Targeting Metastasis-Initiating Cells PDF

ARTICLE
doi:10.1038/nature20791
Targeting metastasis-initiating cells

through the fatty acid receptor CD36
GloriaPascual1, AlexandraAvgustinova1, StefaniaMejetta2, MercMartn1, AndrsCastellanos1,

CamilleStephan-OttoAttolini1, AntoniBerenguer1, NeusPrats1, AgustToll3, JuanAntonioHueto4,
CoroBescs4, LucianoDiCroce2,5,6 & SalvadorAznarBenitah1,5
The fact that the identity of the cells that initiate metastasis in most human cancers is unknown hampers the development
of antimetastatic therapies. Here we describe a subpopulation of CD44bright cells in human oral carcinomas that do not
overexpress mesenchymal genes, are slow-cycling, express high levels of the fatty acid receptor CD36 and lipid metabolism
genes, and are unique in their ability to initiate metastasis. Palmitic acid or a high-fat diet specifically boosts the metastatic
potential of CD36+ metastasis-initiating cells in a CD36-dependent manner. The use of neutralizing antibodies to block
CD36 causes almost complete inhibition of metastasis in immunodeficient or immunocompetent orthotopic mouse models
of human oral cancer, with no side effects. Clinically, the presence of CD36+ metastasis-initiating cells correlates with
a poor prognosis for numerous types of carcinomas, and inhibition of CD36 also impairs metastasis, at least in human
melanoma- and breast cancer-derived tumours. Together, our results indicate that metastasis-initiating cells particularly
rely on dietary lipids to promote metastasis.
The mechanisms whereby some tumour cells detach from the primary
lesion to colonize distant sites are still largely unknown. Pro-metastatic
events common to the majority of solid tumours might include the
reversible transition of tumour cells from an epithelial to a mesenchymal state as well as their interactions with stromal components
or tumour-activated stromal cells117. Some tumours also secrete
metastasis-promoting exosomes that contain proteins, mRNAs and
microRNAs to establish a distant pro-metastatic niche9,13,18,19. However,
whether a subpopulation of metastasis-initiating cells exists among
primary tumour-initiating cells is not clear.
LRCs express lipid metabolism genes
When cell lines and patient-derived cells (PDCs) arising from human
oral carcinomas (Methods) were pulsed with a lipophilic fluorescent
dye (DiD) that non-specifically binds to membranes and is diluted
upon cell division20, and were orthotopically injected into the oral
cavity of NOD.Cg-PrkdcscidIl2rgtm1Wjl/SzJ (NSG) mice, we observed
a small percentage of slow-cycling CD44bright/dye+ long-term labelretaining cells (LRCs) within oral lesions (Fig. 1a, b and Extended Data
Fig. 1ak). Thus, the CD44bright population, which have been shown
to have the highest tumour-initiating potential in oral s quamous cell
carcinomas (OSCCs), displayed cell cycle heterogeneity in vivo2123.
Although the transcriptomes of LRCs (CD44bright dye+) and nonLRCs (dye) sorted by fluorescence-activated cell sorting (FACS)
from orthotopic tongue tumours derived from the OSCC cell line
SCC-25 were more similar to each other than to the differentiated
CD44dim population, they still displayed a number of differentially
expressed genes (Extended Data Fig. 2a and Supplementary Table 1a).
Gene ontology analysis indicated that the CD44bright dye signature
was associated with chromosomal instability, cell transformation
and neoplasm and genes involved in the cell cycle, as expected from
a proliferative tumour population (Extended Data Fig. 2b, c and
Supplementary Table 1b). On the other hand, the signature of LRCs
included an over-representation of genes associated with lymphatic

metastasis, neoplasm metastasis, response to lipids and lipid metabolic
process (Fig. 1c, d and Extended Data Fig. 2c, d); this was c onfirmed
in LRCs sorted from a second OSCC tumour line (Detroit-562;
Extended Data Fig. 2e and Supplementary Table 1c, d). We validated
the d
ifferential expression of several of these genes by quantitative
PCR with reverse transcription (RTqPCR) in five biological replicates
(Extended Data Fig. 2f). Notably, 69 genes were found in the dye+
signatures from both the SCC-25 and the Detroit-562 cell lines; the
main functions of these genes were related to neoplasm metastasis,
lymphatic metastasis, response to stimulus, lipid distribution and
translocation and response to lipid, underscoring their r elevance
in defining the LRC population (Extended Data Fig. 2g and
Supplementary Table 1e).
LRCs expressed genes involved in different aspects of fatty acid
metabolism, including lipid uptake and transport, fatty acid - and
-oxidation, lipid biosynthesis and intracellular lipid storage (Extended
Data Fig. 2d). Among the products of these genes, the receptor CD36 is
at the top of the signalling cascade that takes lipids up from the extracellular environment, allowing cells to obtain ATP energy through
lipid -oxidation2426. As CD36 is a cell surface receptor, we used
it as a surrogate marker to detect and isolate LRCs from tumours,
circumventing fluorescent dyes. Strikingly, LRCs corresponded to the
CD36bright CD44bright cells within primary oral lesions derived from all
cell lines and PDCs tested (Extended Data Figs 2h, 3a).
CD36 is essential for metastasis
Overexpression of CD36 in cell lines or PDCs with low metastatic

potential (SCC-25, VDH-00 and JHU-029) greatly increased their
potential to metastasise to lymph nodes, with penetrance increasing
from less than 20% to 7580% (Extended Data Figs 3b, eg and 4a).
Lymph node metastases generated by OSCC tumours o
verexpressing
CD36 were also more than 40 times the size of those generated by
Institute for Research in Biomedicine (IRB Barcelona), The Barcelona Institute of Science and Technology (BIST), 08028 Barcelona, Spain. 2Centre for Genomic Regulation (CRG), The Barcelona
Institute of Science and Technology (BIST), Dr. Aiguader 88, 08003 Barcelona, Spain. 3IMIM, Department of Dermatology, Hospital del Mar, 08003 Barcelona. 4Vall DHebron Hospital, Barcelona,
Department of Oral and Maxillofacial Surgery, Universitat Autnoma de Barcelona, Barcelona 08035 Spain. 5Catalan Institution for Research and Advanced Studies (ICREA), 08010 Barcelona,
Spain. 6Universitat Pompeu Fabra (UPF), Barcelona 08002, Spain.
0 0 M O N T H 2 0 1 6 | VO L 0 0 0 | NAT U R E | 1
2016 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
RESEARCH ARTICLE
104
102
35.5%
CD44bright
Dye
102
104
100
100
25
20
15
10
5
DiD+
*
CD44
*
Signal transduction pathway (dye+)

Interleukin 1
Phosphatidylinositol
Pharathyroid hormone related protein
CD36 molecule
PPAR-
Matrix metalloproteinase
TGF
Rhoa Ras homologue
101
104 P value
Figure 1 | CD44bright LRCs have lymphatic metastasis and lipid

metabolism transcriptome signatures. a, LRC populations separated
by flow cytometry from SCC-25-derived tumours. Three independent
experiments, n=6 mice per experiment. b, Percentage of dividing cells.
*P< 0.05; two-tailed t-test; six independent experiments, n= 3 mice
per experiment; data are mean s.e.m. c, Top ten disease categories
upregulated in dye+ cells. d, Over-represented processes and signal
transduction pathways in dye+ cells.
the parental cells (Extended Data Figs 3bd, 4a). By contrast, primary
tumour growth was only slightly, if at all, enhanced by CD36 overexpression (less than twofold) (Extended Data Figs 3b, 4a). Most
importantly, short hairpin RNA (shRNA)-mediated depletion of CD36
significantly reduced the penetrance of metastasis to lymph nodes, in
some cases by 80100%. Either no or, in one tumour case, only slight
effects on oral lesion growth were observed (Fig. 2a, b and Extended
Data Fig. 4b). Conversely, CD36 depletion greatly reduced the size
of lymph node metastases in all tumour lines, and inhibited lung
metastasis by FaDu cells (Extended Data Fig. 4b-f).
CD36+cells respond to dietary lipids
The histological analyses of the few lymph node metastases that

grew from CD36-depleted cells presented an intriguing pattern
of large swollen cells that were filled with lipid droplets containing
non-metabolized lipids (Fig. 2c, d and Extended Data Fig. 4g). These
structures were not present in the oral lesions generated by OSCC cells
depleted of CD36 (Extended Data Fig. 4h). We therefore hypothesized
that CD36+ cells might specifically require lipid m
etabolism to exert
their metastatic potential. In fact, CD36+ CD44bright cells isolated from
primary oral orthotopic tumours, but not their CD36+ CD44bright
counterparts, expressed numerous genes involved in lymphatic
metastasis and lipid metabolism, which overlapped with the dye+
signature (Extended Data Fig. 5ad and Supplementary Table 2ac).
Second, CD36+ CD44bright cells expressed higher levels of three key
enzymes involved in fatty acid -oxidation (ACADVL, ACADM
and HADHA; Extended Data Fig. 5e). Third, depletion of ACSL1,
which adds an acyl-coenzyme A moiety to fatty acids to activate their
oxidation27,28, significantly reduced the lymph node metastatic
penetrance of parental OSCC cells and OSCC cells overexpressing
CD36, but not primary tumour uptake (Extended Data Fig. 5fj).
Notably, NSG mice fed with a high-fat diet developed more and
larger lymph node metastases, in a CD36-dependent manner (Fig. 3a,
Extended Data Fig. 6a and Supplementary Table 3). The boost in
metastatic potential of OSCCs in high-fat diet-fed mice correlated with
an increase in the percentage of CD36+ cells in oral and m
etastatic
lesions, suggesting that the expression of CD36 might be sensitive to the
concentration of fatty acids (Extended Data Fig. 6a). Indeed, the percentage of CD36+ cells was also strongly elevated when OSCC cells were
co-cultured with OP9-derived adipocytes, but not with non-adipogenic
Tumour
Tumour-free
Metastasis
(%)
Met
Met-free
75
50
25
0
PLKO #98 #99
Cd36 shRNA
Undifferentiated
SCC
PLKO #98
Primary tumour
1.5
5 105
5 106
LDs
Subcapsular
sinus
Metastasis
**
1.0
0.5
#99
Cd36 shRNA
Cd36 shRNA
PLKO #98 #99
Cd36 shRNA
GFP
CD44
DAPI
LDs
PT
shCD36#99
BrdU+ cells (%)
DiD
30
LNMet
(dye+)
Biological process
Response to lipid
Response to bacterium
Lipid metabolic process
Response to oxygen compound
Skin development
102
105 P value
40
35
1018
Primary tumours
(%)
100
PT
CD44
108
PT
shCD36#98
8.07%
CD44dim
Psoriasis
Neoplasms, squamous cell
Skin diseases, papulosquamous
Stomach neoplasms
Carcinoma, squamous cell
Prostatic diseases
Lymphatic metastasis
Neoplasm metastasis
Skin neoplasms
1028 P value
SCC-25
Disease (dye+)
Squamous
differentiation
1,000 m
PLKO #98 #99
Cd36 shRNA
LDs
25 m
CD44bright
Dye+
2.37%
CD44bright
Dyelow
16.8%
PLKO
DiD dye
GFP+Lin
106
Relative photon flux
GFP
DAPI
Bodipy-568
100 m
Figure 2 | Modulation of CD36 expression strongly affects metastatic

penetrance and growth. a, b, Bioluminescence imaging (BLI)
quantification (a) and frequency of tumours (b) in mice inoculated with
tumour cells expressing empty vector (PLKO) or shRNA targeting Cd36.
PLKO, n=9 mice; shCD36#98, n=7 mice; shCD36#99, n= 6 mice
(*P=0.003, Fishers exact test). Relative photon flux primary tumour,
*P=0.04 and metastasis, *P= 0.03 and **P= 0.006, two-tailed t-test;
data are mean s.e.m. c, Representative haematoxylin and eosin staining
of metastatic lymph nodes (LDs, lipid droplets). d, Immunostaining of
lymph node metastases arising from cells expressing Cd36 shRNA. Inset,
lipid droplets surrounding GFP+ CD44+ cells and showing free fatty
acid content. Source data from mouse experiments are in Supplementary
Information.
OP9 parental cells or with squamous cell c arcinoma-associated

f ibroblasts (Extended Data Fig. 6b, c). In addition, exposure of
cultured OSCC cells to palmitic acid, a dietary fatty acid recognized
by CD36 (ref. 29), for 2 days also robustly increased the percentage of
CD36+ cells (Extended Data Fig. 6e). Palmitic acid increased the size
and frequency of lymph node metastases in a manner dependent on
CD36, without affecting primary tumour growth. Notably, palmitic
even promoted lung metastasis in 10% of mice, something we did not
observe in the more than 100 mice inoculated with control SCC-25
tumour cells (Fig. 3b).
Importantly, OSCC cells expressing mutant CD36 (CD36-K164A)
with an impaired ability to internalize lipids30 generated primary
lesions with the same penetrance as cells harbouring wild-type
CD36, but displayed substantially fewer and much smaller metastases
(Fig. 3c and Extended Data Fig. 5fh). Lymph node metastases
generated by CD36-K164A tumour cells contained large lipid droplets
similar to those formed after endogenous CD36 depletion (Extended
Data Fig. 7a). As CD36 can activate fatty acid -oxidation31,32, we
hypothesized that CD36 inhibition leads to accumulation of endogenously synthesized, unmetabolized lipids. This continuous lipid
accumulation would ultimately result in metastatic lipotoxicity and cell
death. Indeed, caspase-3 immunoreactivity was observed within and
surrounding the lipid droplet-rich areas, both in metastases expressing
the CD36-K164A mutant and in those depleted of endogenous CD36
(Extended Data Fig. 7b, c).
CD36+cells are metastasis-initiating
In all OSCC cell lines and PDCs tested, CD36+ CD44bright cells were
always enriched at metastatic sites with respect to primary oral lesions
(Extended Data Fig. 7d). In addition, knockdown and o
verexpression
of CD36 inhibited and upregulated, respectively, the expression of
genes associated with the metastatic process (Extended Data Fig. 7e).
We next FACS-sorted CD36+ and CD36 cells from OSCC cells co-
cultured with adipocytes derived from mouse mesenchymal OP9 cells
and then orthotopically inoculated them separately into NSG mice
(Extended Data Fig. 7f, g). Under these co-culture conditions, Cd36
2 | NAT U R E | VO L 0 0 0 | 0 0 M O N T H 2 0 1 6
ARTICLE RESEARCH
CTD
1.5
2
2
1
0
1.8 107
100
75
50
25
0
Metastasis (%)
PT
et
-M
Lys164mut
Primary tumours (%)
Tumour
Tumour-free
LN
3 105
**
**
2.0
1.5
1.0
Unt Ct shCD36
PA 0.4 mM
Met
Met-free
Unt Ct shCD36
PA 0.4 mM
Cd36 Lys164 Cd36 Lys164

wt
wt
mut
mut
Primary tumour
1.2
1.0
0.8
0.6
0.4
0.2
0
P = 5.47 1021
1/2,000
Metastasis
****
Cd36 Lys164 Cd36 Lys164

wt
wt
mut
mut
Figure 3 | CD36+ cells require fatty acid internalization to promote

metastasis. a, BLI monitoring and frequency of tumours from mice fed
either with high-fat diet (HFD) or control diet (CTD). PLKO, n=5 mice;
Cd36 shRNA, n=5 mice. Primary tumour, P=0.007 and P<0.0001;
metastasis, P=0.01 and P=0.04, two-tailed t-test. b, BLI monitoring and
frequency of tumours generated from palmitic acid (PA)-treated cells.
LN-Met, lymph node metastasis; PT, primary tumour. PLKO, n=10 mice;
Cd36 shRNA, n=10 mice (*P=0.05, **P=0.03, ***P=0.0001, Fishers
exact test). BLI quantifications: primary tumour, *P=0.01 and **P=0.007;
metastases, **P=0.005, two-tailed t-test. Unt, control vector untreated; Ct,
control vector treated with palmitic acid (PA) and Cd36 shRNA (shCD36)
treated with palmitic acid (PA). c, BLI monitoring and frequency of tumours
generated by primary OSCC cells transduced with wild-type (wt) CD36
(n=10 mice) or CD36-K164A (n=10 mice). *P=0.03, Fishers exact test,
and ****P<0.0001, two-tailed t-test. BLI data in ac are the means.e.m.
Source data from mouse experiments are in Supplementary Information.
mRNA expression was significantly lower in CD36 cells than in those

isolated from OSCC cell monocultures (Extended Data Fig. 6b).
Notably, CD36 cells did not generate even a single lymph node
metastasis, whereas CD36+ cells formed lymph node metastases with
an even higher penetrance than the parental cells (Fig. 4a and Extended
Data Fig. 7h). On the other hand, both CD36+ and CD36 cells
generated oral lesions with 100% efficiency and similar tumour size
(Fig. 4a and Extended Data Fig. 7i). Orthotopic inoculation of limiting
dilutions determined that approximately 1/3,000 CD36+ CD44bright
cells harbour metastatic potential, a tenfold increase compared to the
entire CD44bright population (Fig. 4b, c, Supplementary Table 4a-b,
and Extended Data Fig. 7j, k). CD36+ CD44bright cells were equally
tumorigenic in the oral cavity as their CD36 CD44bright counterparts
or the entire CD44bright population (Fig. 4c, Extended Data Fig. 7ik and
Supplementary Table 4b). Furthermore, the gene expression signatures
of CD36+ and CD36 cells sorted from oral lesions (generated upon
inoculation of parental Detroit-562 cells) or lymph node metastases
(generated upon inoculation of CD36+ Detroit-562 cells) were very
similar and strongly defined by lipid metabolism (Extended Data
P = 0.147 (NS)
P = 0.0747 (NS)
1/500
1/665
1/3,000
1/5,000
1/1,000
1/25,000
ht
ig
SCC-25
ht
Detroit-562
ig
SCC-25
D3
D4 6
4 br
Detroit-562
D3
6+
D3
C
D4 6
4 br
1/2,000
1/5,0000
25
25
P = 0.772 (NS)
P = 6.06 106
75
50
TIC frequency
P = 3.43 1010
1/2,500
50
MIC frequency
Tumour
Tumour-free
Metastasis (%)
**
* ***
Primary tumours (%)
D3
6+
Unt Ct shCD36
PA 0.4 mM
Met
Met-free
**
Met
Met-free
HFD
VDH-02
CD36wt
75
0.5
0
CTD
2.5
**
0.5
PL
sh KO
C
D3
6
PL
sh KO
C
D3
6
PL
sh KO
C
D3
6
PL
sh KO
C
D3
6
HFD
1.0
CTD
Primary tumours (%)
-M
et PT
LN
HFD
25
Unt Ct shCD36
PA 0.4 mM
Metastasis
Primary tumour
Metastasis
6
** ****
Primary tumour
4
6 106
*
100
100
6 105
HFD
CTD
CTD
HFD
25
Metastasis (%)
50
PLKO
Palmitic acid 0.4 mM
Cd36 shRNA
Palmitic acid 0.4 mM
CTD
50
75
et
8 105
Metastasis (%)
Met
Met-free
Cd36
PLKO shRNA
lu
ng
-
Tumour
Tumour-free
Cd36
PLKO shRNA
75
PLKO
untreated
shCD36#98#99
PLKO
100
LN
shCD36#98#99
PLKO
Tumour
Tumour-free
100
SCC-25
Pa
re
n
C tal
D3
6+
C
D3
6
Pa
re
nt
C al
D3
C 6+
D3
6
Pa
re
n
C tal
D3
C 6+
D3
6
Pa
re
n
C tal
D3
C 6+
D3
6
b
-M
et PT
8 106
HFD
Primary tumours (%)
High-fat diet Control diet
Detroit-562
PT
Figure 4 | CD36-positive cells initiate and promote OSCC metastasis.

a, Frequency of tumours generated by CD44bright CD36+ or CD44bright
CD36 cells compared to the parental cell lines (P= 0.01, P= 0.009,
two-tailed Fishers exact test). b, c, Frequency of metastasis-initiating cells
(MIC) and tumour-initiating cells (TIC) for CD44bright CD36+, CD44bright
CD36 or CD44bright populations determined by extreme limiting dilution
analysis (ELDA). Goodness-of-fit tests, estimated slope= 0.643; likelihood
ratio test of single-hit model, P=0.01; and score tests of heterogeneity,
P= 5 105. Source data from mouse experiments are in Supplementary
Information. NS, not significant.
Fig. 8ac and Supplementary Table 5ac). Thus, when transplanted,

CD36+ cells are not only uniquely capable of initiating metastasis but
can also recapitulate their molecular and cellular heterogeneity from
the primary origin, and hence represent bona fide metastasis-initiating
cells.
CD36-based anti-metastatic therapy
We next used two different anti-CD36 neutralizing antibodies:

one reported to inhibit all known functions of CD36, including its
interactions with thrombospondin, collagens and fatty acids (FA6.152),
and the other reported to block fatty acid and oxidized low-density
lipoprotein uptake (JC63.1)33,34. Treating mice that had been orally
inoculated with tumour cells with an intraperitoneal injection of
either antibody every 3 days completely inhibited metastasis initiation
without affecting the percentage of mice that developed primary
tumours (Extended Data Fig. 9a). More strikingly, treating mice that
had already developed lymph node metastasis (from OSCC cell lines
or PDCs) with daily intraperitoneal injections of JC63.1 reduced the
size of lymph node metastases by 8090%, and resulted in complete
remission in 15% of mice, in a dose-dependent manner and without
affecting oral lesion sizes by the endpoint (Fig. 5ac and Extended Data
Fig. 9bd). We observed lipotoxicity and high caspase-3 immunoreactivity around large lipid-swollen cells that accumulated in the lymph
node metastases of mice treated with CD36-neutralizing antibodies,
which were similar to those observed after CD36 knockdown or
expression of CD36-K164A (Fig. 5d and Extended Data Fig. 9e). Mice
did not lose weight when treated continuously with CD36-neutralizing
antibodies (Supplementary Table 6).
To further assess the possible therapeutic use of CD36-neutralizing
antibodies to inhibit metastasis, we tested them in immunocompetent C3H/HeJ mice inoculated with a murine Ln-7-derived OSCC
cell line. Mice were treated with either IgA or JC63.1 and examined
post-mortem. Treatment with JC63.1 strongly inhibited Ln7-induced
metastasis, with some cases of full regression of lymph node and lung
metastases (Extended Data Fig. 9f, g). No changes were observed in
body weight (Supplementary Table 7a) or liver and kidney weight or
histology (Extended Data Fig. 8h), or in the levels of liver transaminases,
haemoglobin, haematocrit, white blood cells, lymphocytes, g ranulocytes,
MID (mid-range absolute count of monocytes, eosinophils, basophils,
blasts and other precursor white cells), lymphocytes, red blood cells or
platelets, as compared to IgA-treated controls (Supplementary Table 7b).
CD36 correlates with poor prognosis
We next analysed publicly available data and found that CD36

expression, or its associated signature, correlated strongly with poor
0 0 M O N T H 2 0 1 6 | VO L 0 0 0 | NAT U R E | 3
RESEARCH ARTICLE
b
LN-Met BLI
6 Anti-CD36 JC63.1 (g per day)
[5]
5
[10]
[20]
4
3 Daily injection
start of treatment
2
1
0
12
75
50
25
0
18 Days
Primary tumour
(NS)
Online Content Methods, along with any additional Extended Data display items and
Source Data, are available in the online version of the paper; references unique to
these sections appear only in the online paper.
IgA
Anti-CD36
JC63.1
Anti-CD36
JC63.1
5 10
10
20
20 g per day
Met remission
Met
100
Met remission (%)
100
16
**
1.0
0.5
metastasis of ovarian cancer cells2. We therefore expect that targeting

CD36 and CD36+ metastasis-initiating cells could provide a breakthrough therapy to specifically target the metastatic process.
Metastasis
IgA
Tumour-free
Tumour
Primary tumours (%)
1.5
Photon flux ( 106)
IgA Anti-CD36
JC63.1
LDs
75
50
25
0
LN-Met
IgA
500 m
LN-Met
Anti-CD36 JC63 .1
500 m
IgA Anti-CD36
JC63.1
Figure 5 | CD36-blocking antibodies therapeutically inhibit metastasis

of OSCC tumours. a, b, Doseresponse in BLI signal of tumours from
mice treated daily with monoclonal JC63.1 (5, 10 or 20g; n=7, 12 or
7 mice, respectively) or the IgA isotype control (n=13 mice) for 3weeks.
**P=0.008, IgA vs JC63.1 (10g) *P=0.01 and IgA vs JC63.1 (5g)
*P= 0.05, two-tailed t-test. Data are mean s.e.m. c, Frequency of
tumour remission. d, Haematoxylin and eosin staining of metastatic
lymph nodes of animals treated daily with JC63.1. Dashed line shows areas
surrounded by lipid droplets. Source data from mouse experiments are in
Supplementary Information.
disease-free survival and overall survival in patients with lung SCC,

bladder cancer or luminal A breast cancer (Extended Data Fig. 10a and
Supplementary Table 8). In addition, CD36 amplification s pecifically
correlated with metastasis in a large number of human tumours,
including highly aggressive ones such as melanoma35. When NSG mice
were inoculated intravenously with a luminal A breast cancer (MCF-7)
or human melanoma (501mel) cell line lacking CD36, both the number
and size of lung, bone and liver metastases were strongly reduced as
compared to controls (Extended Data Fig. 10b, c).
We have identified a small population of CD36+ cells that are highly
predisposed to promote metastasis and are predominantly defined by
a lipid metabolism signature. Lipid -oxidation is the most energetically efficient type of oxidation in terms of moles of ATP per molecule
metabolized36. Hence, CD36+ metastatic cells might take advantage
of this feature to obtain the high amount of energy that is likely to
be required for them to anchor and survive at sites distant from the
primary tumour. Although blocking lipid uptake resulted in s ubstantial
metastatic lipotoxicity, we cannot rule out the possibility that inhibition
of other functions of CD36 might also contribute to this anti-metastatic
effect26,34. In fact, our preliminary results indicate that CD36 knockdown substantially reduced the initial homing of OSCC cells to the
lungs, without affecting their attachment to the oral cavity or their
survival in circulation (Extended Data Fig. 10d). Furthermore, CD36+
cells expressed overall lower levels of epithelial-to-mesenchymal (EMT)
transition-associated genes than their CD36 counterparts in primary
lesions and lymph node metastases (Extended Data Fig. 10e). These
results suggest that EMT might not be required for the metastatic
process in all types of cancer3741, or that metastasis-initiating cells
are likely to require cooperation with other tumour cells that have
undergone EMT (that is, CD36 cells) or with tumour stroma cells to
leave the primary lesion39,42.
Regarding the therapeutic significance of our findings, several lines
of evidence suggest that CD36+ cells constitute a general mechanism
of metastasis. Besides our results in OSCC, melanoma, luminal breast
cancer43, bladder cancer, and lung SCC, CD36 drives glioblastoma
progression44, is amplified in metastases in most types of human
tumours35, and promotes motility of hepatocellular carcinoma cells45.
Furthermore, lipid metabolism is part of the signature that defines
the most invasive cells in murine epidermal SCC46 and is required for
Received 16 September 2015; accepted 16 November 2016.

Published online 7 December 2016.
1. Oskarsson, T., Batlle, E. & Massagu, J. Metastatic stem cells: sources, niches,
and vital pathways. Cell Stem Cell 14, 306321 (2014).
2. Nieman, K. M. et al. Adipocytes promote ovarian cancer metastasis and
provide energy for rapid tumor growth. Nat. Med. 17, 14981503 (2011).
3. Kalluri, R. & Zeisberg, M. Fibroblasts in cancer. Nat. Rev. Cancer 6, 392401
(2006).
4. Zhang, X. H. et al. Selection of bone metastasis seeds by mesenchymal signals
in the primary tumor stroma. Cell 154, 10601073 (2013).
5. Calon, A. et al. Dependency of colorectal cancer on a TGF--driven program in
stromal cells for metastasis initiation. Cancer Cell 22, 571584 (2012).
6. Goel, H. L. & Mercurio, A. M. VEGF targets the tumour cell. Nat. Rev. Cancer 13,
871882 (2013).
7. Sevenich, L. et al. Analysis of tumour- and stroma-supplied proteolytic
networks reveals a brain-metastasis-promoting role for cathepsin S. Nat. Cell
Biol. 16, 876888 (2014).
8. Valiente, M. et al. Serpins promote cancer cell survival and vascular co-option
in brain metastasis. Cell 156, 10021016 (2014).
9. Obenauf, A. C. et al. Therapy-induced tumour secretomes promote resistance
and tumour progression. Nature 520, 368372 (2015).
10. Lu, X. et al. VCAM-1 promotes osteolytic expansion of indolent bone
micrometastasis of breast cancer by engaging 41-positive osteoclast
progenitors. Cancer Cell 20, 701714 (2011).
11. Chen, Q., Zhang, X. H. & Massagu, J. Macrophage binding to receptor VCAM-1
transmits survival signals in breast cancer cells that invade the lungs. Cancer
Cell 20, 538549 (2011).
12. Chaffer, C. L. & Weinberg, R. A. A perspective on cancer cell metastasis. Science
331, 15591564 (2011).
13. Peinado, H. et al. Melanoma exosomes educate bone marrow progenitor cells
toward a pro-metastatic phenotype through MET. Nat. Med. 18, 883891
(2012).
14. Oskarsson, T. et al. Breast cancer cells produce tenascin C as a metastatic
niche component to colonize the lungs. Nat. Med. 17, 867874 (2011).
15. McAllister, S. S. & Weinberg, R. A. The tumour-induced systemic environment
as a critical regulator of cancer progression and metastasis. Nat. Cell Biol. 16,
717727 (2014).
16. Paolino, M. et al. The E3 ligase Cbl-b and TAM receptors regulate cancer
metastasis via natural killer cells. Nature 507, 508512 (2014).
17. Wculek, S. K. & Malanchi, I. Neutrophils support lung colonization of
metastasis-initiating breast cancer cells. Nature 528, 413417 (2015).
18. Zomer, A. et al. Invivo imaging reveals extracellular vesicle-mediated
phenocopying of metastatic behavior. Cell 161, 10461057 (2015).
19. Zhou, W. et al. Cancer-secreted miR-105 destroys vascular endothelial barriers
to promote metastasis. Cancer Cell 25, 501515 (2014).
20. Yumoto, K., Berry, J. E., Taichman, R. S. & Shiozawa, Y. A novel method for
monitoring tumor proliferation in vivo using fluorescent dye DiD. Cytometry A
85, 548555 (2014).
21. Prince, M. E. et al. Identification of a subpopulation of cells with cancer stem
cell properties in head and neck squamous cell carcinoma. Proc. Natl Acad. Sci.
USA 104, 973978 (2007).
22. Harper, L. J., Piper, K., Common, J., Fortune, F. & Mackenzie, I. C. Stem cell
patterns in cell lines derived from head and neck squamous cell carcinoma.
J. Oral Pathol. Med. 36, 594603 (2007).
23. Biddle, A. et al. Cancer stem cells in squamous cell carcinoma switch between
two distinct phenotypes that are preferentially migratory or proliferative.
Cancer Res. 71, 53175326 (2011).
24. Coburn, C. T. et al. Defective uptake and utilization of long chain fatty acids in
muscle and adipose tissues of CD36 knockout mice. J. Biol. Chem. 275,
3252332529 (2000).
25. Ibrahimi, A. et al. Muscle-specific overexpression of FAT/CD36 enhances fatty
acid oxidation by contracting muscle, reduces plasma triglycerides and fatty
acids, and increases plasma glucose and insulin. J. Biol. Chem. 274,
2676126766 (1999).
26. Pepino, M. Y., Kuda, O., Samovski, D. & Abumrad, N. A. Structure-function of
CD36 and importance of fatty acid signal transduction in fat metabolism.
Annu. Rev. Nutr. 34, 281303 (2014).
27. Ellis, J. M. et al. Adipose acyl-CoA synthetase-1 directs fatty acids toward
beta-oxidation and is required for cold thermogenesis. Cell Metab. 12, 5364
(2010).
28. Kermorvant-Duchemin, E. et al. Trans-arachidonic acids generated during
nitrative stress induce a thrombospondin-1-dependent microvascular
degeneration. Nat. Med. 11, 13391345 (2005).
29. Glatz, J. F., Luiken, J. J. & Bonen, A. Membrane fatty acid transporters as
regulators of lipid metabolism: implications for metabolic disease. Physiol. Rev.
90, 367417 (2010).
4 | NAT U R E | VO L 0 0 0 | 0 0 M O N T H 2 0 1 6
ARTICLE RESEARCH
30. Holmes, R. S. Comparative studies of vertebrate platelet glycoprotein 4 (CD36).
Biomolecules 2, 389414 (2012).
31. Kennedy, D. J. & Kashyap, S. R. Pathogenic role of scavenger receptor CD36 in
the metabolic syndrome and diabetes. Metab. Syndr. Relat. Disord. 9, 239245
(2011).
32. Shi, Y. & Burn, P. Lipid metabolic enzymes: emerging drug targets for the
treatment of obesity. Nat. Rev. Drug Discov. 3, 695710 (2004).
33. Mwaikambo, B. R., Sennlaub, F., Ong, H., Chemtob, S. & Hardy, P. Activation of
CD36 inhibits and induces regression of inflammatory corneal
neovascularization. Invest. Ophthalmol. Vis. Sci. 47, 43564364 (2006).
34. Klenotic, P. A. et al. Molecular basis of antiangiogenic thrombospondin-1 type 1
repeat domain interactions with CD36. Arterioscler. Thromb. Vasc. Biol. 33,
16551662 (2013).
35. Nath, A. & Chan, C. Genetic alterations in fatty acid transport and metabolism
genes are associated with metastatic progression and poor prognosis of
human cancers. Sci. Rep. 6, 18669 (2016).
36. Weber, J. M. Metabolic fuels: regulating fluxes to select mix. J. Exp. Biol. 214,
286294 (2011).
37. Ye, X. et al. Distinct EMT programs control normal mammary stem cells and
tumour-initiating cells. Nature 525, 256260 (2015).
38. Del Pozo Martin, Y. et al. Mesenchymal cancer cell-stroma crosstalk promotes
niche activation, epithelial reversion, and metastatic colonization. Cell Reports
13, 24562469 (2015).
39. Pein, M. & Oskarsson, T. Microenvironment in metastasis: roadblocks
and supportive niches. Am. J. Physiol. Cell Physiol. 309, C627C638
(2015).
40. Fischer, K. R. et al. Epithelial-to-mesenchymal transition is not required for
lung metastasis but contributes to chemoresistance. Nature 527, 472476
(2015).
41. Zheng, X. et al. Epithelial-to-mesenchymal transition is dispensable for
metastasis but induces chemoresistance in pancreatic cancer. Nature 527,
525530 (2015).
42. Celi-Terrassa, T. et al. Epithelial-mesenchymal transition can suppress major
attributes of human epithelial tumor-initiating cells. J. Clin. Invest. 122,
18491868 (2012).
43. Pavlovic, M. et al. Enhanced MAF oncogene expression and breast cancer bone
metastasis. J. Natl Cancer Inst. 107, djv256 (2015).
44. Hale, J. S. et al. Cancer stem cell-specific scavenger receptor CD36 drives
glioblastoma progression. Stem Cells 32, 17461758 (2014).
45. Nath, A., Li, I., Roberts, L. R. & Chan, C. Elevated free fatty acid uptake via CD36
promotes epithelial-mesenchymal transition in hepatocellular carcinoma.
Sci. Rep. 5, 14752 (2015).
46. Oshimori, N., Oristian, D. & Fuchs, E. TGF-promotes heterogeneity and drug
resistance in squamous cell carcinoma. Cell 160, 963976 (2015).
Supplementary Information is available in the online version of the paper.
Acknowledgements Research in the laboratory of S.A.B. for this project is
supported by the European Research Council (ERC), the Government of
Catalua (SGR grant), and the Fundacin Botn and Banco Santander, through
Santander Universities. We would like to thank the Beug Stiftung Foundation for
their support. S.M. was supported by a La Caixa International PhD fellowship.
A.A. was supported by an EU Cofound postdoctoral fellowship. L.D.C. was
supported by the Spanish Ministerio de Educacin y Ciencia (SAF201348926-P) and the European Commissions 7th Framework Program 4DCellFate
grant number 277899. We thank the Vall DHebron Research Institute Tumor
Biobank for their assistance with the human samples. We also thank R. Wong
for the Ln-7 cell line and J. Zuber for the PMSCV-Luc2-PGKneo-Ires GFP vector.
IRB Barcelona is the recipient of a Severo Ochoa Award of Excellence from
MINECO (Government of Spain). We thank V. Raker for manuscript editing.
Author Contributions G.P. and S.A.B. designed all experiments. G.P. performed
all experiments with the help of M.M. for the histological characterization of the
lipotoxicity and A.C. for the analysis of the gene expression data. A.A. established
the patient-derived cells and the oral cancer orthotopic method. C.S.-O.A. and
A.B. performed statistical analyses. J.A.H., C.B. and A.T. provided the tumours
from patients. S.M. established the dye protocol to detect LRCs. N.P. performed
the histopathology analysis of the mice. L.D.C. analysed expression data.
G.P. and S.A.B. wrote the manuscript.
Author Information Reprints and permissions information is available at
www.nature.com/reprints. The authors declare no competing financial
interests. Readers are welcome to comment on the online version of the paper.
Correspondence and requests for materials should be addressed to
S.A.B. (salvador.aznar-benitah@irbbarcelona.org).
Reviewer Information Nature thanks A. Harris and the other anonymous
reviewer(s) for their contribution to the peer review of this work.
0 0 M O N T H 2 0 1 6 | VO L 0 0 0 | NAT U R E | 5
RESEARCH ARTICLE
METHODS
Animal studies. The institutional Animal Care and Use Committee of IRB
Barcelona approved all animal procedures. NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ
(NSG) mice were purchased from Charles River and crossed in-house. C3H/
HeJ (000659) mice were purchased from the Jackson Laboratory. All mice were
housed under a regimen of 12h light/12h dark cycles and specific pathogen-free
conditions, and all procedures were evaluated and approved by the CEEA (Ethical
Committee for Animal Experimentation) of the Government of Catalonia. To
establish patient-derived tumour cells, tumours were implanted subcutaneously
into NSG mice immediately after tumour resection from the patient. Once the
tumours had grown, they were collected and disaggregated, and tumour cells
were FACS-sorted and cultured as described below. For 5-bromo-2-deoxyuridine
(BrdU) labelling experiments, mice were injected intraperitoneally with a single
dose of 100g g1 BrdU (Invitrogen) and then analysed 24h later. Intra-tongue
injections of OSCC cells were performed as previously described47,48. Briefly,
mice were anaesthetized by intraperitoneal injection with a mixture of 50mg kg1
ketamine and 0.5mg kg1 medetomidin, and OSCC cells resuspended in 30l
PBS were injected into each mouse tongue with a BD ultra-fine 6-mm needle.
For intravenous tumour cell injection, mice were anaesthetized with continuous
administration of isofluorane gas and were injected by retro-orbital injection with
100,000 cells in 100l PBS. Mice were monitored for the luciferase bioluminescent
signal immediately after injection (T0) and once weekly thereafter with a Xenogen
IVIS Imaging System-100 (Caliper Life Sciences). For this, mice were injected
by retro-orbital injection with 50l d-luciferin (Promega) diluted in 1 PBS at
5mg ml1. Continuous administration of isofluorane gas was provided to maintain
anaesthesia of animals during imaging. Data were quantified with Living Image
software version 4.4 (Caliper Life Sciences). Quantifications were calculated with
unsaturated pixels. Colour scale minimum and maximum values are shown in
images.
For intravenous injections of MCF-7 breast cancer cells, a pellet of 17--
oestradiol (0.18mg per pellet, 60-day release; Innovative Research of America)
was implanted subcutaneously into 8-week-old NSG female mice 3days before
intravenous injection. Mice injected with MCF-7 or 501mel cells were analysed
for BLI 6 or 4 weeks post-inoculation, respectively.
To treat mice in vivo with neutralizing anti-CD36 antibodies, mice were injected
intraperitoneally with 100l of physiological serum containing either 2.5g of
neutralizing monoclonal anti-CD36 FA6.152 (Abcam, ab17044) or 2.5g of the
corresponding IgG1 (mouse IgG1 monoclonal NCG01, Abcam, ab81032), and
with 100l of physiological serum containing 5g, 10 g or 20g of neutralizing
monoclonal anti-CD36 JC63.1 (CAYMAN, CAY-10009893-500) or 5g, 10 g or
20g of the corresponding IgA (mouse IgA, kappa [S107], Abcam, ab37322). All
antibodies were azide-free with no added preservatives.
For the toxicological analysis regarding anti-CD36 JC63.1 treatment in
immunocompetent mice, mouse SCC Ln-7 cells were injected into the tongues
of 8-week-old syngeneic C3H/HeJ mice. When metastatic signals appeared, mice
were distributed homogeneously into groups and daily injected i ntraperitoneally
with anti-CD36 JC63.1 (10g) or the corresponding IgA control isotype as
described above.
High-fat diet experiments were performed by feeding the mice with a 60/Fat
Research diet (TD.06414, Harlan) for 7 days before inoculating the mice with
the tumour cells and thereafter. Normal diet was used for control groups. For
doxycycline treatment, mice were treated continuously with 50g ml1 of fresh
doxycycline in light-protected drinking water that contained 5% sucrose. Glucose
levels were measured once a week at a controlled time by using a glucometer
(BAYER, ContourRnext).
For each experiment, mice were killed at the same time, once an experimental
group reached the humane endpoint (46 weeks after the orthotopic injection
as soon as mice started to lose weight owing to the growth of the oral lesion).
The endpoint procedure was approved by the institutional Animal Care and Use
Committee of IRB Barcelona.
Animal tissue was collected and fixed with 4% paraformaldehyde (PFA) for 2h
at room temperature and then either embedded in OCT and frozen at 80 C or
dehydrated and embedded in paraffin. Toxicological analysis was performed at the
Histopathology Facility according to standard procedures.
Total blood samples from mice were collected from the inferior vena cava and
then processed in the Experimental Toxicology and Ecotoxicology Unit (PCB)
following standard procedures.
The investigators were not blinded to allocation during experiments and
outcome assessment.
Characterization of oral carcinoma orthotopic transplants. We used an
orthotopic model of human oral squamous cell carcinoma (OSCC) to study the
cell cycle heterogeneity of tumour cells in vivo49, first examining for slow-cycling
cancer cells, which have been shown to have high primary tumour-initiating
otential in several tumour types, and have been previously identified in OSCC cell
p
lines5053. We implanted human OSCC cells from established cell lines (SCC-25,
FaDu, Detroit-562 and JHU-029) or from three primary PDCs, transduced with a
retroviral vector expressing luciferase-GFP (Luc-GFP), into the tongue of immunosuppressed NSG mice. All cell lines and PDCs generated primary tumours with
100% penetrance, albeit with differing growth kinetics (Extended Data Fig. 1a, b).
This orthotopic model of OSCC recapitulated the metastatic spread to lymph
nodes that occurs in patients. However, the penetrance of lymph node c olonization
between different tumour cells was much more variable than primary tumour
growth, varying from very high to low, and lung metastasis was observed only in
mice injected with FaDu cells (Extended Data Fig. 1ac).
Clinical material. Biological samples were obtained from patients from the
Hospital Vall dHebron (Barcelona, Spain) under informed consent and approval
of the Bank of Tumour Committees of the hospital according to Spanish ethical
regulations. The study followed the guidelines of the Declaration of Helsinki, and
patient identity and pathological specimens remained anonymous in the context
of the study.
Plasmids. The MSCV-IRES-Luciferase-GFP retrovirus was kindly provided by
J. Zuber (Research Institute of Molecular Pathology (IMP), Vienna Biocentre,
Austria)54. Cd36 and Acsl1 knockdown experiments were conducted using
lentiviral shRNAs targeting the selected gene (Sigma Aldrich and Dharmacon,
respectively). A non-targeting shRNA sequence was used as a control (pLKO.1TRC control; Addgene, plasmid #10879) (Supplementary Table 9). CD36
overexpression experiments were conducted by cloning CD36 cDNA into the
lentiviral PMSCV vector. Empty vector was used as control. The CD36-K164A
mutant construct was made using QuikChange II XL Site-Directed Mutagenesis
Kit (Catalog #200521, Agilent Technologies), following the manufacturers
instructions, with the primer pair K164A Fwd and K164A Rvs and the plasmid
pMSVCD36OE as template. The sequence of the CD36Lys164mut plasmid was
checked by Sanger sequencing (Supplementary Table 10).
Cell culture. All cells were cultured in a humidified incubator at 37C with 5%
CO2. No method of cell line authentication was performed for the different cell
lines used. SCC-25 (ATCCR CRL-168TM) and patient-derived cells (VDH-00,
VDH-01 and VDH-02) were grown in keratinocyte serum-free medium (KSFM,
GIBCO) supplemented with 5g ml1 penicillin/streptomycin, 0.025mg ml1
bovine pituitary extract and 0.2g ml1 hEGF. JHU-029 cells (Johns Hopkins
University) were grown in RPMI (GIBCO) supplemented with 5g ml1 penicillin/
streptomycin and 10% fetal bovine serum (FBS; GIBCO). FaDu (ATCCR HTB43TM) and Detroit-562 (ATCCR CCL-138TM) cells were grown in EMEM (LONZA)
supplemented with 5g ml1 penicillin/streptomycin and 10% FBS (GIBCO).
The mouse SCC Ln-7 cell line, kindly provided by R. Wong (Department of
Surgery, Memorial Sloan Kettering Cancer Center, New York, USA)55, was grown
in EMEM (LONZA) s upplemented with 5g ml1 penicillin/streptomycin and
10% FBS (GIBCO). MCF-7 cells (ATCCR HTB-22TM) were grown in EMEM
(LONZA) supplemented with 5g ml1 penicillin/streptomycin, 0.01mg ml1
human recombinant insulin and 10% FBS (GIBCO). The 501mel human cell
line (kindly provided by C. Wellbrock, Manchester Cancer Research Centre,
The University of Manchester, UK) was grown in DMEM supplemented with
5g ml1 penicillin/streptomycin and 10% FBS (GIBCO). HNACFS (head and
neck tumourassociated fibroblasts) were grown in DMEM supplemented with
5g ml1 penicillin/streptomycin, 10% FBS (GIBCO) and 1 insulin transferrin
selenium G supplement (Invitrogen). OP9 cells were cultured and differentiated
into adipocytes as previously described56. Briefly, cells were cultured and amplified
in MEM alpha medium +glutaMAXTM (MEM-, GIBCO) supplemented with
20% FBS (GIBCO) and 5g ml1 penicillin/streptomycin. To differentiate them
into adipocytes, OP-9 cells were grown to confluency and then cultured for
two additional days in MEM-supplemented with 15% KnockOutTM Serum
Replacement (GIBCO) and 5g ml1 penicillin/streptomycin. For co-culture
experiments, OSCC cells were seeded over confluent adipogenic OP9 cells for 12to 48h at 37C with 5% CO2. PhoenixA and 293T cells grown in DMEM/10% FBS
were used for retrovirus and lentivirus production, respectively, after transfection
with the CaCl2 method. For selection, 2.5g ml1 puromycin or 0.3mg ml1
G418 was added to the medium. All cell lines tested negative for mycoplasma
contamination. For label-retaining experiments, cells were trypsinised with 0.25%
trypsin-EDTA (GIBCO), washed twice in 1PBS and incubated with VybrantR
DiD (Molecular Probes, V-22887) at a 1:200 dilution in 1PBS for 20min at
37C. After incubation, cells were washed twice with 1PBS to remove excessive
dye. Note that the dye homogeneously coated all cells in culture and was undetectable by FACS after an average of eight cell divisions. For treatment of OSCC
cells with palmitic acid, sodium palmitate (P9767, SIGMA) was prepared as a
2.5mM stock solution by dissolving it in 1ml of 0.1M NaOH and warming at
80C until clear. The fatty acid solution was complexed with fatty acid-free BSA
(A7030, SIGMA) in a molar ratio fatty acid:BSA of 5:1; briefly, 0.325g of BSA was
ARTICLE RESEARCH
dissolved in 8ml 0.9% NaCl, and the mixture was warmed to 45C. The clear
solution of palmitate was added drop-by-drop by pipette with agitation. The final
stock solution was filtered at 0.45m, aliquoted and stored at 20C. OSCC cells
growing in serum-free medium were treated in vitro with 0.4mM palmitate for
48h. For intra-tongue injections of OSCC cells, cultured cells were trypsinized
with trypsin-0.25% EDTA (GIBCO) diluted in PBS/trypan blue and resuspended
in 1PBS. For SCC-25 and JHU-029 cells, 100,000 cells per mouse were injected;
for FaDu, Detroit-562, Ln-7 and patient-derived cells, 50,000 cells per mouse were
injected. For l imiting dilution assays, serial dilutions (50,000, 20,000, 10,000, 1,000
or 100 cells) of sorted CD44bright, CD36+ CD44bright or CD36 CD44bright OSCC
cells were injected immediately after sorting into the tongues of NSG mice. For
intravenous i njection, 100,000 cells in 1PBS were injected by retro-orbital
injection. Mice were monitored by bioluminescence determination for positive
response of primary tumour and/or metastasis growth. The active cell frequency
of tumour-initiating cells or metastasis-initiating cells was estimated by s tatistical
analysis 60days after injection (humane endpoint) using the ELDA software
(http://bioinf.wehi.edu.au/software/elda/)57.
The fatty acid uptake assay was performed using the QBT Fatty Acid Uptake
Assay Explorer Kit (Molecular Devices), according to the manufacturers
instructions. Briefly, OSCC cells were grown in adipogenic-conditioned medium
for 72h and collected, and seeded in a 96-well-plate at a density of 50,000 cells per
well, with 100l adipogenic-conditioned medium. After 24h, cells were deprived of
serum and incubated for 1h at 37C with 5% CO2. Plates were read using a Biotek
FL600 Fluroescence instrument, using bottom reading, medium sensitivity (150),
excitation 488/emission 515, and a filter cutoff of 495nm.
Tumour disaggregation from xenografted mice. Tumours were isolated from
mice, and connective tissue was removed to the largest extent possible. Tissue was
chopped in 0.5% trypsin 1-300 (MP Biomedical) in KSFM medium (GIBCO)
in a McIlwain Tissue Chopper. After complete homogenization, samples were
incubated at 37C for 90min with shaking. Homogenates were filtered sequentially
in 100m, 70 m and 40m BD strainers and centrifuged at 1,000 rpm for 10min
at 4C. Supernatant was discarded, and each pellet was resuspended in 1 PBS/4%
calcium-chelated FBS7 for antibody staining and subsequent FACS analysis.
FACS analysis. For orthotopic transplant analysis, samples were incubated for
45min at room temperature with anti-CD36-PercP-EFluor 710 (ref. 46-036941, E Bioscience) and anti-CD44-PeCy7 (CD44 Clone GG-26, ref. 560533,
BD Pharmigen) at 1:100 dilution, and, to exclude contaminant mouse cells, a
lineage-negative cocktail conjugated to biotin composed of anti-CD31 clone
MEC13.3 (ref. 553371, BD Pharmigen, 1:200 dilution), anti-CD45 clone 30-F11
(ref. 13-045-81, eBioscience, 1:200 dilution) and anti-H2Kd (ref. 553564,
eBioscience, 1:200 dilution). After the first incubation, samples were washed in
1 PBS/calcium-chelated FBS58, spun for the second incubation with Brilliant
Violet (BV) 605 streptavidin (ref. 405229, Biolegend, 1:50 dilution) for 30min at
room temperature and then resuspended in 1PBS/4% calcium-chelated FBS with
DAPI at a 1:1,000 dilution, for FACS analysis. To analyse the co-cultures of OSCC
cells with adipocytes or tumour-associated fibroblasts, samples were incubated
for 45min at room temperature with anti-CD36-PercP-EFluor 710 (ref. 46-036941, E Bioscience) as well as with a mouse/rat anti-CD29-APC, clone HM11
(ref. 1002216) to exclude adipocytes. Unstained and single-colour controls were
used in each case. Samples were analysed in a BD FACS ARIA FUSION instrument.
For FACS sorting, cells were selected on the basis of their forward and side scatter
excluding cellular debris. Doublets and dead cells were eliminated by DAPI or
propidium iodide. GFP-positive cells were gated excluding the lineage-BV positive
cells. This population was selected for further analysis or were directly injected
into the tongues of the mice.
Fatty acid oxidation enzymes were measured by flow cytometry with the Fatty
Acid Oxidation Human Flow Cytometry Kit (Abcam, ab118183), according to the
manufacturers specifications.
Immunofluorescence and histological analysis. Cryo- or de-paraffinized
antigen-retrieved 8-m sections (treated for 10min in boiling 0.01 M citric acid,
pH 6.0) were permeabilized for 25min in 0.25% Triton X-100/PBS and blocked
for 90min in 0.25% gelatin/PBS. Primary antibodies were incubated overnight
at 4C, and secondary antibodies were incubated for 2h at room temperature in
0.25% gelatin/PBS. Nuclei were stained with DAPI (1:5,000, Roche), and slides
were mounted in Mowiol. The primary antibodies used were rat anti-CD44 at
1:100 (eBioscience), rabbit anti-caspase3 (abcam, ab13847) and rabbit anti-GFP at
1:1,000 (Life Technologies). DiD (D-7757, Molecular Probes) dye was analysed by
direct immunofluorescence. The secondary antibodies used were conjugated with
Alexa Fluor (R37118, A-11006, A-10042, A-11077, Molecular Probes). For neutral
lipid droplet visualization, frozen cryosections were stained with BODIPY 558/568
C12 (Molecular Probes, D-3835) as previously described59. Haematoxylin and eosin
staining was done according to the standard protocol. Images were acquired using
a Nikon E600+Olympus DP72, Leica SPE or a Leica TCS SP5 confocal microscope.
Representative pictures were selected in each case.
Gene expression analysis. RNA isolation and cDNA amplification were p
erformed
as described60. In brief, cells were sorted into lysis buffer, and RNA was p
urified
using magnetic beads (RNAClean XP beads, Agencourt). RNA was reverse
transcribed, and cDNA was amplified using whole transcriptome a mplification
chemistry (WTA2, Sigma Aldrich). Amplification was monitored by SYBR
Green addition to the reaction and was stopped when the SYBR Green signal
reached a plateau. cDNA was purified using a Purelink Quick PCR purification
kit (Invitrogen). cDNA was labelled using GeneChip Mapping 250K Nsp Assay Kit
(Affymetrix; catalogue # 900766), according to the manufacturers instructions.
Affymetrix PrimeView arrays were hybridized with 8mg labelled cDNA, washed,
stained and scanned according to the protocol described in the GeneChip 3 IVT
Express Kit User Manual. Arrays were scanned with GeneChip scanner 3000
(Affymetrix). Normalized expression signals were calculated from Affymetrix CEL
files using the RMA algorithm61. Log2 RMA expression was estimated.
For Agilent microarrays, total RNA was extracted from FACS-sorted cells
with the RNAClean XP kit Agentcourt A63987 following the manufacturers
instructions and immediately amplified using the TransPlex Complete Whole
Transcriptome Amplification Kit (Sigma WTA2) to generate cDNA. Cyanine-3
(Cy3)-labelled cDNA was prepared from 500ng double-stranded cDNA using the
DNA Enzymatic Labelling Kit (Agilent 5190-0449) according to the manufacturers
instructions, followed by column purification (Amicon 30 kDa). Dye incorporation
and cDNA yield were checked with a NanoDrop ND-1000 Spectophotometer.
Arrays were scanned with an Agilent G2539A scanner at 3-m resolution and
100% PMT. Intensity data of the individual hybridizations were extracted, and
data quality was assessed using Feature Extraction software 10.7 (Agilent). Raw
data were corrected for background noise using the normexp method. Quantile
normalization was applied to assure comparability across samples. Normalized log2
intensity values were estimated. The relative log2 ratios of DiD/DiD+, DiD/
CD44dim and DiD+/CD44dim were calculated, respectively, and genes were filtered
based on a log2 ratio DiD/DiD+ equal to or greater than 1.5; P 0.005.
Gene expression analysis for comparing the transcriptomes of CD36+ and
CD36+ cells was performed using R62 and Bioconductor63. Affymetrix microarrays
were processed using RMA normalization as implemented in the Bioconductor
R package affy63. Data were annotated using probeset information provided by
Affymetrix in its product support website (http://www.affymetrix.com/support).
Differential expression analysis between CD36+ and CD36+ samples was carried
out at the probeset level using moderated t-statistics by empirical Bayes shrinkage
as implemented in the limma R package64. For this, a linear model was fitted
using biological replicate as a random effect (functions duplicateCorrelation and
lmFit)19. The BenjaminiHochberg correction65 was applied for multiple contrast
adjustment. Geneset enrichment analysis (GSEA)66 was used to assess the enrichment of a fatty acid metabolism geneset from the Hallmark geneset collection
provided by the Broad Institute Molecular Signature Database (MsigDB-H; KEGG
FATTY ACID METABOLISM; downloaded 15/07/2015)66. In these analyses,
probesets were collapsed to gene level by selecting the probeset showing the h
ighest
standard deviation within each gene. Genes were ranked according to their fold
changes between conditions.
Genomatix software suite v3.4 (http://www.genomatix.de/)67 was also used
for gene ontology and pathway analysis. Using the Genomatix Software, pathway
analysis is based on data mining of PubMed, referring to Biocarta, STKE or KEGG.
The complete data set was deposited in the National Center for Biotechnology
Information Gene Expression Omnibus Database (GEO)68 with the accession
number GSE72939.
Real-time PCR. Real-time PCR using TaqMan gene expression probes (Applied
Biosystems) (Supplementary Table 11) was performed and analysed using a 7900HT Fast Real-time PCR Instrument (Applied Biosystems). Relative expression
levels were determined by normalization to human -2-microglobulin (b2m) using
the Ct method.
Bioinformatics analysis. Data from the TCGA project were downloaded from
cbioportal69,70. Expression data computed as mRNA z-scores (RNA Seq V2 RSEM,
Agilent) were compared to the expression distribution of each gene tumour
diploid for the queried gene. For luminal A breast cancer, staging information
mapped to the seventh edition of the Cancer Genome Atlas Network was used, as
previously described71. Stage subclassifications were collapsed to stages I, II, III
and IV. Expression of the gene signature was computed by averaging all genes in
signature for each patient. Both the CD36 and signature expressions were scaled
before fitting the model. A Cox model was fitted to overall survival and disease-free
survival data, using stage, gender, age and histological subtype as covariates. The
significance of the association between expression and survival was assessed with
a Likelihood Ratio Test as implemented in the R function drop1.
RESEARCH ARTICLE
Statistical analysis. For all the experiments, adequate sample size was determined
using the results of pilot studies. No statistical method was used to determine
sample size. All animals that fulfilled proper experimental conditions during
the experimental procedures were included in the analysis. Based on the results
of pilot studies, homogeneous groups of 8- to 10-week-old male and female
experimental mice and their control littermates were used for the s tudies. No
statistically differences were found regarding the gender of mice, and no previously
described randomization method was used. For FACS and RTpPCR analysis,
animals belonging to each experimental group were sub-divided generally
into n=34groups (containing at least two animals per group), pulled within
group and n was considered as a biological replicate. Data are generally shown
as the means.e.m. Statistical significance was analysed using Prism 6 software
(GraphPad) by using a two-tailed t-test, MannWhitney U test, Fishers exact test
or hypergeometric test. Significance was considered at P 0.05.
Data availability. The data sets generated during and/or analysed during the
current study are available in the GEO repository (accession number GSE72939).
47. Myers, J. N., Holsinger, F. C., Jasser, S. A., Bekele, B. N. & Fidler, I. J. An
orthotopic nude mouse model of oral tongue squamous cell carcinoma.
Clin. Cancer Res. 8, 293298 (2002).
48. Benaich, N. et al. Rewiring of an epithelial differentiation factor, miR-203, to
inhibit human squamous cell carcinoma metastasis. Cell Reports 9, 104117
(2014).
49. Myers, J. N., Holsinger, F. C., Jasser, S. A., Bekele, B. N. & Fidler, I. J. An
orthotopic nude mouse model of oral tongue squamous cell carcinoma.
Clin. Cancer Res. 8, 293298 (2002).
50. Pece, S. et al. Biological and molecular heterogeneity of breast cancers
correlates with their cancer stem cell content. Cell 140, 6273 (2010).
51. Roesch, A. et al. A temporarily distinct subpopulation of slow-cycling
melanoma cells is required for continuous tumor growth. Cell 141, 583594
(2010).
52. Bragado, P. et al. Analysis of marker-defined HNSCC subpopulations reveals a
dynamic regulation of tumor initiating properties. PLoS One 7, e29974 (2012).
53. Qin, J. et al. The PSA(-/lo) prostate cancer cell population harbors selfrenewing long-term tumor-propagating cells that resist castration. Cell Stem
Cell 10, 556569 (2012).
54. Zuber, J. et al. Mouse models of human AML accurately predict chemotherapy
response. Genes Dev. 23, 877889 (2009).
55. Yu, Z. et al. Sensitivity of squamous cell carcinoma lymph node metastases to
herpes oncolytic therapy. Clin. Cancer Res. 14, 18971904 (2008).
56. Wolins, N. E. et al. OP9 mouse stromal cells rapidly differentiate into
adipocytes: characterization of a useful new model of adipogenesis. J. Lipid
Res. 47, 450460 (2006).
57. Hu, Y. & Smyth, G. K. ELDA: extreme limiting dilution analysis for comparing
depleted and enriched populations in stem cell and other assays. J. Immunol.
Methods 347, 7078 (2009).
58. Nowak, J. A. & Fuchs, E. Isolation and culture of epithelial stem cells. Methods
Mol. Biol. 482, 215232 (2009).
59. Spangenburg, E. E., Pratt, S. J., Wohlers, L. M. & Lovering, R. M. Use of BODIPY
(493/503) to visualize intramuscular lipid droplets in skeletal muscle.
J. Biomed. Biotechnol. 2011, 598358 (2011).
60. Gonzalez-Roca, E. et al. Accurate expression profiling of very small cell
populations. PLoS One 5, e14418 (2010).
61. Irizarry, R. A. et al. Exploration, normalization, and summaries of high
density oligonucleotide array probe level data. Biostatistics 4, 249264
(2003).
62. R: A Language and Environment for Statistical Computing (R Foundation for
Statistical Computing, 2008).
63. Gentleman, R. C. et al. Bioconductor: open software development for
computational biology and bioinformatics. Genome Biol. 5, R80 (2004).
64. Smyth, G. in Bioinformatics and Computational Biology Solutions using R and
Bioconductor 397420 (Springer, 2005).
65. Benjamini, Y. H. Y. Controlling the False Discovery Rate: A Practical and
Powerful Approach to Multiple Testing. J. R. Stat. Soc. B 57, 289300
(1995).
66. Subramanian, A. et al. Gene set enrichment analysis: a knowledge-based
approach for interpreting genome-wide expression profiles. Proc. Natl Acad.
Sci. USA 102, 1554515550 (2005).
67. Berriz, G. F., King, O. D., Bryant, B., Sander, C. & Roth, F. P. Characterizing gene
sets with FuncAssociate. Bioinformatics 19, 25022504 (2003).
68. Barrett, T. & Edgar, R. Gene expression omnibus: microarray data storage,
submission, retrieval, and analysis. Methods Enzymol. 411, 352369
(2006).
69. Gao, J. et al. Integrative analysis of complex cancer genomics and clinical
profiles using the cBioPortal. Sci. Signal. 6, pl1 (2013).
70. Cerami, E. et al. The cBio cancer genomics portal: an open platform for
exploring multidimensional cancer genomics data. Cancer Discov. 2, 401404
(2012).
71. Cancer Genome Atlas Network. Comprehensive molecular portraits of human
breast tumours. Nature 490, 6170 (2012).
ARTICLE RESEARCH
Extended Data Figure 1 | See next page for caption.
RESEARCH ARTICLE
Extended Data Figure 1 | Orthotopically inoculated human oral
squamous cell carcinomas contain a slow-cycling sub-population of
CD44bright cells. a, Overview of the tumorigenic and metastatic activities
of the different OSCC cell lines injected into the tongues of NSG mice.
b, Tumour development from mice injected with OSCC-pLuc-GFP
cells (using the cell lines indicated). Tumour growth was monitored by
bioluminescence imaging (BLI) over a four-week period. Data are given
as the mean s.e.m. c, Frequency of metastases in the lymph nodes.
ac, Detroit-562 cells, two independent experiments: exp. 1 n= 10 mice;
exp. 2 n=11 mice; VDH-02, n=20 mice; VDH-01, n=20 mice; VDH-00,
n=8 mice; SCC-25, three independent experiments: exp. 1 n= 13 mice,
exp. 2 n=17 mice, exp. 3 n=7 mice; JHU-029, three independent
experiments, n=12 mice per experiment; FaDu, two independent
experiments, exp. 1 n=14 mice, exp. 2 n= 5 mice. d, Immunofluorescence
analysis of in vitro cultured OSCC-RFP cells pulsed with DID and grown
in 2D culture for 16days. e, f, Flow cytometry analysis of dye-pulsed OSCC
cells in vitro showing the kinetics of dye dilution. Data are given as mean
fluorescence intensity. g, FACS strategy to FACS-sort CD44bright dye+,

CD44bright dye and CD44dim cells from OSCC-pLucGFP oral tumours.
Viable single cells were selected if GFP+ but negative for a lineage (Lin)
cocktail of antibodies (H2KD, CD31 and CD45), to select human cells.
GFP+ Lin cells were gated for CD44 and dye. Percentages from the total
GFP+ Lin SCC-25 parental tumour are shown. h, Representative flow
cytometry analyses to detect quiescent slow-cycling CSCs from OSCC cell
lines. g, h, n=8 animals per OSCC cell line. i, Global quantification of
CD44bright dye+, CD44bright dye and CD44dim cells from OSCC-pLucGFP
tumours reported in g and h. j, Immunofluorescence analysis of SCC-25pLucGFP and JHU-029-pLucGFP primary tumours, collected five weeks
after OSCC inoculation, to detect dye+ quiescent slow-cycling cancer stem
cells (CSCs). Insets show a magnification of dye+ cells that co-localized
with the CD44 marker. SCC-25, n=5 tumours; JHU-029, n= 5 tumours.
k, Percentage of dividing cells by flow cytometry analysis in the dye+,
dye and CD44dim populations. Source data from mouse experiments are
in Supplementary Information.
ARTICLE RESEARCH
RESEARCH ARTICLE
Extended Data Figure 2 | Oral SCC label-retaining cells are defined
by a lipid metabolism and metastasis transcriptome signature.
a, Microarray analysis and heatmap of mRNA expression showing
differentially expressed genes in dye+, dye and CD44dim cells. n= 4
biological replicates and 8 mice per replicate. b, Gene ontology (GO)
analysis showing the top categories for diseases, biological processes and
signal transduction pathways that were upregulated in the proliferative
active (DID) as compared to LR-CSCs (DID+) populations. The resulting
GO terms highlighted cell cyclerelated categories. c, Over-represented
genes in Dye+ and Dye populations. d, Lipid metabolism genes
over-represented in dye+ cells. e, Gene ontology (GO) analysis showing
top diseases and biological processes categories upregulated in the DID+
(LR-CSCs) and DID (proliferative) sorted populations from dye-pulsed
Detroit-562 tumours analysed by microarrays. f, RTqPCR validation by
human-specific TaqMan gene expression assays of differentially expressed

genes by microarray in the CD44+ DID+ and CD44+ DID populations.
Data are given as relative expression levels. Human -2-microglobulin
was used as internal control gene. n= 5, *P< 0.05, **P< 0.005, two-tailed
t-test. g, Gene expression overlapping analysis of the LR-CSC signatures
from SCC-25 and Detroit-562 tumours showing the top represented
common diseases and biological processes. Metastatic processes and lipid
metabolism-related categories are highlighted in red. P= 2.10 1049,
hypergeometric test. h, Correlation between CD36 expression and DiD
content for orthotopic transplants of SCC-25, JHU-029, Detroit-562,
FaDu, VDH-00, VDH-01 and VDH-02 cells (n=8 animals per cell line).
Numbers indicate percentages from the total GFP+Lin OSCC parental
tumour. Results are given as the mean s.e.m. (n=7 OSCC orthotopic
transplants; ***P= 0.0008, *P=0.03, two-tailed t-test).
ARTICLE RESEARCH
RESEARCH ARTICLE
Extended Data Figure 3 | LRCs correspond to CD36+ cells, and CD36
overexpression promotes metastatic initiation and progression.
a, CD36+ CD44bright OSCC cells detected by flow cytometry analysis of
tumours from orthotopic transplants. Tumours were obtained from OSCC
Detroit-562 (three independent experiments: exp. 1 n=3, exp. 2 n= 3,
exp. 3 n=4 mice), JHU-029 (three independent experiments: exp. 1 n= 3,
exp. 2 n=3, exp. 3 n=4 mice), SCC-25 (n=8 mice), FaDu (n= 8 mice),
VDH-00 (n=8 mice), VDH-01 (n=8 mice) and VDH-02 cells (n= 8
mice). Numbers indicate CD44bright CD36bright or CD44bright CD36low cells
in the represented gate, expressed as percentages from the total GFP+
Lin OSCC parental tumour. Histograms show the correlation between
CD36 expression and the DID content. The average counted events as
a function of dye fluorescence intensity is reported for each population
CD44bright CD36bright and CD44bright CD36low. b, BLI monitoring of
tumours generated by SCC-25 cells (empty vector (EV) n= 7 and Cd36
overexpression (OE) n=17 mice), or JHU-029 cells (EV n= 19 and Cd36

OE n=24 mice), transduced with PMSCV-EV (empty vector) or Cd36overexpression vector. Graphs show the frequency of developed tumours
(SCC-25 ***P= 0.05, JHU-029 ***P=0.03, Fisher exact test) and BLI
signal quantifications (primary tumour, *P= 0.01 and ****P< 0.0001;
metastasis, **P= 0.007, *P= 0.01, two-tailed t-test). Data are given as the
mean s.e.m. c, d, Haematoxylin and eosin staining (c) and anti-human
CD44 immunostaining (d) of lymph nodes isolated from animals reported
in a (n=5 animals per group). e, RTqPCR analysis of OSCC parental
and CD36OE cells. Human -2-microglobulin was used as internal
control gene (n=3 biological replicates, **P< 0.005, *P< 0.05, two-tailed
t-test), data are given as the mean s.e.m. f, g, Flow cytometry analysis
of OSCC tumours derived from PMSCV-EV or CD36OE cell transplants
(n=5 animals per group). Source data from mouse experiments are in
ARTICLE RESEARCH
RESEARCH ARTICLE
Extended Data Figure 4 | Depletion of CD36 inhibits metastatic
initiation and progression. a, BLI signal quantifications (*P= 0.01, twotailed t-test) and frequency of developed tumours (*P= 0.04, two-tailed
Fishers exact test) of PMSCV-EV and CD36overexpressing tumours
from VDH-00 primary cell line (PMSCV-EV, n= 7; CD36OE, n= 8).
b, BLI monitoring of tumours from FaDu cell line transduced with
either PLKO or shRNA CD36 (two independent experiments: exp1. and
exp.2, n=5 mice per group). Graphs show the frequency of developed
tumours, and BLI signal quantification (metastasis lymph node, *P= 0.05;
metastasis lung, **P= 0.002; two-tailed t-test). c, d, Flow cytometry
analysis of tumours from OSCC cells transduced with PLKO or shRNA
CD36#99. Numbers indicate the percentages of CD44bright CD36+,
CD44bright CD36 or CD44dim cells in the represented gate (n= 6 animals
per group). e, Relative RNA levels of CD36 in SCC-25 parental and

shRNA CD36 cells, determined by RTqPCR analysis using TaqMan gene
expression assay. Human -2-microglobulin was used as internal control
gene (n=3 biological replicates, ****P< 0.005, two-tailed t-test). Data
in a, b, e, are given as the mean s.e.m. f, Representative images of lungs
from mice transplanted with PLKO or shRNACD36 FaDu cells (PLKO,
n=5 mice; shRNA CD36#99, n= 5 mice). g, Haematoxylin and eosin
staining of metastatic lymph nodes from cells transduced with PLKO or
shRNA Cd36. h, Representative haematoxylin-eosin staining of primary
tumours from transplanted SCC-25 cells transduced with PLKO or
Cd36 shRNA (n=5 mice per group). Source data from mouse experiments
are in Supplementary Information.
ARTICLE RESEARCH
RESEARCH ARTICLE
Extended Data Figure 5 | CD36+ cells are defined by a lipid metabolism
and metastatic signature, and require the fatty acid -oxidation
enzyme ACSL1 to promote metastasis. a, b, Top categories for diseases
(a) and biological process (b) upregulated in CD36+ CD44bright cells.
c, Gene set enrichment analysis (GSEA) plot of CD36-associated
signatures, highlighting strong enrichment for fatty acid metabolism.
NES denotes normalized enrichment score. d, Comparative analysis
of overlapping genes between CD36+ CD44bright and CD44bright DID+
upregulated signature, highlighting over-represented genes associated
with lipid metabolism, cancer invasion and metastasis and transport
and metabolism of nucleoside drugs. P= 1.359 1016, hypergeometric
test. e, Flow cytometry analysis of in vitro SCC-25 cells co-cultured with
adipogenic OP-9 cells, showing the expression of three enzymes of fatty
acid -oxidation (ACADVL, ACADM and HADHA). Histograms show
the average normalized number of events as a function of fluorescence
intensity for the three enzymes (n=2 biological replicates). f, BLI
monitoring of tumours generated from OSCC cells transduced with
either scrambled shRNA (SCR, n=5 mice) or shRNA ACSL1#936
(n=5 mice). Graphs show the frequency of developed tumours and

the BLI signal quantification (**P= 0.001 and *P= 0.003, two-tailed
t-test). g, Haematoxylin and eosin staining of metastatic lymph nodes
from animals reported in f, showing the smaller metastases arising from
Acsl1 shRNA transplants as compared to the control SCR (n= 5 animals
per group). h, BLI monitoring of orthotopic transplants from CD36overexpressing JHU-029 cells transduced with either control (SCR, n= 10
mice) or shRNA ACSL1#936 (n=10 mice). Graphs show the BLI signal
quantification (metastasis: *P= 0.03 and *P= 0.03, two-tailed t-test) and
the frequency of developed tumours (CT vs OE-SCR *P=0.03 and OESCR vs OE-shACSL1 *P=0.04, Fisher exact test). i, Histogram shows the
average normalized number of events as a function of CD36 fluorescence
intensity. j, Relative RNA levels of OSCC cells reported in j, by RTqPCR
analysis. Human -2-microglobulin was used as internal control gene
(n=3 biological replicates, P= 0.03, two-tailed t-test). Data in f, h, j, are
given as the means.e.m. Source data from mouse experiments are in
ARTICLE RESEARCH
RESEARCH ARTICLE
Extended Data Figure 6 | CD36+ cells are stimulated by a high-fat
diet or adipocyte-conditioned medium, and require the ability of
CD36 to internalize fatty acids for their pro-metastatic potential.
a, Flow cytometry analysis of orthotopic transplants of Detroit-562 cells
transduced with PLKO or shRNACD36#98 or #99, from mice fed with
high-fat diet (HFD) or control diet (CD), analysed 4weeks after OSCC
injection. Numbers indicate CD44bright CD36+, CD44bright CD36 and
CD44dim (differentiated) cells in the represented gate, expressed as
percentages from the total GFP+ Lin OSCC parental tumour. n= 5
animals per group. b, Flow cytometry analysis of co-cultured SCC-25/
OP-9, SCC-25/adipogenic OP9 or SCC-25/HNCAFS (head and neck
cancerassociated fibroblasts) cells. Numbers indicate CD36+cells in the
represented gate, expressed as percentage. c, FACS analysis of co-cultured
Detroit-562 or SCC-25 with OP9 (control) or adipogenic OP9, showing
an increase in the percentage of CD36-positive cells in the adipogenic cocultures. Numbers indicate CD44bright CD36+ and CD44bright CD36 from
the total GFP+CD29 OSCC cells. d, CD36 mRNA relative expression
levels, measured by RTqPCR, from SCC-25 CD36 sorted cells either
co-cultured with adipogenic OP9 (Ad.OP9) cells or not, or from SCC-25
CD36+ sorted cells co-cultured with Ad.OP9 cells. In bd, OSCC were
co-cultured in vitro for 2days. e, Flow cytometry analysis of OSCC cells
co-cultured with adipogenic OP-9 cells or with 0.4mM palmitic acid
(PA). Histograms show the average normalized number of events as a
function of CD36 and CD44 fluorescence intensity. f, cDNA and amino
acid sequence of the CD36 receptor at the level of the point mutation
introduced to generate the fatty acid-binding site mutant, CD36-K164A
(left). Fatty acid uptake assay is shown for SCC-25 cells not transduced
(as control, CT) or transduced with CD36wt (overexpressing wild-type
CD36), shRNA Cd36 or CD36-K164A. g, BLI monitoring of transplants
from SCC-25 cells overexpressing CD36wt (wild-type, n= 10) or
CD36-K164A (n=10). Frequency of developed tumours is expressed as
percentage (*P=0.02, Fisher exact test), and BLI signal quantification is
expressed as the relative normalized photon flux (* P=0.05, two-tailed
t-test). Data are given as the mean s.e.m. h, FACS analysis of OSCC
cells overexpressing either CD36 wild-type (wt) or mutant (Lys164mut).
Histograms show the average normalized number of events as a function
of CD36 and CD44 fluorescence intensity. Source data from mouse
experiments are in Supplementary Information.
ARTICLE RESEARCH
RESEARCH ARTICLE
Extended Data Figure 7 | Inhibition of CD36 results in metastatic
lipotoxicity, and CD36+ cells are the only cells capable of initiating
metastasis. a, Representative haematoxylin and eosin staining of
metastatic lymph nodes from SCC-25-pLucGFP transplants with
overexpressed wild-type CD36 or CD36-K164A. Dashed line denotes the
areas surrounded by lipid droplets in the CD36-K164A-expressing cells.
b, c, Caspase-3 immunostaining of the metastases reported in a and in
Cd36 shRNA FaDu-pLucGFP metastatic lymph nodes, showing activated
casp-3-positive apoptotic cells in the vicinity of droplets. d, Relative
expression levels expressed as percentages of four populations, CD36+
CD44bright, CD36+ CD44dim, CD36 CD44bright and CD36 CD44dim, as
determined by FACS analysis of the primary tumour and metastasis of the
OSCC cell lines SCC-25, JHU-029, Detroit-562 and FaDu and the PDCs
VDH-00, VDH-01 and VDH-02 (n=4 biological replicates per cell line).
e, Genes differentially expressed between CD36+ CD44bright and CD36+
CD44bright populations validated by RTqPCR with human-specific
TaqMan gene expression assays in SCC-25 EV (empty vector), SCC-25
CD36-overexpressing and SCC-25 Cd36 shRNA cells grown

in vitro. Human -2-microglobulin was used as internal control gene
(n=4 biological replicates, *P< 0.05, **P< 0.005, ***P< 0.0005,
two-tailed t-test). f, OSSC cells were co-cultured with adipogenic OP9
cells, FACS-sorted and injected into the oral cavity of NSG mice. g, FACS
strategy to isolate CD36+ CD44bright, CD36 CD44bright and CD44bright
cells from in vitro SCC-25 cells co-cultured with adipogenic OP-9 cells.
Serial limiting dilutions of the different populations were injected
immediately after FACS sorting. h, i, BLI monitoring (h) and primary
tumour quantification (i) of mice injected with CD44bright CD36+ or
CD44bright CD36 cells. Yellow arrows denote increased affinity in injected
OSCC for the metastatic place, observed in some animals. j, k, Metastasisinitiating cell (MIC) frequency (j) and tumour-initiating cell (TIC)
frequency (k) of the three different populations in g, as determined by
ELDA software statistical analysis. Source data from mouse experiments
are in Supplementary Information.
ARTICLE RESEARCH
Extended Data Figure 8 | CD36+ cells recapitulate the cellular and

molecular heterogeneity of primary tumours and metastases when
orthotopically transplanted. a, Overview of experimental set-up.
Detroit-562 cells co-cultured with adipogenic OP-9 cells were FACSsorted to select the CD44bright and CD36+ CD44bright populations. Selected
cells were then injected orthotopically into NSG mice. Tumours were
collected after 4weeks, and cells were isolated for gene expression analysis
by microarray. b, CD36-associated signatures from lymph node metastases
arising from CD36+ CD44bright or primary tumour CD44bright transplants,
showing the top upregulated categories for diseases and biological

processes. c, GSEA analysis of lymph node metastases from CD36+
CD44bright and primary tumours from CD44bright transplants. Ranked lists
of primary tumour comparison versus top 300 genes of lymph node-Met
sorted by fold change (FC) and ranked lists of lymph node-Met CD36+
comparison versus top 300 genes of primary tumour sorted by fold change
(FC). Nominal P<0.0001. All source data from mouse experiments are in
RESEARCH ARTICLE
Extended Data Figure 9 | Anti-CD36 neutralizing antibodies inhibit

metastatic initiation, and cause metastatic regression of oral SCC.
a, BLI quantification of tumours from mice treated with anti-CD36
FA6.152 (anti-CD36 FA6.152, n=3 mice; IgG1, n= 3 mice; **P= 0.004,
two-tailed t-test). b, d, BLI monitoring of tumours from mice treated
daily with anti-CD36 JC63.1 (anti-CD36: n=5 mice; anti-IgA isotype
control, n=5 mice). Graphs show the BLI signal quantification (*P= 0.04,
two-tailed t-test). c, Representative pictures of metastatic lymph nodes
of animals treated daily with JC63.1 or IgA for 2.5weeks. e, Activated
caspase-3 immunostaining of metastatic lymph nodes of Detroit-562
transplants from mice treated with monoclonal anti-CD36 JC63.1

(10g per 100l), or with the IgA isotype control. f, BLI monitoring of
immunocompetent C3H/HeJ mice treated daily with monoclonal JC63.1
or IgA. Graphs show BLI signals from tumours (*P= 0.05, two-tailed
t-test). g, Fold change in metastasis BLI signal of the animals reported
in d. h, Representative haematoxylin and eosin staining of liver, spleen,
thymus and kidney of mice from f. No pathological differences related to
anti-CD36 treatment were found (n=10 animals per group). Data in
a, d, f, g are given as the mean s.e.m.
ARTICLE RESEARCH
Extended Data Figure 10 | Expression of CD36 correlates with poor

prognosis in several human tumours, and inhibition of CD36 inhibits
metastasis of human melanoma and luminal A breast carcinoma cell
lines. a, Correlation of CD36-associated signature expression or CD36
expression with overall and disease-free survival for patients. Red and
green lines denote patients whose tumours expressed signatures or CD36
higher and lower than the median, respectively. b, BLI signals from
metastasis developed in NSG mice injected with MCF-7 (PLKO, n= 10;
Cd36 shRNA, n=10 mice) and 501mel (PLKO, n= 10; Cd36 shRNA,
n=10 mice) cells (for breast MCF-7, *P= 0.04, two-tailed t-test and for
melanoma 501mel, ***P=0.0001 in liver metastasis and **P= 0.0003
in lung metastasis, two-tailed t-test). c, Relative proportion of developed
metastases from mice in a (*P=0.05, two-tailed Fishers exact test). d, BLI
signals from primary tumours and relative blood and lung GFP RNA levels
measured by qPCR analysis after intravenous injection of Detroit-562 and
SCC-25 cells transduced with empty vector (control) or shRNA Cd36.

Samples were collected immediately after injection (T-0h) and 12 and
48h (T-12h and T-48h, respectively) after injection (n=3 animals per
time point in each of the groups; *P 0.05, two-tailed t-test). Data in b,
d, are given as the mean s.e.m. e, GSEA of EMT genes in CD36+ and
CD36 cells sorted from primary oral lesions (generated from CD44bright
inoculated cells), or from lymph node metastases (generated from CD36+
CD44bright inoculated cells). CD36 cells express higher levels of EMT
genes than CD36+ cells in both the primary lesion and lymph node
metastases. Genes are ranked by t-statistic value. Enriched populations are
indicated for each of the plots. Lower panels show the GSEA analysis of the
same cohort of EMT genes compared between lymph node metastases and
primary oral lesions within CD36+ cells or CD36 cells. Source data from
mouse experiments are in Supplementary Information.

Targeting Metastasis-Initiating Cells PDF

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Targeting Metastasis-Initiating Cells PDF

Hochgeladen von

Copyright:

Verfügbare Formate

ARTICLE

Targeting metastasis-initiating cells

GloriaPascual1, AlexandraAvgustinova1, StefaniaMejetta2, MercMartn1, AndrsCastellanos1,

LRCs express lipid metabolism genes

included an over-representation of genes associated with lymphatic

CD36 is essential for metastasis

Overexpression of CD36 in cell lines or PDCs with low metastatic

Signal transduction pathway (dye+)

Figure 1 | CD44bright LRCs have lymphatic metastasis and lipid

CD36+cells respond to dietary lipids

The histological analyses of the few lymph node metastases that

PLKO #98 #99

PLKO #98 #99

BrdU+ cells (%)

PLKO #98 #99

Relative photon flux

Figure 2 | Modulation of CD36 expression strongly affects metastatic

OP9 parental cells or with squamous cell c arcinoma-associated

CD36+cells are metastasis-initiating

Primary tumours (%)

Cd36 Lys164 Cd36 Lys164

Cd36 Lys164 Cd36 Lys164

Figure 3 | CD36+ cells require fatty acid internalization to promote

mRNA expression was significantly lower in CD36 cells than in those

Primary tumours (%)

Primary tumours (%)

Relative photon flux

Relative photon flux

Relative photon flux

Primary tumours (%)

High-fat diet Control diet

Figure 4 | CD36-positive cells initiate and promote OSCC metastasis.

Fig. 8ac and Supplementary Table 5ac). Thus, when transplanted,

CD36-based anti-metastatic therapy

We next used two different anti-CD36 neutralizing antibodies:

CD36 correlates with poor prognosis

We next analysed publicly available data and found that CD36

Met remission (%)

metastasis of ovarian cancer cells2. We therefore expect that targeting

Photon flux ( 106)

Figure 5 | CD36-blocking antibodies therapeutically inhibit metastasis

disease-free survival and overall survival in patients with lung SCC,

Received 16 September 2015; accepted 16 November 2016.

Extended Data Figure 1 | See next page for caption.

fluorescence intensity. g, FACS strategy to FACS-sort CD44bright dye+,

Extended Data Figure 2 | See next page for caption.

human-specific TaqMan gene expression assays of differentially expressed

Extended Data Figure 3 | See next page for caption.

overexpression (OE) n=17 mice), or JHU-029 cells (EV n= 19 and Cd36

Extended Data Figure 4 | See next page for caption.

per group). e, Relative RNA levels of CD36 in SCC-25 parental and

Extended Data Figure 5 | See next page for caption.

(n=5 mice). Graphs show the frequency of developed tumours and

Extended Data Figure 6 | See next page for caption.

Extended Data Figure 7 | See next page for caption.

CD36-overexpressing and SCC-25 Cd36 shRNA cells grown

Extended Data Figure 8 | CD36+ cells recapitulate the cellular and

showing the top upregulated categories for diseases and biological

Extended Data Figure 9 | Anti-CD36 neutralizing antibodies inhibit

transplants from mice treated with monoclonal anti-CD36 JC63.1

Extended Data Figure 10 | Expression of CD36 correlates with poor

SCC-25 cells transduced with empty vector (control) or shRNA Cd36.

Das könnte Ihnen auch gefallen