancer cells. Oncogenesis 5, e190.
Eastoe, J.E. (1955). The amino acid composition of mammalian collagen and
gelatin. Biochem. J. 61, 589–600.
Eelen, G., de Zeeuw, P., Treps, L., Harjes, U., Wong, B.W., and Carmeliet, P.
(2018). Endothelial cell metabolism. Physiol. Rev. 98, 3–58.
Epstein, M. (2016). Matrix Gla-protein (MGP) not only inhibits calcification in
large arteries but also may be renoprotective: connecting the dots.
EBioMedicine 4, 16–17.
Félétou, M., Huang, Y., and Vanhoutte, P.M. (2011). Endothelium-mediated
control of vascular tone: COX-1 and COX-2 products. Br. J. Pharmacol. 164,
894–912.
Fitzgerald, G., Soro-Arnaiz, I., and De Bock, K. (2018). The Warburg effect in
endothelial cells and its potential as an anti-angiogenic target in cancer.
Front. Cell Dev. Biol. 6, 100.
Folkman, J. (1982). Angiogenesis: initiation and control. Ann. N Y Acad. Sci.
401, 212–227.
Garcı́a-Caballero, M., Zecchin, A., Souffreau, J., Truong, A.-C.K., Teuwen,
L.-A., Vermaelen, W., Martı́n-Pérez, R., de Zeeuw, P., Bouché, A., Vinckier,
S., et al. (2019). Role and therapeutic potential of dietary ketone bodies in
lymph vessel growth. Nature Metabolism 1, 666–675.
Goveia, J., Rohlenova, K., Taverna, F., Treps, L., Conradi, L.C., Pircher, A.,
Geldhof, V., de Rooij, L.P.M.H., Kalucka, J., Sokol, L., et al. (2020). An inte-
grated gene expression landscape profiling approach to identify lung tumor
endothelial cell heterogeneity and angiogenic candidates. Cancer Cell 37,
21–36.e13.
Haghverdi, L., Lun, A.T.L., Morgan, M.D., and Marioni, J.C. (2018). Batch ef-
fects in single-cell RNA-sequencing data are corrected by matching mutual
nearest neighbors. Nat. Biotechnol. 36, 421–427.
€nzelmann, S., Castelo, R., and Guinney, J. (2013). GSVA: gene set variation
Ha
analysis for microarray and RNA-seq data. BMC Bioinformatics 14, 7.
Haraldsdóttir, H.S., Cousins, B., Thiele, I., Fleming, R.M.T., and Vempala, S.
(2017). CHRR: coordinate hit-and-run with rounding for uniform sampling of
constraint-based models. Bioinformatics 33, 1741–1743.
Hart, T., Komori, H.K., LaMere, S., Podshivalova, K., and Salomon, D.R.
(2013). Finding the active genes in deep RNA-seq gene expression studies.
BMC Genomics 14, 778.
Hazell, G.G., Peachey, A.M., Teasdale, J.E., Sala-Newby, G.B., Angelini, G.D.,
Newby, A.C., and White, S.J. (2016). PI16 is a shear stress and inflammation-
regulated inhibitor of MMP2. Sci. Rep. 6, 39553.
Heindl, L.M., Kaser-Eichberger, A., Schlereth, S.L., Bock, F., Regenfuss, B.,
Reitsamer, H.A., McMenamin, P., Lutty, G.A., Maruyama, K., Chen, L., et al.
(2015). Sufficient evidence for lymphatics in the developing and adult human
choroid? Invest. Ophthalmol. Vis. Sci. 56, 6709–6710.
Heirendt, L., Arreckx, S., Pfau, T., Mendoza, S.N., Richelle, A., Heinken, A.,
Haraldsdóttir, H.S., Wachowiak, J., Keating, S.M., Vlasov, V., et al. (2019).
Creation and analysis of biochemical constraint-based models using the
COBRA Toolbox v.3.0. Nat. Protoc. 14, 639–702.
Hong, F., Breitling, R., McEntee, C.W., Wittner, B.S., Nemhauser, J.L., and
Chory, J. (2006). RankProd: a bioconductor package for detecting differentially
expressed genes in meta-analysis. Bioinformatics 22, 2825–2827.
Huang, H., Vandekeere, S., Kalucka, J., Bierhansl, L., Zecchin, A., Bru €ning, U.,
Visnagri, A., Yuldasheva, N., Goveia, J., Cruys, B., et al. (2017). Role of gluta-
mine and interlinked asparagine metabolism in vessel formation. EMBO J. 36,
2334–2352.
Hwangbo, C., Wu, J., Papangeli, I., Adachi, T., Sharma, B., Park, S., Zhao, L.,
Ju, H., Go, G.W., Cui, G., et al. (2017). Endothelial APLNR regulates tissue fatty
acid uptake and is essential for apelin’s glucose-lowering effects. Sci. Transl.
Med. 9, https://doi.org/10.1126/scitranslmed.aad4000.
Iñiguez, M.A., Rodrı́guez, A., Volpert, O.V., Fresno, M., and Redondo, J.M.
(2003). Cyclooxygenase-2: a therapeutic target in angiogenesis. Trends Mol.
Med. 9, 73–78.
Jain, R.K. (2014). Antiangiogenesis strategies revisited: from starving tumors
to alleviating hypoxia. Cancer Cell 26, 605–622.
Jeong, H.W., Hernández-Rodrı́guez, B., Kim, J., Kim, K.P., Enriquez-Gasca,
R., Yoon, J., Adams, S., Schöler, H.R., Vaquerizas, J.M., and Adams, R.H.
(2017). Transcriptional regulation of endothelial cell behavior during sprouting
angiogenesis. Nat. Commun. 8, 726.
€ning, U.,
Kalucka, J., Bierhansl, L., Conchinha, N.V., Missiaen, R., Elia, I., Bru
Scheinok, S., Treps, L., Cantelmo, A.R., Dubois, C., et al. (2018). Quiescent
endothelial cells upregulate fatty acid b-oxidation for vasculoprotection via
redox homeostasis. Cell Metab. 28, 881–894.e13.
Kanda, T., Brown, J.D., Orasanu, G., Vogel, S., Gonzalez, F.J., Sartoretto, J.,
Michel, T., and Plutzky, J. (2009). PPARgamma in the endothelium regulates
metabolic responses to high-fat diet in mice. J. Clin. Invest. 119, 110–124.
Khan, S., Taverna, F., Rohlenova, K., Treps, L., Geldhof, V., de Rooij, L., Sokol,
L., Pircher, A., Conradi, L.C., Kalucka, J., et al. (2019). EndoDB: a database of
endothelial cell transcriptomics data. Nucleic Acids Res. 47, D736–D744.
Kim, M.K., and Lun, D.S. (2014). Methods for integration of transcriptomic data
in genome-scale metabolic models. Comput. Struct. Biotechnol. J. 11, 59–65.
Koina, M.E., Baxter, L., Adamson, S.J., Arfuso, F., Hu, P., Madigan, M.C., and
Chan-Ling, T. (2015). Evidence for lymphatics in the developing and adult hu-
man choroid. Invest. Ophthalmol. Vis. Sci. 56, 1310–1327.
Lambert, V., Lecomte, J., Hansen, S., Blacher, S., Gonzalez, M.L., Struman, I.,
Sounni, N.E., Rozet, E., de Tullio, P., Foidart, J.M., et al. (2013). Laser-induced
choroidal neovascularization model to study age-related macular degenera-
tion in mice. Nat. Protoc. 8, 2197–2211.
Langmead, B., and Salzberg, S.L. (2012). Fast gapped-read alignment with
Bowtie 2. Nat. Methods 9, 357–359.
Levandowsky, M., and Winter, D. (1971). Distance between Sets. Nature
234, 34.
Li, X., Kumar, A., and Carmeliet, P. (2019). Metabolic pathways fueling the
endothelial cell drive. Annu. Rev. Physiol. 81, 483–503.
Cell Metabolism 31, 862–877, April 7, 2020 875
Luo, W., and Brouwer, C. (2013). Pathview: an R/Bioconductor package for
pathway-based data integration and visualization. Bioinformatics 29,
1830–1831.
Lyssiotis, C.A., and Kimmelman, A.C. (2017). Metabolic interactions in the tu-
mor microenvironment. Trends Cell Biol. 27, 863–875.
Mahadevan, R., and Schilling, C.H. (2003). The effects of alternate optimal so-
lutions in constraint-based genome-scale metabolic models. Metab. Eng. 5,
264–276.
Manavski, Y., Lucas, T., Glaser, S.F., Dorsheimer, L., Gu€nther, S., Braun, T.,
Rieger, M.A., Zeiher, A.M., Boon, R.A., and Dimmeler, S. (2018). Clonal expan-
sion of endothelial cells contributes to ischemia-induced neovascularization.
Circ. Res. 122, 670–677.
Marneros, A.G., She, H., Zambarakji, H., Hashizume, H., Connolly, E.J., Kim, I.,
Gragoudas, E.S., Miller, J.W., and Olsen, B.R. (2007). Endogenous endostatin
inhibits choroidal neovascularization. FASEB J. 21, 3809–3818.
McCandless, E.E., Piccio, L., Woerner, B.M., Schmidt, R.E., Rubin, J.B.,
Cross, A.H., and Klein, R.S. (2008). Pathological expression of CXCL12 at
the blood-brain barrier correlates with severity of multiple sclerosis. Am. J.
Pathol. 172, 799–808.
McCormick, S.M., Eskin, S.G., McIntire, L.V., Teng, C.L., Lu, C.M., Russell,
C.G., and Chittur, K.K. (2001). DNA microarray reveals changes in gene
expression of shear stressed human umbilical vein endothelial cells. Proc.
Natl. Acad. Sci. USA 98, 8955–8960.
McDonald, A.I., Shirali, A.S., Aragón, R., Ma, F., Hernandez, G., Vaughn, D.A.,
Mack, J.J., Lim, T.Y., Sunshine, H., Zhao, P., et al. (2018). Endothelial regen-
eration of large vessels is a biphasic process driven by local cells with distinct
proliferative capacities. Cell Stem Cell 23, 210–225.e6.
McGarrity, S., Anuforo, Ó., Halldórsson, H., Bergmann, A., Halldórsson, S.,
Palsson, S., Henriksen, H.H., Johansson, P.I., and Rolfsson, Ó. (2018).
Metabolic systems analysis of LPS induced endothelial dysfunction applied
to sepsis patient stratification. Sci. Rep. 8, 6811.
McLeod, D.S., Lefer, D.J., Merges, C., and Lutty, G.A. (1995). Enhanced
expression of intracellular adhesion molecule-1 and P-selectin in the diabetic
human retina and choroid. Am. J. Pathol. 147, 642–653.
Mondor, I., Jorquera, A., Sene, C., Adriouch, S., Adams, R.H., Zhou, B.,
Wienert, S., Klauschen, F., and Bajénoff, M. (2016). Clonal proliferation and
stochastic pruning orchestrate lymph node vasculature remodeling.
Immunity 45, 877–888.
Muir, A., Danai, L.V., and Vander Heiden, M.G. (2018). Microenvironmental
regulation of cancer cell metabolism: implications for experimental design
and translational studies. Dis. Model. Mech. 11, https://doi.org/10.1242/
dmm.035758.
Murphy, E.J., Joseph, L., Stephens, R., and Horrocks, L.A. (1992).
Phospholipid composition of cultured human endothelial cells. Lipids 27,
150–153.
Naito, H., Kidoya, H., Sakimoto, S., Wakabayashi, T., and Takakura, N. (2012).
Identification and characterization of a resident vascular stem/progenitor cell
population in preexisting blood vessels. EMBO J. 31, 842–855.
€tz, M., Britzen-Laurent, N.,
Naschberger, E., Liebl, A., Schellerer, V.S., Schu
Kölbel, P., Schaal, U., Haep, L., Regensburger, D., Wittmann, T., et al.
(2016). Matricellular protein SPARCL1 regulates tumor microenvironment-
dependent endothelial cell heterogeneity in colorectal carcinoma. J. Clin.
Invest. 126, 4187–4204.
Neve, A., Cantatore, F.P., Maruotti, N., Corrado, A., and Ribatti, D. (2014).
Extracellular matrix modulates angiogenesis in physiological and pathological
conditions. BioMed Res. Int. 2014, 756078.
O’Brien, E.J., Monk, J.M., and Palsson, B.O. (2015). Using genome-scale
models to predict biological capabilities. Cell 161, 971–987.
Pagliarini, R., Castello, R., Napolitano, F., Borzone, R., Annunziata, P.,
Mandrile, G., De Marchi, M., Brunetti-Pierri, N., and di Bernardo, D. (2016).
In silico modeling of liver metabolism in a human disease reveals a key enzyme
for histidine and histamine homeostasis. Cell Rep. 15, 2292–2300.
Patel, J., Seppanen, E.J., Rodero, M.P., Wong, H.Y., Donovan, P., Neufeld, Z.,
Fisk, N.M., Francois, M., and Khosrotehrani, K. (2017). Functional definition of
876 Cell Metabolism 31, 862–877, April 7, 2020
progenitors versus mature endothelial cells reveals key SoxF-dependent dif-
ferentiation process. Circulation 135, 786–805.
Patella, F., Schug, Z.T., Persi, E., Neilson, L.J., Erami, Z., Avanzato, D.,
Maione, F., Hernandez-Fernaud, J.R., Mackay, G., Zheng, L., et al. (2015).
Proteomics-based metabolic modeling reveals that fatty acid oxidation
(FAO) controls endothelial cell (EC) permeability. Mol. Cell. Proteomics 14,
621–634.
Perez-Riverol, Y., Csordas, A., Bai, J., Bernal-Llinares, M., Hewapathirana, S.,
Kundu, D.J., Inuganti, A., Griss, J., Mayer, G., Eisenacher, M., et al. (2019). The
PRIDE database and related tools and resources in 2019: improving support
for quantification data. Nucleic Acids Res. 47 (D1), D442–D450.
Phang, J.M. (2019). Proline metabolism in cell regulation and cancer biology:
recent advances and hypotheses. Antioxid. Redox Signal. 30, 635–649.
Potente, M., Gerhardt, H., and Carmeliet, P. (2011). Basic and therapeutic as-
pects of angiogenesis. Cell 146, 873–887.
Pusztaszeri, M.P., Seelentag, W., and Bosman, F.T. (2006).
Immunohistochemical expression of endothelial markers CD31, CD34, von
Willebrand factor, and Fli-1 in normal human tissues. J. Histochem.
Cytochem. 54, 385–395.
Rastogi, B.K., and Nordøy, A. (1980). Lipid composition of cultured human
endothelial cells. Thromb. Res. 18, 629–641.
Red-Horse, K., Ueno, H., Weissman, I.L., and Krasnow, M.A. (2010). Coronary
arteries form by developmental reprogramming of venous cells. Nature 464,
549–553.
Ritchie, M.E., Phipson, B., Wu, D., Hu, Y., Law, C.W., Shi, W., and Smyth, G.K.
(2015). limma powers differential expression analyses for RNA-sequencing
and microarray studies. Nucleic Acids Res. 43, e47.
Roca, C., and Adams, R.H. (2007). Regulation of vascular morphogenesis by
Notch signaling. Genes Dev. 21, 2511–2524.
Ryu, J.Y., Kim, H.U., and Lee, S.Y. (2015). Reconstruction of genome-scale
human metabolic models using omics data. Integr. Biol. 7, 859–868.
Santos, C.R., and Schulze, A. (2012). Lipid metabolism in cancer. FEBS J. 279,
2610–2623.
Santos, A., Wernersson, R., and Jensen, L.J. (2015). Cyclebase 3.0: a multi-or-
ganism database on cell-cycle regulation and phenotypes. Nucleic Acids Res.
43, D1140–D1144.
Sardana, M., Vasim, I., Varakantam, S., Kewan, U., Tariq, A., Koppula, M.R.,
Syed, A.A., Beraun, M., Drummen, N.E., Vermeer, C., et al. (2017). Inactive ma-
trix Gla-protein and arterial stiffness in type 2 diabetes mellitus. Am. J.
Hypertens. 30, 196–201.
Satija, R., Farrell, J.A., Gennert, D., Schier, A.F., and Regev, A. (2015). Spatial
reconstruction of single-cell gene expression data. Nat. Biotechnol. 33,
495–502.
Sawada, N., and Arany, Z. (2017). Metabolic regulation of angiogenesis in dia-
betes and aging. Physiology (Bethesda) 32, 290–307.
Schoors, S., De Bock, K., Cantelmo, A.R., Georgiadou, M., Ghesquière, B.,
Cauwenberghs, S., Kuchnio, A., Wong, B.W., Quaegebeur, A., Goveia, J.,
et al. (2014). Partial and transient reduction of glycolysis by PFKFB3 blockade
reduces pathological angiogenesis. Cell Metab. 19, 37–48.
Schoors, S., Bruning, U., Missiaen, R., Queiroz, K.C., Borgers, G., Elia, I.,
Zecchin, A., Cantelmo, A.R., Christen, S., Goveia, J., et al. (2015). Fatty acid
carbon is essential for dNTP synthesis in endothelial cells. Nature 520,
192–197.
Scialdone, A., Natarajan, K.N., Saraiva, L.R., Proserpio, V., Teichmann, S.A.,
Stegle, O., Marioni, J.C., and Buettner, F. (2015). Computational assignment
of cell-cycle stage from single-cell transcriptome data. Methods 85, 54–61.
Semenza, G.L. (2011). Hypoxia-inducible factor 1: regulator of mitochondrial
metabolism and mediator of ischemic preconditioning. Biochim. Biophys.
Acta 1813, 1263–1268.
Shlomi, T., Benyamini, T., Gottlieb, E., Sharan, R., and Ruppin, E. (2011).
Genome-scale metabolic modeling elucidates the role of proliferative adapta-
tion in causing the Warburg effect. PLoS Comput. Biol. 7, e1002018.
Stoilov, I., Starcher, B.C., Mecham, R.P., and Broekelmann, T.J. (2018).
Measurement of elastin, collagen, and total protein levels in tissues.
Methods Cell Biol. 143, 133–146.
Strasser, G.A., Kaminker, J.S., and Tessier-Lavigne, M. (2010). Microarray
analysis of retinal endothelial tip cells identifies CXCR4 as a mediator of tip
cell morphology and branching. Blood 115, 5102–5110.
Swainston, N., Smallbone, K., Hefzi, H., Dobson, P.D., Brewer, J., Hanscho,
M., Zielinski, D.C., Ang, K.S., Gardiner, N.J., Gutierrez, J.M., et al. (2016).
Recon 2.2: from reconstruction to model of human metabolism.
Metabolomics 12, 109.
Takabatake, Y., Sugiyama, T., Kohara, H., Matsusaka, T., Kurihara, H., Koni,
P.A., Nagasawa, Y., Hamano, T., Matsui, I., Kawada, N., et al. (2009). The
CXCL12 (SDF-1)/CXCR4 axis is essential for the development of renal vascu-
lature. J. Am. Soc. Nephrol. 20, 1714–1723.
Thiele, I., and Palsson, B.O. (2010). A protocol for generating a high-quality
genome-scale metabolic reconstruction. Nat. Protoc. 5, 93–121.
Thiriot, A., Perdomo, C., Cheng, G., Novitzky-Basso, I., McArdle, S.,
Kishimoto, J.K., Barreiro, O., Mazo, I., Triboulet, R., Ley, K., et al. (2017).
Differential DARC/ACKR1 expression distinguishes venular from non-venular
endothelial cells in murine tissues. BMC Biol. 15, 45.
van den Brink, S.C., Sage, F., Vértesy, Á., Spanjaard, B., Peterson-Maduro, J.,
Baron, C.S., Robin, C., and van Oudenaarden, A. (2017). Single-cell
sequencing reveals dissociation-induced gene expression in tissue subpopu-
lations. Nat. Methods 14, 935–936.
Vandekeere, S., Dubois, C., Kalucka, J., Sullivan, M.R., Garcı́a-Caballero, M.,
Goveia, J., Chen, R., Diehl, F.F., Bar-Lev, L., Souffreau, J., et al. (2018). Serine
synthesis via PHGDH is essential for heme production in endothelial cells. Cell
Metab. 28, 573–587.e13.
Vanlandewijck, M., He, L., Ma €e, M.A., Andrae, J., Ando, K., Del Gaudio, F.,
Nahar, K., Lebouvier, T., Laviña, B., Gouveia, L., et al. (2018). A molecular atlas
of cell types and zonation in the brain vasculature. Nature 554, 475–480.
Vartanian, R.K., and Weidner, N. (1994). Correlation of intratumoral endothelial
cell proliferation with microvessel density (tumor angiogenesis) and tumor cell
proliferation in breast carcinoma. Am. J. Pathol. 144, 1188–1194.
Wagenseil, J.E., and Mecham, R.P. (2012). Elastin in large artery stiffness and
hypertension. J. Cardiovasc. Transl. Res. 5, 264–273.
Wakabayashi, T., Naito, H., Takara, K., Kidoya, H., Sakimoto, S., Oshima, Y.,
Nishida, K., and Takakura, N. (2013). Identification of vascular endothelial side
population cells in the choroidal vessels and their potential role in age-related
macular degeneration. Invest. Ophthalmol. Vis. Sci. 54, 6686–6693.
Wakabayashi, T., Naito, H., Suehiro, J.I., Lin, Y., Kawaji, H., Iba, T., Kouno, T.,
Ishikawa-Kato, S., Furuno, M., Takara, K., et al. (2018). CD157 marks tissue-
resident endothelial stem cells with homeostatic and regenerative properties.
Cell Stem Cell 22, 384–397.e6.
Welti, J., Loges, S., Dimmeler, S., and Carmeliet, P. (2013). Recent molecular
discoveries in angiogenesis and antiangiogenic therapies in cancer. J. Clin.
Invest. 123, 3190–3200.
Whitfield, M.L., Sherlock, G., Saldanha, A.J., Murray, J.I., Ball, C.A., Alexander,
K.E., Matese, J.C., Perou, C.M., Hurt, M.M., Brown, P.O., and Botstein, D.
(2002). Identification of genes periodically expressed in the human cell cycle
and their expression in tumors. Mol. Biol. Cell 13, 1977–2000.
Yang, S., Zhao, J., and Sun, X. (2016). Resistance to anti-VEGF therapy in neo-
vascular age-related macular degeneration: a comprehensive review. Drug
Des. Devel. Ther. 10, 1857–1867.
Ying, Y., Ueta, T., Jiang, S., Lin, H., Wang, Y., Vavvas, D., Wen, R., Chen, Y.G.,
and Luo, Z. (2017). Metformin inhibits ALK1-mediated angiogenesis via activa-
tion of AMPK. Oncotarget 8, 32794–32806.
You, L.R., Lin, F.J., Lee, C.T., DeMayo, F.J., Tsai, M.J., and Tsai, S.Y. (2005).
Suppression of Notch signalling by the COUP-TFII transcription factor regu-
lates vein identity. Nature 435, 98–104.
Yu, G., Wang, L.G., Han, Y., and He, Q.Y. (2012). clusterProfiler: an R package
for comparing biological themes among gene clusters. OMICS 16, 284–287.
Yu, P., Alves, T.C., Kibbey, R.G., and Simons, M. (2018). Metabolic analysis of
lymphatic endothelial cells. Methods Mol. Biol. 1846, 325–334.
Zanetta, L., Marcus, S.G., Vasile, J., Dobryansky, M., Cohen, H., Eng, K.,
Shamamian, P., and Mignatti, P. (2000). Expression of Von Willebrand factor,
an endothelial cell marker, is up-regulated by angiogenesis factors: a potential
method for objective assessment of tumor angiogenesis. Int. J. Cancer 85,
281–288.
Zhao, Q., Eichten, A., Parveen, A., Adler, C., Huang, Y., Wang, W., Ding, Y.,
Adler, A., Nevins, T., Ni, M., et al. (2018). Single-cell transcriptome analyses
reveal endothelial cell heterogeneity in tumors and changes following antian-
giogenic treatment. Cancer Res. 78, 2370–2382.
Cell Metabolism 31, 862–877, April 7, 2020 877
 DON’T BE THE
LAST TO KNOW

Give your research a boost with alerts

from Cell Press. You’ll be glad you did.
Get first access with immediate, regular
electronic table of contents (eToCs) alerts
delivered directly to your desktop, free
of charge, that keep you informed of
breakthroughs in your field.
Exciting extra features like video abstracts,
podcasts, and blog posts give you additional
depth and context and you can access news
and commentary, normally not accessible
online, in advance of publication.

Register today
Visit cell.com/alerts
 Viruses in Health and Disease
October 24–26, 2021 — Sitges, Spain

This meeting will bring together research that encompasses basic and translational
Abstract virology. We envision this meeting showcasing the ubiquity of viruses while highlighting
submission commonalities among researchers, from technology through a standard goal of
deadline: understanding how viruses behave and adapt and how this may be exploited for
prediction of the next outbreak or leveraged for therapeutic development. The scientific
June 18, response to SARS-CoV-2/COVID-19 continues to showcase how science has
2021 progressed, as the rapidity of basic and translational research has been second to
none, while also revealing our weaknesses as a population. This meeting is both a
Early celebration of how far virology has come and a forecast of what is next and the inroads
registration to get us there.
deadline:
Our topics will include:
August 13, 1) Viruses Within: defining the virome; interaction with the microbiome;
2021 endogenous retroviruses in health and disease; tools and technologies
2) Viral Pathogens: evolution, genomics, spread, prediction, and pandemic
preparedness of emerging and re-emerging viruses
3) Virus-Host Interaction: immunology, genetics, and host factors
4) Viruses as Therapeutics: diagnostics, phage therapy, viral vectors, and
disease therapy
5) Targeting Viruses: Antivirals, Vaccines, Antibodies: lessons from the past;
successes versus failures

Organizers
Erica Saphire, La Jolla Institute of Immunology, USA
Kanta Subbarao, Doherty Institute, Australia
Amanda Monahan, Scientific Editor, Cell Host & Microbe
Sri Narasimhan, Deputy Editor, Cell

cell-symposia.com/Viruses-2021
cell.com/symposia
 stands for interdisciplinary

iScience is an interdisciplinary
open access journal from Cell Press
iScienceW\ISPZOLZIHZPJHUKHWWSPLKYLZLHYJO[OH[HK]HUJLZHZWLJPÄJÄLSKHJYVZZSPMLWO`ZPJHS
HUKLHY[OZJPLUJLZ0[»ZHUVWLUHJJLZZQV\YUHS^P[OJVU[PU\V\ZW\ISPJH[PVUZVYLZLHYJOPZ
PTTLKPH[LS`HJJLZZPISL6\YUVUVUZLUZLHWWYVHJO[VZ\ITPZZPVUZPZZPTWSLMHZ[HUKMHPYHUK
V\YJVTTP[TLU[[VPU[LNYP[`TLHUZ^LW\ISPZO[YHUZWHYLU[TL[OVKZ^P[OOPNOLKP[VYPHSZ[HUKHYKZ

So if you’re looking for a place to be inspired by science, welcome home.

Submit your paper to iScience.

cell.com/iscience
pub_NanoSIMS_cell-press2_Cell press selection 03/11/2020 14:24 Page 3

Cutting-Edge Research
Instruments for Elemental
& Isotopic Microanalysis
IMS xf series SIMS NanoSIMS 50L LEAP 5000 APT
®

CAMECA IMS 7f series Secondary Ion
Mass Spectrometers (SIMS) can map
major and trace species in biological
samples with high sensitivity and a
lateral resolution reaching ~0.5 μm.
In a recent study, an IMS 7f was used to
localize uranium at very low concen-
tration levels within cells, showing for the
first time the presence of soluble
uranium within the nuclei after 15 min
to 24 hours exposure to concentrations
lower than 100 μM.
The CAMECA NanoSIMS 50L is a
unique imaging Secondary Ion Mass
Spectrometer (SIMS) enabling quanti-
tative measurements as well as 2D and
3D imaging of molecular distribution,
kinetics and fluxes in tissues or at intra-
cellular level through a strategy of stable
isotope or rare element labelling.
14
N
The LEAP 5000 is CAMECA’s state-of-
the-art Atom Probe Tomography (APT)
microscope delivering 3D sub-nanometer
chemical analysis of wide variety of
samples. Specific specimen preparation
methods open the path to a large range
of APT applications in life sciences.

10 µm
Analysis of impurities in chiton teeth: APT
15
N/14N allows 3D atomic-scale analysis of these
impurities in the surrounding matrix in 3D
providing unique insights into the
arrangement of the atoms.
From D. Joester, et al. Organic Materials and Organic /
norganic Heterostructures in Atom Probe Tomography (2012)
Microscopy Today. Vol.20, Issue 3

10 µm

Cardiomyocyte cell cycle activity in the

Uranium distribution inside HepG2 cells
after 30 min. exposure to 100 μM of uranyl
nitrate. Superposition of 238U+, 23Na+ and
40
Ca+ SIMS ion images and 238U mass
spectra. Red arrows indicate 238U
precipitates. White circles delimit the
nucleus. Field of view: 50 x 50 μm2.
From from D. Suhard et al.,
Microsc Res Tech. (2018)
DOI: 10.1002/jemt.23047
exercised heart: 15N-thymidine (labelling
DNA) was administered for 8 weeks to
adult mice undergoing voluntary wheel
running versus sedentary activity. Mass 14N
image shows histological details, while
15
N/14N image demonstrates nuclear 15N
labeling (red, 150% above natural ratio) of Scan the code
a cardio-myocyte while the cytoplasm and or visit
other cells are at natural abundance level
(blue, 0% above natural ratio).
cameca.com/focus/tuto
From A. Vujic et al. Exercise induces new cardiomyocyte for free guides on
generation in the adult mammalian heart. (2018) SIMS & APT
Nat Commun 2018;9:1659

www.cameca.com cameca.info@ametek.com
pub_NanoSIMS_cell-press2_Cell press selection 03/11/2020 14:24 Page 4

Upgrade your knowledge on

microanalytical techniques!

The Essential Guide on

Dynamic Secondary Ion Mass
Spectrometry (SIMS)
Still the most sensitive microanalytical technique available, Dynamic
SIMS provides localized elemental, isotopic and molecular characterization
of the sample surface.
Our free Dynamic SIMS guide provides a simple introduction to the basic principles of this essential
microanalytical technique, outlining its advantages over alternative surface analysis techniques. Four
case studies allow readers to explore current practices and applications of the NanoSIMS and other
dynamic SIMS instruments including research conducted at Harvard Medical School (see extract below).

Scan the code or visit

cameca.com/focus/sims-tuto
to download a free guide
on Dynamic SIMS!

Extract from Dynamic SIMS guide, case study Nr 4:

Protein turnover quantification by feeding animals with a 15N-labeled
precursor amino acid to measure appearance of new protein.
In contrast to previous studies, direct measurement with NanoSIMS
multi-isotope imaging mass spectrometry revealed unexpectedly low
turnover of protein in hair-cell stereocili. Measured volume: 8x8x2μm.

From: Multi-isotope imaging mass spectrometry reveals slow

protein turnover in hair-cell stereocilia. Duan-Sun Zhang et al.
NATURE, vol 481, 26 January 2012

cameca.info@ametek.com
www.cameca.com