Mesenchymal stem cells within tumour
stroma promote breast cancer metastasis
Antoine E. Karnoub1, Ajeeta B. Dash2, Annie P. Vo1, Andrew Sullivan2, Mary W. Brooks1, George W. Bell1,
Andrea L. Richardson3, Kornelia Polyak4, Ross Tubo2& Robert A. Weinberg1
Mesenchymal stem cells have been recently described to localize to breast carcinomas, where they integrate into the
tumour-associated stroma. However, the involvement of mesenchymal stem cells (or their derivatives) in tumour
when mixed with otherwise weakly metastatic human breast carcinoma cells, cause the cancer cells to increase their
metastatic potency greatly when this cell mixture is introduced into a subcutaneous site and allowed to form a tumour
xenograft. The breast cancer cells stimulate de novo secretion of the chemokine CCL5 (also called RANTES) from
mesenchymal stem cells, which then acts in a paracrine fashion on the cancer cells to enhance their motility, invasion and
metastasis. This enhanced metastatic ability is reversible and is dependent on CCL5 signalling through the chemokine
receptor CCR5. Collectively, these data demonstrate that the tumour microenvironment facilitates metastatic spread by
eliciting reversible changes in the phenotype of cancer cells.
The origins of the invasive and metastatic phenotypes of carcinoma
cells have been the subjects of intense investigation. Whereas some
current models depict these phenotypes as cell-autonomous altera-
tions specified by the genomes of cancer cells, alternative views pro-
cancer cells to paracrine signals that they receive from mesenchymal
cell types within the tumour-associated stroma. Although several
lines of evidence demonstrate the contributions of stromal cells to
primary tumour growth1, direct experimental demonstration of the
influence of these various cells on the metastatic abilities of cancer
of the mesenchymal cell types that are recruited into the stroma, and
to the elusive nature of the putative paracrine signals that are
exchanged between the mesenchymal and epithelial compartments
mesenchymal stem cell (MSC) is a cell type that is recruited in large
numbers to the stroma of developing tumours2. To characterize bet-
ter the role of this stromal cell in tumorigenesis, we set out to deter-
mine whether MSCs could supply contextual signals that serve to
promote cancer metastasis.
Mesenchymal stem cells are pluripotent progenitor cells that con-
tribute to the maintenance and regeneration of a variety of connect-
ive tissues, including bone, adipose, cartilage and muscle3. Although
MSCs reside predominantly in the bone marrow, they are also dis-
tributed throughout many other tissues, where they are thought to
serve as local sources of dormant stem cells4,5. The contributions of
MSCs to tissue formation become apparent only in cases of tissue
remodelling after injury or chronic inflammation. These conditions
from the injured or inflamed tissue that are then transmitted to the
bonemarrow,leadingtothemobilization ofmulti-potent MSCsand
have been shown to contribute to the formation of fibrous scars after
The formation of breast carcinomas is often accompanied by a
well-orchestrated desmoplastic reaction, which involves the recruit-
activities1. Such response closely resembles wound healing and scar
formation, and entails the constant deposition of growth factors,
cytokines and matrix-remodelling proteins that render the tumour
site a ‘wound that never heals’8. This suggests that, similar to sites of
injury,actively growingtumoursrecruitMSCsthrough thereleaseof
various endocrine and paracrine signals. Indeed, as we have found,
mouse stroma prepared from developing human MCF7/Ras or
MDA-MB-231 breast cancer xenografts is rich in cells with an ability
to generate fibroblastoid colony-forming units (CFU-F) in vitro
(Supplementary Fig. 1a), a hallmark of MSCs3. The absence of such
colonies from control Matrigel plugs or from neighbouring tissues
murine MSCs localize specifically to sites of neoplasia.
To investigate whether human breast cancer cells also have the
ability to attract human MSCs, we established a transwell assay
in which bone-marrow-derived human MSCs were allowed to
migrate towards media derived from MCF7/Ras or MDA-MB-231
cultures. We found that human MSCs migrated much more avidly
(,11-fold more) towards media derived from these cancer cells
than towards control media (Supplementary Fig. 1b). More im-
portantly, green fluorescent protein (GFP)-labelled human MSCs
infused into the venous circulation of mice bearing MCF7/Ras
or MDA-MB-231 human breast cancer xenografts localized speci-
fically to the developing tumours, with no observable accumulation
in other tissues, such as the kidneys (Supplementary Fig. 1c), liver
and spleen (data not shown). Such findings indicated that MSCs
arespecifically recruited bysubcutaneous breast xenografts, andcor-
roborated recent studies that described the localization of system-
ically infused MSCsto other types of malignancy, such as gliomas9,10,
1Whitehead Institute for Biomedical Research and Massachusetts Institute of Technology, Cambridge, Massachusetts 02142, USA.2Genzyme Corporation, Framingham,
Massachusetts01701,USA.3DepartmentofPathology,BrighamandWomen’s Hospital,Boston,Massachusetts02115,USA.4DepartmentofMedicalOncology,Dana-Farber Cancer
Institute, Harvard Medical School, Boston, Massachusetts 02115, USA.
Vol 449|4 October 2007|doi:10.1038/nature06188
MSCs enhance breast cancer metastasis
To investigate the functional consequences of the heterotypic inter-
MDA-MB-435 and HMLER (see Methods) human breast cancer
cells (BCCs) were mixed with bone-marrow-derived human MSCs
(hereafter referred to as MSCs) and injected subcutaneously into
immunocompromised mice. The growth kinetics of the MSC-
containing tumours (BCCs plus MSCs) were compared to those of
BCCs injected alone (BCCs) over the subsequent 8–12weeks, after
which the histopathology of the resulting tumours was studied.
We found that MSCs accelerated the growth of MCF7/Ras
tumours without affecting the kinetics of MDA-MB-231-, MDA-
MB-435- or HMLER-containing tumours (Fig. 1a). More impor-
tantly, whereas mice carrying tumours composed only of BCCs
exhibited few microscopic metastases in the lungs (Fig. 1b, d), mice
bearing the mixed MCF7/Ras1MSC, MDA-MB-2311MSC, MDA-
MB-4351MSC and HMLER1MSC tumours displayed a marked
increase in the numbers of micro- and macroscopic lung metastases
of BCC1MSC-bearing mice compared to their BCC-control litter-
mates revealed two-, three-, four- and sevenfold enhancements in
the overall numbers of detectable HMLER, MDA-MB-435, MCF7/
Ras and MDA-MB-231 metastatic deposits, respectively (Fig. 1c).
Furthermore, in contrast to the MDA-MB-231-bearing mice, the
MDA-MB-2311MSC-bearing mice showed metastases to various
other tissues, including the mammary glands (Supplementary
Table 1). Although all four of the tested cell lines exhibited enhanced
metastatic potential after admixture of MSCs, we chose to focus
further analysis on the MDA-MB-231 tumour model, because it
displayed the greatest relative increase in MSC-induced metastasis
without any concomitant effect on either tumour cell proliferation
(as revealed by Ki67 staining; Supplementary Fig. 2) or overall prim-
ary tumour growth kinetics.
We note that admixture of other types of mesenchymal cells—
specifically WI-38 or BJ human fibroblasts (Supplementary Fig. 3
into host mice did not result in either enhanced growth kinetics
(Supplementary Fig. 3a, b) or increased numbers of lung metastases
(Supplementary Fig. 3c, d). Taken together, these observations indi-
cated that the metastasis-enhancing powers were a specific property
of admixed MSCs or derivatives thereof.
in nearby separate sites of injection did not affect the metastatic
potential of the resulting primary tumours (data not shown), indi-
cating that MSCs could enhance cancer metastasis only when they
were in close proximity to the engrafted BCCs. This influence might
be ascribed to various effects that MSCs exert on the commingled
carcinoma cells. Thus, the MSCs might favour the outgrowth of rare
variants within the MDA-MB-231 cell populations that exhibit
unusually high metastatic powers. Alternatively, the MSCs might
cause otherwise weakly metastatic MDA-MB-231 cells to acquire
enhanced metastatic abilities. This latter mechanism suggests the
possibility that the acquisition of the metastatic phenotype might
be reversible, in that carcinoma cells might revert to a lower meta-
static state once they were no longer in close contact with MSCs.
To resolve between these two mechanisms, explants of MDA-MB-
231 cells were prepared from BCC plus MSC primary tumours (T-
explants) as well as from their derived lung metastases (L-explants),
expanded in vitro, cleared from contaminating stromal components,
and then re-injected into subcutaneous sites in host mice in order
to evaluate their respective metastatic powers (Fig. 2a). Although
the growth rate of the resulting L-explant primary tumours was
marginally enhanced compared to their T-explant counterparts
(Fig. 2b, c), these L-explant cells were no more metastatic than the
parental T-explant cancer cells (Fig. 2d). This suggested that the
Days after injection
Tumour volume (mm3)
10192431 38 4610 17 24 31 38 46 72 80 8913 19 26 33 40 47 54 61 68 75 81
1 mm1 mm
Metastasis index (fold)
021 31 38 49 56 63 70 78
Figure 1 | MSCs promote breast cancer metastasis. a, Tumour volume
measurements (mean6s.e.m.) of 500,000 GFP-labelled BCCs injected
subcutaneously into nude mice with or without 1.53106MSCs.
alone, n55–7 mice per group; squares, BCCs plus MSCs, n55–8 mice per
group. b, Representative bright-field/fluorescence images of lungs of mice
bearing the indicated tumours. Cancer colonies are in green. MCF7/Ras-
grow to comparable sizes to their MCF7/Ras1MSC counterparts. c, The
lung metastasis indices pooled within each cohortof mice in a are expressed
as fold increase (6s.e.m.) over controls. Data shown are representative of
multiple repeats. Asterisk, P,0.01, double asterisk, P,0.05 using one-
tailed Student’s t-test. d, Representative haematoxylin-and-eosin-stained
sections of lungs of mice bearing the indicated tumours. Metastases are
delineated by a dashed line.
NATURE|Vol 449|4 October 2007
MSC-induced metastatic powers reflected a reversibly induced trait
oftheMDA-MB-231 cells,andthattheability ofthesecellstometas-
tasize to the lungs wasa consequence of their‘education’ by MSCsin
the primary tumour rather than the selection of rare variants of
MDA-MB-231 cells that display elevated metastatic potency in a
The effects that the MSCs exerted on the BCCs might have
occurred within the site of primary tumour formation. Alterna-
tively, the MSCs might have accompanied the metastasizing BCCs
to sites of metastasis formation. To distinguish between these two
possibilities, we admixed ds-red-labelled MSCs to GFP-labelled
MDA-MB-231 cells and implanted the mixture subcutaneously in
host mice. We found that the tumour-derived lung metastases con-
tained green-labelled MDA-MB-231 cells but no detectable red-
labelled MSCs (or their derivatives; Supplementary Fig. 4a) when
scored 4, 5 or 6weeks after primary tumour implantation. The
absence of red-labelled MSCs from the lung metastatic sites cannot
be ascribed to an inhospitable lung parenchyma, as MSCs that lodge
environment for ,6weeks after injection (Supplementary Fig. 4b).
Hence, it appeared that the admixed MSCs do not migrate in large
numbers to the sites of metastasis, and that they exerted their pro-
metastatic effects on BCCs in the context of primary tumours.
CCL5 in MSC-induced metastasis
The aforementioned observations indicate that MSCs supply locally
acting paracrine cues that induce BCCs within primary tumours to
metastasize. To understand this crosstalk better, in vitro co-cultures
of MDA-MB-231 breast cancer cells and MSCs were established and
their conditioned media were screened for the levels of various cyto-
kines, chemokines and growth factors using the Luminex-based Bio-
Plex suspension array system (Fig. 3a). In some cases, the resulting
Metastasis index (fold)
Days after injection
(× 100 mm3)
14 17 21 24 28 31 34 38 42 45 48 66 71 77
Primary tumour explants
Tumour mass (g)
Figure 2 | MSC-induced increase in the metastasis of MDA-MB-231 cells
involves reversible mechanisms. a, BCCs were recovered from lung or
primary tumour tissues, cleared of stromal contaminants by culture in
blasticidin-containing media (5mgml21), and re-injected as primary
subcutaneous tumours in recipient animals. b, Tumour growth
tumour-derived (T-explant) MDA-MB-231 cells inoculated subcutaneously.
Data shownare representative ofmultiple independentexperiments inwhich
in parallel. MDA-MB-231-T-explant (n58 mice); MDA-MB-231-L-explant
one-tailed Student’s t-test and indicates no statistical significance. d, Lung
t-test and indicates no statistical significance.
MDA+MSC (2:1 ratio)
CCL5 levels (pg ml–1)
Fold CCL5 induction
CCL5 (A.U. × 100)30
MSC (from MDA tumour)
BCC (from MDA tumour)
Figure 3 | The interaction of BCCs with MSCs causes a rise in the levels of
CCL5. a, MDA-MB-231, MSCs, or MDA-MB-2311MSCs were cultured in
were normalized to the levels observed in the media of BCCs cultured alone.
Data are expressed as fold induction6s.d. of triplicates. Asterisk indicates
fromco-cultured MSCs by a 0.4-mmmembrane.CCL5 levels were probedby
d, CCL5 ELISA on the supernatants of MSC-siluc (MSC.c), MSC-siCCL5.1
(MSC.1) and MSC-siCCL5.5 (MSC.5) co-cultured with MDA-MB-231-siluc
(MDA.c), MDA-MB-231-siCCL5.1 (MDA.1), or MDA-MB-231-siCCL5.5
(MDA.5). Data are expressed as means6s.d. of triplicates in arbitrary units
(A.U.). e, RT–PCR analyses of CCL5 in MSCs and BCCs sorted from
GFP–MSC1MDA-MB-231 tumours (3:1 ratio) 4weeks after tumour
implantation. Tissue-cultured MSCs (TC-MSC) and MDA-MB-231/CCL5
cells were used as controls. GAPDH was used for equal loading.
NATURE|Vol 449|4 October 2007
levels of certain released factors (for example, interferon-c or
tumour-necrosis factor-a) reflected the additive contributions of
the two cell types when cultured on their own. Notably, the levels
of only one cytokine, CCL5, reflected a synergistic interaction
between the MSCs and BCCs, as it accumulated to levels ,60-fold
higher than those produced by pure BCC cultures (Fig. 3a). This
cooperative induction of CCL5 was proportional to the numbers of
ent as early as the third day of co-culture (Fig. 3b). Moreover, this
induction required close physical contact between MSCs and cancer
cells, because it failed to occur when the two cell populations were
separated by a permeable membrane (Fig. 3c).
We undertook to determine the source of the CCL5 produced
under conditions of co-culture. To do so, we stably reduced the
expression of CCL5 in MDA-MB-231 cells by.80% using short
hairpin (sh)RNA (variant siCCL5.1; Supplementary Fig. 6). Impor-
with MSCs continued to allow accumulation of CCL5 in the culture
supernatants to levels that were comparable to those observed in the
that the source of CCL5 was the admixed neighbouring MSCs.
Indeed, inhibition of CCL5 protein expression in MSCs using the
same shRNA hairpin vector (MSC.1; Fig. 3d) resulted in more than
75% reduction of CCL5 protein levels in the co-cultures, indicating
that the MSCs were the major source of the CCL5 observed on co-
CCL5 levels in the media of MSCs or MDA-MB-231 cells separated
from one another after 3days of co-culture indicated astrong induc-
tion of CCL5 in the culture of MSCs, but not that of BCCs (Sup-
plementary Fig. 5b). Finally, polymerase chain reaction with reverse
culture-derived MSCs (Supplementary Fig. 5c), as well as from the
MSCs isolated from MDA-MB-2311MSC tumours ,4weeks after
tumour implantation (Fig. 3e), indicated a strong accumulation of
CCL5 messenger RNA, suggesting that an active signal transduction
pathway is triggered in MSCs by the nearby BCCs.
example, CCL5 levels in the plasma of breast cancer patients have
been correlated with the severity of the disease, and localized CCL5
protein expression was found to be elevated in invasive tumours
when compared to in situ ductal tumours or benign lesions16,17.
However, the precise contributions of CCL5 to cancer development
and progression are poorly understood. To investigate further the
possible causal role of CCL5 in cancer cell metastasis, we over-
expressed this chemokine in the MDA-MB-231 BCCs (Supple-
mentary Fig. 7a) and analysed its effects on cancer cell growth and
tumorigenesis. The overexpressed CCL5 did not confer any pro-
liferative advantage on cultured cancer cells when compared with
those lacking such overexpression (Supplementary Fig. 7b), and had
no effect on the ability of BCCs either to grow in an anchorage-
independent fashion in vitro (Supplementary Fig. 7c), or to form
primary subcutaneous tumours in immunocompromised mice (Fig.
4a). However, these tumours exhibited a ,5-fold enhancement in
their metastatic potential when compared with control tumours
lacking ectopic CCL5 (Fig. 4a). Similarly, overexpression of CCL5
in WI-38 fibroblasts sufficed to enable these cells to promote the
metastasis of admixed MDA-MB-231 BCCs (Fig. 4b), indicating that
MSC-induced metastasis by the BCCs.
CCL5 promotes lung colonization
Previous reports have described an important role for CCL5 as a
chemoattractant for stromal cells, such as macrophages, that express
one of the receptors for CCL5, CCR5 (refs 18, 19). Furthermore,
CCL5 expression has been associated with increased tumour neovas-
cularization, suggestingthatendothelial cells,which expressavariety
of chemokine receptors, may also be attracted by CCL5 to sites of
tumour formation, thereby enhancing tumour angiogenesis20. Such
observations suggest that CCL5 may contribute to breast cancer
metastasis through the recruitment of a number of stromal cell types
to sites of primary tumour growth.
However, immunohistochemical analyses indicated that the
MDA-MB-231 control and CCL5-overexpressing MDA-MB-231
(MDA-MB-231/CCL5) tumours exhibited comparable numbers of
tumour-infiltrating macrophages and had similar vessel densities (as
evident by F4/80 and MECA-32 staining for macrophages and
endothelial cells, respectively; Supplementary Fig. 8). In addition,
of other stromal cells, such as SMA-positive cells, in the examined
the observed CCL5-induced metastasis could not be attributed to
significant effects on the numbers of the major constituents of the
stroma or to the vascularity of these tumour xenografts.
facilitated by their transdifferentiation through the process termed
Extravasated clusters (mean)
LY – – + +
Nodules per lung (mean)
Tumour mass (mg)
Tumour mass (mg)
Metastasis index (fold) Metastasis index (fold)
Figure 4 | CCL5 enhances breast cancer cell migration, invasion and
metastasis. a, A total of 500,000 MDA-MB-231/vector (ctrl) or MDA-MB-
231/CCL5 cells were injected subcutaneously in NOD/SCID mice. Tumour
masses (mean6s.e.m., n56 each group) were taken at 10weeks. Lung
metastasisindices areexpressed as fold increase (6s.e.m.)over controls. Data
shown are representative of multiple repeats. Asterisk, P,0.01 in one-tailed
Student’s t-test. b, A total of 500,000 MDA-MB-231 cells were admixed to
250,000 WI-38 fibroblast controls (WI-38/vector) or WI-38 fibroblasts
overexpressing CCL5 (WI-38/CCL5) and were injected subcutaneously in
NOD/SCID mice. Tumours (n55 per group) were excised and weighed at
12weeks. Masses shown represent mean6s.e.m. Lung metastasis indices are
expressed as fold increase (6s.e.m.) over controls. Asterisk, P,0.01 in one-
tailed Student’s t-test. c, A total of 800,000 indicated BCCs were introduced
into the circulation of NOD/SCID hosts. GFP-positive cancer colonies in the
231 controls, n516 mice; MDA-MB-231/CCL5, n518 mice). Asterisk,
P,0.01 in one-tailed Student’s t-test. d, Western blot analysis of lysates of
MDA-MB-231 control or MDA-MB-231/CCL5 cells. b-Actin was used as a
loading control. e, Transwell migration or Matrigel invasion assays on 50,000
P,0.05; double asterisk, P,0.05; triple asterisk, P,0.01 in one-tailed
Student’s t-test. f, One million GFP-labelled BCCs were injected into the tail
vein of NOD/SCID mice. Lungs were processed 48h later and examined for
extravasated cells. Bars represent means6s.e.m. (MDA-MB-231 cells, n57
mice; MDA-MB-231/CCL5, n510 mice). Asterisk, P,0.01 in one-tailed
orMDA-MB-231/CCL5cells platedwithorwithoutthe phosphatidylinositol-
3-OH kinase inhibitor LY290042 (0.5mM); representative experiment shown;
asterisk, P,0.01 in one-tailed Student’s t-test.
NATURE|Vol 449|4 October 2007
the epithelial-to-mesenchymal transition (EMT), in which cells shed
their epithelial characteristics and acquire instead a series of mesen-
chymal markers that enable their invasiveness and intravasation21.
of mesenchymal markers such as fibronectin (data not shown), the
MDA-MB-231 cells studied here exist in an intermediary phenotypic
in vitro and are still responsive to EMT-inducing stimuli in culture. In
fact, we observed that ectopic CCL5 expression did not cause MDA-
the expression of mesenchymal markers closely associated with the
EMT process, namely vimentin, N-cadherin (Supplementary Fig. 9c)
not directly promote the EMT programme of MDA-MB-231 cells.
We proceeded to explore an alternative possibility: that CCL5
expression affected some of the later, critical steps of the invasion–
metastasis cascade, namely the lodging of cancer cells in secondary
organs and the subsequent step of colonization. For that purpose,
MDA-MB-231/CCL5 cells were injected intravenously into host
mice, and the lungs of these hosts were examined ,6weeks later
using fluorescence microscopy. These experiments revealed that
CCL5-overexpressing cells indeed had a significant ,1.8-fold
advantage over their control counterparts in colonizing the lungs
(Fig. 4c), suggesting that CCL5 exposure has effects on later steps
of the invasion–metastasis cascade. We note once again that this
enhanced tissue-colonizing ability was not due to CCL5’s effects on
cellular proliferation measured either in vitro (Supplementary Fig.
7b) or in vivo (Supplementary Fig. 7g, Ki67 staining).
Because improved colonization can be due to enhanced cellular
survival, we tested whether CCL5 protects against apoptosis.
Notably, we found that MDA-MB-231/CCL5 cells exhibited higher
levels of the Ser473-phosphorylated, activated form of Akt, but
exhibited no difference in the levels of other pro-survival proteins,
such as Bcl-XL or Bcl-2 (Fig. 4d), or a reduction in the levels of
pro-apoptotic molecules such as BAX or BAD (data not shown).
Moreover, we found that overexpression of CCL5 had no effect on
the ability of MDA-MB-231 cells to withstand serum deprivation
(Supplementary Fig. 7b), loss of substrate anchorage (Supplemen-
tary Fig. 7d), or hyperoxia (data not shown). We also observed that
ectopic CCL5 expression did not protect MDA-MB-231 cells from
doxorubicin-induced apoptosis monitored using western blots for
cleaved caspase-3 (CC3) and cleaved PARP (as markers ofapoptosis;
Supplementary Fig. 7e), or TdT-mediated dUTP nick end labelling
(TUNEL) assays (Supplementary Fig. 7f). Finally, immunohisto-
chemical analyses on control and CCL5-overexpressing tumours
revealed only minor differences in the levels of apoptotic CC3-
positive cancer cells among the examined groups (Supplementary
Fig. 7g, h). Together, these observations suggested that CCL5 does
that the observed enhancement of lung colonization was not a con-
sequence of significant anti-apoptotic activities of CCL5.
Akt serves as a key relay switch for upstream signals that promote
Akt phosphorylation did not correlate with enhanced protection
against apoptosis, we tested whether the CCL5-enhanced lung col-
onization could be due to an increased ability of MDA-MB-231/
CCL5 cells to invade from the microvasculature into the lung
parenchyma through the process of extravasation. Indeed, ectopic
expression of CCL5 enhanced the motility of MDA-MB-231 cells
fold in either high or low serum conditions, respectively (Fig. 4e).
Notably, when we flushed the lungs of mice 48h after BCC tail-vein
injection—in order to remove most cells that remained within the
microvasculature of the lungs and thus had not extravasated—we
found twice as many deposits in the MDA-MB-231/CCL5-injected
clear effect of CCL5 on cancer cell extravasation.
Finally, we investigated the role of Akt in mediating the actions of
CCL5 on cellular motility by using the phosphatidylinositol-3-OH
Metastasis index (fold) Metastasis index (fold)
Figure 5 | CCL5–CCR5 interaction is essential
for the MSC-induced metastasis.
a, Immunofluorescence analysis of CCR5
distribution in MDA-MB-231 cells cultured with
MSCs. DAPI (for nuclei staining) is in blue;
CCR5 detected in green. Arrowheads denote
MSCs. b, Western blot analysis showing CCR5
expression in MDA-MB-231/silacZ, MDA-MB-
231/siCCR5(809) and MDA-MB-231/
siCCR5(186) lysates. b-Actin was used as a
loading control. c, A total of 500,000 cells of the
MDA-MB-231 variants in b were co-mixed with
nude mice. Mice were killed when tumours
reached 1cm in diameter and the metastasis
index was calculated for each cohort (n55 per
group). Results represent means6s.e.m.;
asterisk, P,0.05 using one-tailed Student’s
t-test. d, Anti-CCL5 neutralizing antibody or
control IgG was administered intraperitoneally
(n59) or MDA-MB-2311MSC tumours
(n511). Representative lung pictures of the
indicated cohorts are shown. e, Lung metastasis
indices of mice in d. Data shown are
representative of means6s.e.m. Asterisk,
P,0.05 in one-tailed Student’s t-test.
NATURE|Vol 449|4 October 2007
kinase inhibitor LY294002. Drug concentrations that did not inhibit
the basal motility levels of MDA-MB-231 cells blocked the elevation
of motility induced by ectopic CCL5 expression (Fig. 4g). These
results, when taken together, suggest that the observed CCL5-
enhanced lung colonization could be ascribed, in significant part,
to its ability to promote extravasation and/or motility of cancer cells
at sites of dissemination rather than promoting the survival and/or
proliferation of these cells.
Essential role for the CCL5–CCR5 loop
CCL5 acts through three G-protein-coupled receptors, termed
CCR1, CCR3 and CCR5 (ref. 23). CCR5 has been determined to be
surface expression through dominant-negative mutants abrogated
the ability of these cells to respond to CCL5 chemotaxis24. We there-
fore focused our efforts on evaluating the importance of the CCL5–
CCR5 interactions in MSC-induced metastasis.
We confirmed that CCR5 is expressed by MDA-MB-231 cells and
not by MSCs (Fig. 5a), supporting the notion that MSC-derived
CCL5 acts primarily in a paracrine fashion on MDA-MB-231 cells
in the BCC and MSC mixed cell populations described above. To
probe whether the observed MSC-induced metastasis required
CCL5–CCR5 interactions, we inhibited CCR5 expression in MDA-
MB-231 cellsby morethan 85%through shRNAknockdown (ref.25
and Fig. 5b), and mixed these cells with MSCs before implantation
into host mice. Indeed, inhibition of CCR5 expression in the BCCs,
achieved using either of two different shRNA constructs, abrogated
the ability of MSCs to enhance the metastasis of MDA-MB-231 cells
(Fig. 5c). Furthermore, neutralization of CCL5 protein using intra-
peritoneal injections of an anti-human CCL5 monoclonal antibody
also abrogated the MSC-induced metastasis by MDA-MB-231 cells
(Fig. 5d, e). In addition, MSCs in which CCL5 expression was inhi-
bited by shRNA knockdown failed to promote metastasis of the
results underscore the critical importance of the CCL5–CCR5 para-
crine interactions in enabling MSCs to induce metastasis of the
Certain models of metastatic progression propose that cancer cell
invasion and metastasis from the primary tumour site are strongly
influenced by contextual signals emanating from the stroma of the
primary tumour. It follows that if carcinoma cells are subsequently
deprived of such signals, they may revert to an earlier phenotypic
state in which they no longer display the traits of high-grade malig-
nancy. Indeed, such a model has been proposed previously by others
on the basis of indirect evidence21. Here, we demonstrate that at least
one mesenchymal cell type, the MSC, can expedite tumour meta-
populations into their midst, subsequent interactions between the
MSCs (or their derivatives) and the BCCs endow the latter with
invasive and metastatic properties.
Although the recruitment of labelled MSCs to tumour xenografts
esis, there is currently no available way to quantify with any accuracy
thenumber of MSCs inactual human tumours,inpartbecausenoset
ofmarkershas been identifiedthatcanuniquely stain thesecells with-
out concomitantly staining other mesenchymal types in the tumour-
associated stroma6. Our demonstration that the stroma derived from
tumour xenografts contained appreciable numbers of murine MSCs
in developing tumours. Interestingly, the use of CD10—one of the
of human primary invasive breast carcinomas yielded a population of
cells that expresses a number of other markers collectively used to
characterize human MSCs (for example, CD44, CD105 and CD106;
Fig. 6a). This suggested that, similar to tumour xenografts, human
carcinomas also acquire significant numbers of MSCs. Furthermore,
signature associated with poor prognosis of breast cancers26(SFT;
Fig. 6b, c), is also enriched in the leukocyte- and endothelial cell-free
stroma of primary invasive ductal carcinomas (Fig. 6d), specifically in
the CD10-positive compartment27(Fig. 6e). Collectively, these obser-
vations argue strongly for a significant association between stromal
CCL5 levels, MSCs and human invasive breast cancers.
–2.0 –1.4 –0.9 –0.3 0.3 0.91.4 2.0
Figure 6 | Stromalfibroblasticcellsofhumaninvasiveductalcarcinomasare
rich in MSC markers and overexpress CCL5. a, SAGE TreeView display of
MSC markers expressed in stromal CD10-positive cells from invasive
(green) to high (red). Wide blocks indicate expression ratios of tumours
classified as desmoid-type fibromatosis (DTF; yellow outline, n510) or
solitary fibrous tumours (SFT; blue outline, n513); narrow blocks are other
soft-tissue tumours (n532). c, Box plot showing that CCL5 expression is
between SFT and DTF was tested with the Welch’s test. d, CCL5 Affymetrix
gene expression in the stroma of human invasive ductal cancers compared to
thatinnormal cancer-free breast tissue (indicated as‘Normal’; see Methods).
e, CCL5expressionismostlyrestricted to the CD10-positive fibroblasticcells
derived from invasive ductal cancers. The heatmap shown is a cluster of
various groups indicated in a, d and e are found in ref. 27.
NATURE|Vol 449|4 October 2007
Although we have focused here on CCL5 in the MSC–MDA-MB-
231 cell interactions, CCL5 seems to have an equally critical involve-
ment in the functional interaction of MSCs with MDA-MB-435
human BCCs. CCL5 levels accumulate synergistically when the two
cell types are co-cultured together (Supplementary Fig. 10a), and
down failed to promote metastasis by MDA-MB-435 cells to which
MSC-induced metastasis of MCF7/Ras or HMLER cells, which may
depend on other paracrine factors such as VEGF and interleukin-8.
the possible utility of a variety of CCL5 analogues and CCR5 antago-
tothe lungthat are,onisolation andre-injection into recipient mice,
no more metastatic than their predecessors in the primary tumour
(Fig. 2e). This indicated that acquisition of increased metastatic
powers by these tumour cells was reversible, and suggested that the
maintenance of this phenotype depends on continuing contact with
stromal cells. If extended to other tumour types, the present results
hold important implications for the molecular analysis of malignant
progression. They suggest that many of the cellular functions assoc-
iated with invasion and metastasis are often not expressed constitu-
tively by carcinoma cells, but rather only transiently in response to
environment. If so, analysis of the gene expression patterns of bulk
primary tumour populations may fail to detect the expression of key
genes mediating invasiveness and metastasis, if only because they are
being transiently expressed in minor subpopulations of cells within
such tumours. Additionally, attempts at determining the metastatic
propensities of tumours may need to be focused on the genes and
proteins that confer responsiveness of primary tumour cells to stro-
malsignals, rather than onthe genes andproteins that directly medi-
ate the cellular phenotypes of invasion and metastasis.
Cells labelled with GFP or ds-red, or harbouring various overexpression or
shRNA constructs, were generated by viral transduction followed by FACS
enrichment or antibiotic selection. Xenograft experiments were conducted in
nude or NOD/SCID mice and metastasis was estimated using fluorescence
microscopy. The levels of cytokines, growth factors and chemokines were
by intraperitoneal injections. See Methods for detailed information regarding
cell culture, viral infections, in vivo colonization and extravasation assays, RT–
PCR, TUNEL and anoikis assays, immunohistochemical and immunofluore-
scence determinations, western blotting, and antibodies used.
Full Methods and any associated references are available in the online version of
the paper at www.nature.com/nature.
Received 13 April; accepted 14 August 2007.
Hall, B., Andreeff, M. & Marini, F. The participation of mesenchymal stem cells in
tumor stroma formation and their application as targeted-gene delivery vehicles.
Handb. Exp. Pharmacol. 180, 263–283 (2007).
Pittenger, M. F. et al. Multilineage potential of adult human mesenchymal stem
cells. Science 284, 143–147 (1999).
Young, H. E. et al. Clonogenic analysis reveals reserve stem cells in postnatal
in the connective tissues of skeletal muscle and dermis derived from fetal, adult,
and geriatric donors. Anat. Rec. 264, 51–62 (2001).
cells: molecular control of expansion and differentiation. Exp. Cell Res. 306,
on the malignant phenotype. Mol. Med. Today 6, 324–329 (2000).
cells in a rat glioma model. Gene Ther. 11, 1155–1164 (2004).
10. Nakamizo, A. et al. Human bone marrow-derived mesenchymal stem cells in the
treatment of gliomas. Cancer Res. 65, 3307–3318 (2005).
11.Hung, S. C. et al. Mesenchymal stem cell targeting of microscopic tumors and
tumor stroma development monitored by noninvasive in vivo positron emission
tomography imaging. Clin. Cancer Res. 11, 7749–7756 (2005).
12. Menon, L. G. et al. Differential gene expression associated with migration of
mesenchymal stem cells to conditioned medium from tumor cells or bone
marrow cells. Stem Cells 25, 520–528 (2007).
13. Komarova, S., Kawakami, Y., Stoff-Khalili, M. A., Curiel, D. T. & Pereboeva, L.
Mesenchymal progenitor cells as cellular vehicles for delivery of oncolytic
adenoviruses. Mol. Cancer Ther. 5, 755–766 (2006).
14. Khakoo, A. Y. et al. Human mesenchymal stem cells exert potent antitumorigenic
effects in a model of Kaposi’s sarcoma. J. Exp. Med. 203, 1235–1247 (2006).
15. Studeny, M. et al. Bone marrow-derived mesenchymal stem cells as vehicles for
interferon-b delivery into tumors. Cancer Res. 62, 3603–3608 (2002).
16. Luboshits, G. et al. Elevated expression of the CC chemokine regulated on
activation, normal T cell expressed and secreted (RANTES) in advanced breast
carcinoma. Cancer Res. 59, 4681–4687 (1999).
in patients with breast or cervical cancer. Clin. Cancer Res. 7, 285–289 (2001).
18. Azenshtein, E.etal.TheCCchemokineRANTESinbreastcarcinomaprogression:
regulation of expression and potential mechanisms of promalignant activity.
Cancer Res. 62, 1093–1102 (2002).
19. Robinson, S. C. et al. A chemokine receptor antagonist inhibits experimental
breast tumor growth. Cancer Res. 63, 8360–8365 (2003).
20. Hillyer, P.& Male, D.Expression ofchemokines on the surface ofdifferent human
endothelia. Immunol. Cell Biol. 83, 375–382 (2005).
21. Thiery, J. P. Epithelial-mesenchymal transitions in tumour progression. Nature
Rev. Cancer 2, 442–454 (2002).
22. Toker, A. & Yoeli-Lerner, M. Akt signaling and cancer: surviving but not moving
on. Cancer Res. 66, 3963–3966 (2006).
23. Tanaka, T. et al. Chemokines intumor progression and metastasis. Cancer Sci. 96,
24. Mira, E. et al. A role for chemokine receptor transactivation in growth factor
signaling. EMBO Rep. 2, 151–156 (2001).
cells by lentiviral-mediated delivery of small interfering RNA against CCR5. Proc.
Natl Acad. Sci. USA 100, 183–188 (2003).
26. West, R. B. et al. Determination of stromal signatures in breast carcinoma. PLoS
Biol. 3, e187 (2005).
27. Allinen, M. et al. Molecular characterization of the tumor microenvironment in
breast cancer. Cancer Cell 6, 17–32 (2004).
29. Karnoub, A. E. & Weinberg, R. A. Chemokine networks and breast cancer
metastasis. Breast Dis. 26, 75–85 (2006).
30. Palani,A.&Tagat,J.R.Discovery anddevelopmentofsmall-molecule chemokine
coreceptor CCR5 antagonists. J. Med. Chem. 49, 2851–2857 (2006).
Supplementary Information is linked to the online version of the paper at
Acknowledgements We thank F. Reinhardt for assistance in animal studies, A. Lu
for technical help, J. Yao for SAGE data analysis and the MIT Comparative
Pathology Laboratory for immunohistochemical analyses. We are grateful to
of the Susan G. Komen Breast Cancer Foundation. R.A.W. is an American Cancer
research is supported by grants from the Breast Cancer Research Foundation
(R.A.W.), the Ludwig Trust (R.A.W.), the Susan G. Komen Breast Cancer
Foundation (R.A.W.) and the Dana-Farber/Harvard Cancer Center Specialized
Program of Research Excellence (SPORE) in Breast Cancer (A.E.K., R.A.W. and
Author Contributions A.E.K. conceived and designed this study, and performed
most experiments; R.A.W. supervised research; A.E.K. and R.A.W. wrote the
manuscript; A.B.D. and R.T. provided human MSCs; A.B.D. helped in in vivo CCL5
neutralization; A.S. helped in the Luminex screens; A.P.V. and M.W.B. provided
technical support in tissue culture, ELISA, western blot, RT–PCR and soft-agar
analyses; G.W.B. performed CCL5 analysis on soft tumour expression data; A.L.R.
obtained and classified the clinical specimens; K.P. fractionated the clinical
samples and performed SAGE analyses; and A.L.R. performed the microarray
analysis on sorted stroma.
at http://www.ncbi.nlm.nih.gov/geo, GSE8977. Reprints and permissions
information is available at www.nature.com/reprints. Correspondence and
requests for materials should be addressed to R.A.W. (firstname.lastname@example.org).
NATURE|Vol 449|4 October 2007
Cell lines. The MCF7/Ras31and MDA-MB-231 (ATCC HTB-26) cancer cells
were infected with pWZL-blasticidin-GFP-expressing retroviral particles and
100 unitsml21penicillin, 100mgml21streptomycin and 2mM L-glutamine
(complete medium) at 37uC in 5% CO2. The MDA-MB-435 and human mam-
mary epithelial cells HMLER32were infected with pRRL3-GFP-expressing lenti-
virus and grown in complete medium or MEGM media with bovine pituitary
extract,respectively.Bone-marrow-derivedhumanMSCs were isolatedfrom hip
aspirates of healthy volunteers, propagated as previously described33, and used
from three different donors were assayed and exhibited consistent results. MSCs
ds-red (Clontech) particles. WI-38 human embryonic lung fibroblasts (ATCC
CCL-75) were grown in Dulbecco’s modified Eagle medium supplemented with
10% calf serum and were used before the 20th passage. The MDA-MB-231 cells
infection of parental cells with pLZ-CCL5-IRES-gfp24viral particles. Control
cultures were infected with pLZ-IRES-gfp retrovirus. WI-38 fibroblasts overex-
Blasticidin-CCL5-expressing viral particles. MDA-MB-231 cells lacking CCR5
expression were generated by lentiviral infection of parental cells with FG12-
with the control FG12 vector harbouring shRNA against bacterial lacZ (FG12-
silacZ) were used as a control cell line. All infected cells were enriched for GFP
expression using FACS.
Animal studies. All mouse studies were performed under the supervision of
MIT’s Division of Comparative Medicine and were done in accordance with
protocols approved by the Institutional Animal Care and Use Committee.
Athymic female nude (NCR nude, nu/nu) mice were purchased from Taconic
Laboratories and NOD/SCID mice were bred in-house. Animals were housed
under pathogen-free conditions and were given autoclaved food and water ad
libitum. For xenograft experiments, cancer cells were implanted alone, or were
mixed with MSCs or WI-38 fibroblasts and injected subcutaneously into recipi-
ent animals as previously described32. Nude mice were used at ,10–13weeks of
18–24h before injection. Female NOD/SCID mice were used at 12–14weeks of
age. Tumours were measured twice weekly using precision calipers. Tumour
volume was calculated as 4/3pr3where r is the estimated radius. Tumours were
dissected out at the end of the experiments and weighed.
CFU-F studies. Tumour xenografts were implanted in recipient NOD/SCID
females and allowed to grow for 4weeks. Tumours were then excised, treated
positive cancer cells using FACS. CFU-F culture assays were performed on the
sorted mouse stroma as standard (Stem Cell Technologies). Colonies were
entire lungs were removed, washed in PBS, and placed in ice-cold Hank’s buffer
(HBSS, Gibco). Excised lungs were immediately dissected into their various
lobes under bright-field microscopy and examined under fluorescence micro-
disseminated cancer cells. Lung metastasis burden was estimated as the number
of GFP-positive colonies observed under fluorescence microscopy. The lung
metastasis index for each mouse was calculated as the ratio of the number of
GFP-positive colonies observed in the lungs divided by the mass of the primary
tumour (in grams). Indices were pooled within each cohort and were expressed
Immunoassays. The levels of cytokines, growth factors and chemokines in the
culture media were assessed by xMAP Bio-Plex cytokine array (Bio-Rad Life
Sciences) using a Luminex 100 plate reader (Bio-Rad Life Sciences) according
natants were measured using enzyme-linked immunosorbent assay (ELISA;
implanted subcutaneously in recipient NOD/SCID mice. Four weeks later,
tumours were excised, and GFP-labelled MSCs were purified using FACS and
primers: CCL5-left, 59-TGCAGAGGATCAAGACAGCA-39, and CCL5-right,
59-GAGCACTTGCCACTGGTGTA-39. RT–PCR on cultured cells was per-
formed as standard.
Migration assays. Cancer cells were seeded in the upper well of a 24-well trans-
well Boyden chamber (8mm pore size; Costar) and migration was assessed 18h
transwell chamber and allowed to migrate towards cell-free media derived from
MCF7/Ras or MDA-MB-231 cells placed in the bottom wells. Membranes were
processed as standard. Migrating cells were stained with crystal violet and
counted using bright-field microscopy.
mice were anaesthetized and their thoracic cage opened. Texas-red lectin (to
visualize blood vessels; from Lycopersicon esculentum, Vector) was then intro-
duced into the left cardiac ventricle, followed by 4% PFA and then 20ml of cold
PBS. Frozen lung tissue was prepared and sections were processed for fluore-
scence microscopy as standard.
Anchorage-independent growth assays. We carried out soft agar assays as
Western blot analyses. Western blotting was done using standard protocols.
We used primary antibodies against phosphorylated S473-Akt (4051, Cell
Signaling), Bcl-XL (2762, Cell Signaling), Bcl-2 (2872, Cell Signaling), b-actin
(ab8224, Abcam), CCR5 (ab21653, Abcam), cleaved caspase-3 (9664, Cell
Signaling), PARP (9542, Cell Signaling), N-cadherin (205606, Calbiochem)
and vimentin (V9, NeoMarkers). We used goat antibodies to mouse (115-035-
146) and to rabbit (111-035-144) conjugated with horseradish peroxidase as
ECL (Dura, Pierce).
Immunohistochemistry. Immunohistochemical analyses were performed on
formalin-fixed, paraffin-embedded tissues. Sections (4-mm thick) were depar-
affinized, re-hydrated and subjected to antigen retrievalproceduresas described
Immunofluorescence. Cells were plated on 0.2% gelatin-coated coverslips in
complete medium overnight, washed in PBS, permeabilized in 0.1% Triton-
X100, blocked in 1% BSA/10% serum, fixed in 3.6% PBS-buffered para-
formaldehyde, and processed for indirect immunofluorescence analyses as
Anti-CCL5 treatment. MDA-MB-231 cancer cells alone or admixed with MSCs
were injected subcutaneously into recipient mice. Anti-CCL5 (32mg per mouse;
AF-278-NA, R&D) or control isotype IgG (32mg per mouse) antibodies were
injected into the peritoneum 48h after tumour implantation and twice weekly
for the duration of the experiments (10weeks).
Anoikis assays. Cancer cells were starved in 1% IFS/DME for 24h, then trypsi-
nized and suspended in serum-free media. Tubes were continuously rotated at
37uC for the duration of the experiments. Viable cells were counted using the
Trypan blue exclusion assay.
TUNEL assays. Apoptosis quantification analysis was performed using the
that the GFP-labelled MDA-MB-231/control and MDA-MB-231/CCL5 cells
were fixed for 60min in ice-cold methanol (instead of 15min in 3.6% para-
formaldehyde) to quench the GFP fluorescence.
Gene expression analysis. Expression ratios (relative to reference mRNA) and
classes of soft-tissue tumours were obtained from ref. 26 via NCBI GEO
(GSE4305). CCL5 expression was calculated as the mean of probes for
IMAGE:1325655 and IMAGE:840753. For Affymetrix gene expression analysis,
RNA extraction, cRNA synthesis and hybridization to Human Genome U133
Plus 2.0 Arrays were performed as described previously36. Raw expression data
obtained using Affymetrix GENECHIP software was normalized and analysed
using DNA-Chip Analyser (dChip) custom software (W. H. Wong and C. Li,
http://www.dChip.org/).Array probedata werenormalized tothemean expres-
sion level of each of the three CCL5 probes across stromal samples prepared
from a set of 15 normal and 7 IDC (invasive ductal carcinoma) specimens.
Leukocyte- and endothelial cell-free stroma was isolated as previously
described27. Comparisons between‘Normal’ and ‘Tumour’ (IDCbreast stroma)
were performed using the dChip ‘Compare Sample’ function. SAGE data were
obtained from http://cgap.nci.nih.gov/SAGE/AnatomicViewer and performed
as previously described27. Data were normalized, log-transformed and clustered
using average linkage uncentred analysis. Detailed purification methodologies
and sample identification numbers have been previously published27.
31. Orimo, A. et al. Stromal fibroblasts present in invasive human breast carcinomas
promote tumor growth and angiogenesis through elevated SDF-1/CXCL12
secretion. Cell 121, 335–348 (2005).
32. Elenbaas, B. et al. Human breast cancer cells generated by oncogenic
transformation of primary mammary epithelial cells. Genes Dev. 15, 50–65
33. Lodie, T. A. et al. Systematic analysis of reportedly distinct populations of
multipotent bone marrow-derived stem cells reveals a lack of distinction. Tissue
Eng. 8, 739–751 (2002).
34. Hahn, W. C. et al. Creation of human tumour cells with defined genetic elements.
Nature 400, 464–468 (1999).
35. Kuperwasser, C. et al. Reconstruction of functionally normal and malignant
human breast tissues in mice. Proc. Natl Acad. Sci. USA 101, 4966–4971 (2004).
36. Richardson, A. L. et al. X chromosomal abnormalities in basal-like human breast
cancer. Cancer Cell 9, 121–132 (2006).