Critical role for lysyl oxidase in mesenchymal stemcell-driven breast cancer malignancyChristelle P. El-Haibia, George W. Bellb, Jiangwen Zhangc, Anthony Y. Collmanna, David Woodd, Cally M. Scherbere,Eva Csizmadiaf, Odette Marianig, Cuihua Zhua, Antoine Campagnea,g, Mehmet Tonere, Sangeeta N. Bhatiad,Daniel Irimiae, Anne Vincent-Salomong, and Antoine E. Karnouba,1

aDepartment of Pathology, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, MA 02215; bWhitehead Institute for Biomedical Research,Cambridge, MA 02142; cCenter for Systems Biology, Faculty of Arts and Sciences, Harvard University, Cambridge, MA 02138; dHarvard–Massachusetts Instituteof Technology Division of Health Sciences and Technology, Koch Institute for Integrative Cancer Research, Cambridge, MA 02139; eBioMEMS Resource Center,Center for Engineering in Medicine and Surgical Services, Massachusetts General Hospital/Harvard Medical School and Shriners Hospital for Children, Boston,MA 02114; fDepartment of Medicine, Beth Israel Deaconess Medical Center, Boston, MA 02215; and gDepartment of Pathology, Institut Curie, 75248 ParisCedex 05, France

Edited* by Joan S. Brugge, Harvard Medical School, Boston, MA, and approved August 9, 2012 (received for review April 20, 2012)

Mesenchymal stem cells (MSCs) are multipotent progenitor cellswith the ability to differentiate into multiple mesoderm lineagesin the course of normal tissue homeostasis or during injury. Wehave previously shown that MSCs migrate to sites of tumorigenesis,where they become activated by cancer cells to promote metastasis.However, the molecular and phenotypic attributes of the MSC-induced metastatic state of the cancer cells remained undetermined.Here, we show that bone marrow-derived human MSCs promotede novo production of lysyl oxidase (LOX) from human breastcarcinoma cells, which is sufficient to enhance the metastasis ofotherwise weakly metastatic cancer cells to the lungs and bones.We also show that LOX is an essential component of the CD44-Twist signaling axis, in which extracellular hyaluronan causesnuclear translocation of CD44 in the cancer cells, thus triggeringLOX transcription by associating with its promoter. Processed andenzymatically active LOX, in turn, stimulates Twist transcription,which mediates the MSC-triggered epithelial-to-mesenchymaltransition (EMT) of carcinoma cells. Surprisingly, although induc-tion of EMT in breast cancer cells has been tightly associated withthe generation of cancer stem cells, we find that LOX, despite beingcritical for EMT, does not contribute to the ability of MSCs topromote the formation of cancer stem cells in the carcinoma cellpopulations. Collectively, our studies highlight a critical role forLOX in cancer metastasis and indicate that the signaling pathwayscontrolling stroma-induced EMT are distinct from pathways regu-lating the development of cancer stem cells.

Neoplastic epithelial cells within breast carcinomas are oftengreatly outnumbered by a variety of connective tissue cell

types, which collectively form the tumor-associated stroma (1).This mesenchymal microenvironment is integral for tumor ini-tiation and growth, because it regulates the survival and pro-liferation of neoplastic cells and the overall dynamics of tumordevelopment (2). Therefore, defining the nature of the signalsexchanged between the stromal niche and the cancer cells shouldprovide insights into how breast cancers develop and progressand reveal therapeutic modalities based on intercepting the tumor–stroma cross-talk.Mesenchymal stem cells (MSCs) are connective tissue pro-

genitor cells that contribute to fibrotic reactions during tissueremodeling and repair in places of wounding and inflammation(3). They reside primarily in the bone marrow, and are mobilizedto the circulation and recruited to their destination in responseto systemic signals emanating from injured tissues (4). Emulatingwounds (5), breast tumors also emit systemic signals that attractMSCs into the tumor stroma (6). Indeed, we and others haveshown previously that bone marrow-derived MSCs home to andbecome incorporated into the stroma of developing breastcarcinoma xenografts (7–9). MSCs were also found at elevatedlevels in the circulation of patients with advanced breast cancer(10) and in the stroma of human primary breast carcinomas (11),observations that are consistent with the systemic mobilization

of MSCs by and their recruitment into breast tumors in theclinical setting.The functions of MSCs in breast cancer pathogenesis, how-

ever, have not been fully elucidated, but accumulating evidenceindicates that they play prominent roles in supporting tumordevelopment (6). Indeed, bone marrow-derived MSCs within thestroma of breast cancer xenografts were shown to enhance thegrowth kinetics of the ensuing tumors and their metastasis tolungs and bones (7, 9). The abilities of MSCs to serve thesemalignant functions have now been described in multiplemodels of breast cancer, and are mediated by a number ofMSC-derived factors, such as chemokine (C-C motif) ligand 5(CCL5), IL-17B, IL-6, or IL-8 (7, 9, 11), with paracrine actions onthe neighboring breast cancer cells that cause their growth, mo-tility, invasion, and/or distant metastasis. However, in contrast tothe increasingly detailed understanding of the cross-talk oper-ating between MSCs and cancer cells and the characteristics ofcancer-associated MSCs, little is known regarding the molecularfeatures of how breast cancer cells respond to the influences ofMSCs. We, therefore, set out to determine the molecular andphenotypic attributes of MSC-activated cancer cells and explorewhether such traits contribute to breast cancer progression.

ResultsBone Marrow-Derived MSCs Stimulate de Novo Production of LysylOxidase in Breast Cancer Cells. GFP-labeled MDA-MB-231 orMCF7/Ras breast cancer cells (BCCs) were cultured alone orwith human bone marrow-derived MSCs (hereafter called MSCs)for 72 h. GFP-BCCs were then recovered using FACS, and theirRNA was processed for gene expression analysis using Affyme-trix-based arrays (Fig. S1A). Compared with cancer cells culturedalone, MSC-stimulated MDA-MB-231 (MDA-MB-231MSC) andMCF7/Ras (MCF7/RasMSC) cells exhibited significant (more thantwofold; P < 0.05) expression changes in 87 and 55 genes, re-spectively, the great majority of which was induced. However,seven of these genes were increased in both the MDA-MB-231MSC and MCF7/RasMSC cells (Fig. S1B). Pathway analysisof the up-regulated genes using gene set enrichment analyses(GSEAs) revealed a significant enrichment for multiple path-ways involved in cancer progression, predominantly in the ECMreceptor interaction gene set (Fig. S1C). In light of its criticalregulation of ECM maturation and remodeling (12) and its

Author contributions: C.P.E. and A.E.K. designed research; C.P.E., G.W.B., J.Z., A.Y.C., C.M.S.,E.C., C.Z., A.C., and A.E.K. performed research; D.W., O.M., M.T., S.N.B., D.I., and A.V.-S.contributed new reagents/analytic tools; C.P.E., G.W.B., J.Z., A.Y.C., C.M.S., E.C., C.Z., A.C.,D.I., A.V.-S., and A.E.K. analyzed data; and C.P.E. and A.E.K. wrote the paper.

The authors declare no conflict of interest.

*This Direct Submission article had a prearranged editor.1To whom correspondence should be addressed. E-mail: akarnoub@bidmc.harvard.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1206653109/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1206653109 PNAS Early Edition | 1 of 6

CELL

BIOLO

GY

ranking among the highest induced genes in the transcriptionarrays of BCCMSC, our attention focused on lysyl oxidase (LOX).LOX is a copper-dependent amine oxidase that catalyzes the

cross-linking of collagens and elastins in the ECM (13). It issecreted to the extracellular space as a 50 kDa proenzyme, thencleaved by bone morphogenetic protein-1 into an 18 kDa pro-peptide (called PP) and a 32 kDa active enzyme (called LOX)(14). LOX is the prototype for four additional proteins that sharesequence and functional similarities called LOX-like or LOXLproteins, all implicated in tumorigenesis (15). In breast cancer,LOX was described to promote cancer cell invasion (16–18) andcontribute to premetastatic niche formation (19). However, howLOX exerts its functions and whether it serves a role in directepithelial to stromal interactions have not been determined. We,therefore, proceeded to investigate the role of LOX in howcancer cells respond to activated MSCs.Quantitative assessment of LOX mRNA levels in BCCMSC

confirmed the ability of MSCs to trigger multifold induction ofLOX in the cancer cells in vitro (Fig. 1A). Similar in vivo analysesconducted on tumor xenografts containing human MSCs revealeda strong enrichment (>50-fold) for LOX mRNA in cancer cellssorted out from the dissociated tumors (Fig. 1B) as well asmarked induction of LOX protein detected by immunohisto-chemistry on the tumor sections (Fig. 1C). LOX up-regulation inthe cancer cells was not triggered by other cocultured mesen-chymal cells, such as human lung embryonic fibroblasts (WI-38),and it was weakly induced by adipose-derived MSCs (approxi-mately fourfold induction above background) (Fig. 1D). Of note,MSCs did not significantly induce the expression of any of thehighly related LOX-like genes (Fig. S2A), suggesting that theeffects of MSCs on LOX induction are specific.Importantly, LOX expression was only induced in the BCCMSC

(and not in the cocultured MSCs) (Fig. S2 B and C), and itresulted in the enrichment of the active 32 kDa mature form ofthe protein (14) (Fig. 1E). Indeed, the MSC-induced LOX wasshown to be enzymatically active in the media of the cancer cellsusing Amplite-based fluorimetric assays, an effect that wassignificantly down-regulated by the LOX inhibitor β-amino-propionitrile (20) (100 μM βAPN) (Fig. 1F). Furthermore,cancer cells expressing a LOX promoter-driven luciferase re-porter construct displayed more than sevenfold increase in lu-ciferase activity when stimulated by MSCs both in vitro (Fig.1G) and in vivo (Fig. 1H), suggesting that MSCs induced de novo

transcriptional up-regulation of LOX in carcinoma cells. Finally, inaddition to their actions on the MDA-MB-231 and MCF7/Ras,MSCs were also able to drive LOX expression in T47D, MCF7,and MDA-MB-435 human BCCs (Fig. 1I), suggesting that theinduction of LOX by MSCs in BCCs is a general phenomenon.Taken together, these observations indicate that bone marrow-derived MSCs induce robust and specific de novo expression ofLOX in BCCs, resulting in potent accumulation of the 32 kDamature form of the protein.

LOX Overexpression Promotes Breast Cancer Invasion, Metastasis, andEpithelial-to-Mesenchymal transition. Next, we examined whetherLOX overexpression in BCCs (BCCLOX) accentuated their in-vasive and metastatic traits. Indeed, forced expression of thecDNA coding for full-length LOX in BCCs (Fig. S3A) resulted inthe accumulation of the active enzyme (Fig. S3 B–D), whichcaused twofold enhancement in BCC motility in Boyden chamberassays (Fig. 2A) as well as promoted two- to fourfold increases intheir average migration velocity on collagen-coated microchannellattices (Fig. 2 B and C and Fig. S3E). In addition, LOXoverexpression enhanced invasion in wound-healing assays(Fig. S3F) and delayed anoikis (Fig. 2D). Most importantly,BCCLOX formed s.c. tumors (Fig. 2E and Fig. S4 A and B) thatwere approximately five to eight times more metastatic thancontrols (Fig. 2 E and F and Fig. S4C), with a marked pre-dilection of disseminated cells to form metastases in bone (Fig.2G). Collectively, these results indicate that elevated levels ofLOX were sufficient in enhancing the abilities of weakly meta-static cancer cells to complete the metastasis cascade.The loss of epithelial phenotypes and the acquisition of mes-

enchymal traits through the process of epithelial-to-mesenchymaltransition (EMT) have been shown to enhance the invasive andmetastatic abilities of carcinoma cells in a variety of cancer models,including breast cancer (21). To determine whether LOX over-expression caused EMT, we assessed the levels of fibronectin,α-smooth muscle actin, vimentin, and N-cadherin in BCCLOX.Indeed, LOX caused a multifold up-regulation in these mesen-chymal markers (Fig. 3 A and B) and a significant reduction in E-cadherin protein in the E-cadherin–rich MCF7/Ras cells (Fig.3B). The stimulation of EMT by LOX was consistent with itsability to notably trigger the EMT master transcriptional regu-lator Twist by >20-fold in both MDA-MB-231 and MCF7/Ras

A C

500600

AP

DH

BM3 M5

556575

M4M4

AP

DH

M3 A-M

B-2

31 Alone +MSC

MCF7/RasMDA-MB-231 MCF7/RasMDA-MB-231

05

200

300

400

LOX

mR

NA

/ G

MSC - + - +05

15M1 M2 M1 M2

LOX

mR

NA

/ G

MSC ++- - - - + ++

MD

MC

F7/R

as

5X

5X

5X

5X

p=0.001

253545

D F

80

100

120

140 p=0.0002

p=0.0001

edia

(fol

d)

250300350

/ GA

PD

H

p=0.006MCF7/RasMDA-MB-231

ECytoplasmicTotal

LOX

DH

32 kDa

MCF7/RasMDA-MB-231

APN (100 μM) - + - + - + - +MSC - - + + - - + +

0

20

40

60p=0.00004

RFU

from

m

050

100150200

LOX

mR

NA

alone +Ad-MSC+BM-MSC +WI-38

alone +Ad-MSC+BM-MSC +WI-38

GA

PH

1

36 kDa

33 kDa+MSCalone

231LOX-p-Luc +MSC

231LOX-p-Luc alone

H I

500600700800

1000900

A /

GA

PD

H

p=0.024G

fold

)

10121416

p=0.0001

05

200300400

LOX

mR

N

MSC - +T47D MCF7

- +MDA-MB-435

- +

p=0.003 p=0.001

RLU

(

0

468

2

mock alone +MSC

MDA-MB-231LOX-p-Luc

p=0.008

p=0.001

p=0.00009

p=0.0002p=0.000014

β

Fig. 1. MSCs trigger de novo LOX expression inBCCs. (A) LOX quantitative RT-PCR (RT-qPCR) onsorted GFP-BCCs cultured ± MSCs for 3 d (1:3 ratio).(B) LOX RT-qPCR on GFP-BCCs sorted from s.c.tumors grown ± MSCs for 9 wk. (C ) Immunohisto-chemistry on tumor sections in B. (D) LOX RT-qPCRon GFP-BCC cultured ± bone marrow MSCs (BM-MSCs), adipose-derived MSCs (Ad-MSCs), or WI-38fibroblasts for 3 d (1:3 ratio). (E ) Western blotanalyses on sorted BCCs cultured ± MSCs. GAPDHand histone (H1) were used as controls. (F ) LOXenzymatic activity on the media of the indicatedcultures in A ± βAPN. Data are expressed as foldchange in relative fluorescensce units (RFU) ± SEM(n = 3). (G) Luciferase assay on BCC expressing LOXpromoter-driven luciferase reporter cultured ±MSCs. Data are expressed as fold induction of rel-ative luminescent units (RLU) ± SEM (n = 3). (H)Bioluminescence imaging on MDA-MB-231 cellsharboring LOX promoter in G 2 wk after their in-jection ± MSCs (1:1 ratio) into nude mice. (I) LOXRT-qPCR on sorted GFP-BCC cultured ± MSCs for 3 d.Data in A, B, D, and I were normalized to GAPDH andare expressed as fold induction ± SEM (n ≥ 3).

2 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1206653109 El-Haibi et al.

cells (Fig. 3C), indicating that LOX regulates a critical driver ofthe EMT machinery.In line with this finding, MSCs triggered strong EMT pheno-

types in cocultured BCCs (Fig. 3A). This transition invoked morepotent up-regulation of Twist mRNA than other EMT-linkedtranscription factors (25- to 50-fold) (Fig. 3C and Fig. S5A) andwas concomitant with the localization of the Twist protein to thenuclei of BCCMSC (Fig. S5 B and C). Accordingly, we proceededto assess the extent to which LOX contributed to the MSC-elicitedTwist induction in BCCs. First, inhibition of LOX expression inthe cancer cells by ∼80% using shRNAs (Fig. S5D) abrogated theability of MSCs to trigger Twist in BCCs (Fig. 3 D and E), andeffectively inhibited MSC-induced EMT altogether (Fig. S5 Eand F). Second, inhibition of the enzymatic activity of LOX byβAPN (100 μM) crippled its ability to induce Twist in BCCLOX

(Fig. 3 F and G), neutralized the ability of MSCs to induceTwist in admixed BCCs (Fig. 3G and Fig. S5G), and com-promised the induction of EMT in BCCLOX (Fig. S5H). Theseobservations provided strong indications that LOX is a criticalmediator of MSC-induced EMT in BCCMSC and that the role ofLOX in inducing EMT depended largely on its enzymatic func-tions. Of note is that βAPN (100 μM) significantly reduced theability of MSCs to trigger LOX in the cancer cells (Fig. 3G) andthat MSCs were not able to induce LOX overexpression incancer cells devoid of Twist (Fig. S6). These results suggest thatTwist is required for LOX induction by the MSCs. How thisregulation is exerted is at present undetermined.

LOX Does Not Mediate the MSC-Induced Expansion of Cancer StemCells. The EMT program has been described to endow cancercells with certain stem cell properties thought to be conducive tometastasis (22). Indeed, BCCs prompted to undergo EMT ac-quire certain stem cell characteristics, such as the ability to formmammospheres and an increased capacity to form tumors inlimited dilution analyses (23, 24). Conversely, stem cell-like cellsderived from mammary epithelium or BCC lines exhibit EMTphenotypes (23, 24). Most importantly, cancer stem cells (CSCs)derived from human breast cancer specimens are enriched forEMT markers (23). These and other similar observations (25)suggested that the EMT and CSC programs are intertwined andthat acquisition of EMT traits by cancer cells is sufficient tobestow on them CSC characteristics. Because LOX was a critical

regulator of EMT, we asked whether it played similar roles inregulating the CSC program in BCCs. Aldehyde dehydrogenase1 (ALDH1)-positive breast cancer cells are enriched for CSCs(26); however, we found that BCCLOX exhibited no augmen-tation in their ALDH1-positive populations as determined byALDEFLUOR-based assays (Fig. S7A). Along the same lines,LOX overexpression did not provide cancer cells with any sig-nificant advantage in mammosphere-forming assays in vitro (Fig.S7 B and C), suggesting that, although LOX could promote robustEMT, it was not sufficient to promote entrance into the CSC state.Recent work showed that MSCs were able to promote a

fourfold increase in the ALDH1-positive population of SUM159BCCs cocultured with MSCs (11). In our own experiments withthe MDA-MB-231 and MCF7/Ras cells, we found that MSCscaused a multifold rise in the ALDH1 positivity of BCCMSC (Fig.S7D), a phenotype mediated by MSCs through contact-dependentmechanisms (Fig. S7 E and F). Furthermore, BCCMSC exhibitedan ∼2- to 12-fold enhancement in the primary (Fig. S7 G and H)and secondary (Fig. S7I) mammosphere-forming capacities, con-sistent with the acquisition of CSC traits. These observationsprompted us to examine whether LOX, although not sufficienton its own in driving the CSC program, was necessary forMSC-induced CSC phenotypes. Surprisingly, βAPN did not af-fect MSC-induced increases in the ALDH1 positivity of BCCMSC

(Fig. S7D) and did not affect their mammosphere-forming ac-tivities (Fig. S7 G–I). The MSC-induced CSC phenotype did notrequire LOX expression either, because MSCs were still able topromote the mammosphere-forming abilities (Fig. S7I) and en-hance the ALDH1 positivity (Fig. S7J) of cancer cells lackingsignificant LOX expression. Interestingly, however, inhibition ofLOX expression significantly compromised the ability of MSCsto enhance tumor growth and metastasis (Fig. S8), suggesting thatLOX-dependent pathways played important roles in MSC-in-duced metastasis. However, these results also suggested thatLOX-independent pathways (such as those pathways inducingCSC phenotypes) contributed substantially to MSC-induced ma-lignancy, perhaps in an equivalent fashion to those pathwaysgoverned by LOX. Collectively, these findings ascribe significantimportance to both LOX-dependent (e.g., EMT) and -in-dependent (e.g., CSC) signaling in MSC-induced metastasis andsuggest that separate stromal triggers feed into such processes.

Am

igra

tion

Bp=0.009

p=0.0003

+βAPN(100 μM)

+vehicle*

2

3 Collagen IVCollagen I

Con

trol

Collagen IVCollagen I

Fold

DC

MDA-MB-231 MCF7/Ras

CtrlLOXLOX

Ctrl CtrlLOXLOX

Ctrl

( μ )

0

1

** * *

G

LOX

MDA-MB-231 MCF7/Ras

ble

cells

(x10

E4)

4

6

8MDA-MB-231 controlMDA-MB-231LOX

MCF7/Ras controlMCF7/RasLOX

p=0.002

MCF7/Ras-Control

Ab controlMCF7/Ras-Control

MCF7/RasLOX

4X4X

e V

eloc

ity (μ

m/m

in)

406080

100120

Collagen I Collagen IV

p=4.075E-13

p=2.88E-07

p=1.15E-23

p=3.74E-08Collagen I Collagen IV

Via

2

Days 1 2 3 4 5

MCF7/Ras-ControlMCF7/RasLOXF 4X4X

20X20X

Lung

E0.3

p=0.138

810

p=0.035

mas

s (g

)A

vera

g

020

MDA-MB-231

Ctrl LOX Ctrl LOXCtrl LOX Ctrl LOX

MCF7/Ras

αGFP-TumorαGFP-Bone

metastasis (fold)

0

0.1

0.2

Control LOX Control LOX0246

Ave

rage

tum

or

MCF7/Ras

MCF7/RasLOX MCF7/RasLOX

p=0.005

Fig. 2. LOX is sufficient in promoting cellular mo-tility, anoikis resistance, and metastasis. (A) Trans-well migration assays in which control cells (Ctrl) orBCCLOX (LOX) were allowed to migrate overnight ±βAPN. Results are represented as fold change ± SEM(n = 3). (B) Representative 10× images from moviesof controls or BCCLOX migrating on collagen-coatedmicrochannels. (C) Average velocity of cells (in μMper min) in Bwere calculated using ImageJ (NationalInstitutes of Health). Data are representative of 50cells/frame. (D) Control cells and BCCLOX were sus-pended in culture media with constant rotation, andcell viability assessed using trypan blue exclusionstaining. (E) Average weight (mean ± SEM ) of GFP-MCF7/RasLOX and control tumors in nude mice (n > 7for each group). GFP-positive lung metastatic nod-ules were enumerated under fluorescence micros-copy and are represented as fold change ± SEM. (F)Representative pictures of GFP-positive metastaticnodules in the lungs of mice in E. (G) Representativeanti-GFP–stained sections of primary tumors and ribsof mice bearing tumors in E. Although ∼50% ofmice bearing MCF7/RasLOX tumors developed bonemetastases, none of the animals bearing the MCF/Ras control tumors developed any. MCF7/RasLOX-bearing mice scoring positive for bone metastasis ex-hibited approximately three colonies per rib section.

El-Haibi et al. PNAS Early Edition | 3 of 6

CELL

BIOLO

GY

Hyaluronan–CD44 Interactions Mediate MSC-Induced LOX and TwistExpression. We focused on determining the nature of the MSC-triggered upstream signaling pathways regulating LOX expres-sion in the cancer cells. Previous reports described the regulationof LOX by hypoxia (18) or mechanical transduction (27), but wefound that neither of these mechanisms mediated MSC-triggeredLOX overexpression in our BCCs (SI Results and Fig. S9 A–E).Furthermore, we found that the ability of MSCs to induce LOXin BCCs was not mediated by soluble factors (SI Results and Fig.S9 E–I) but resided instead in the ECM of the MSCs. Indeed,culture of BCCs on ECM deposited by MSC monolayers (Fig.S9J) resulted in a significant up-regulation of LOX in the cancercells (Fig. S9K). These data indicate that the observed heterotypicsignaling responsible for the induction of LOX in the cancer cellsoriginates from the ECM of MSCs.Influenced by these results, our attention focused particularly

on the glycosaminoglycan hyaluronan (HA) as being a potentialtrigger for LOX in BCCMSC for a number of reasons. First, HA/hyaluronic acid is one of the most abundant proteoglycan pres-ent in the ECM of mammalian cells, particularly the ECM ofMSCs (28, 29). Second, GSEA strongly highlighted the ECM–receptor interaction pathway in BCCMSC (Fig. S1), in which HA issignificantly represented. Third and most importantly, HA wasshown to be a critical modulator of EMT phenotypes in normaland cancerous epithelial cells, and its abundance in the stromalenvironment of cancer cells causes their invasion and metastasis(30). Indeed, treatment of cocultures of MSCs and cancer cellswith hyaluronidase inhibited MSC-induced LOX (and Twist)expression in BCCMSC (Fig. 4A), suggesting a necessary role forHA in regulating LOX induction. To investigate whether HA wassufficient in inducing LOX in the cancer cells, we expandedMDA-MB-231 cells on culture surfaces coated with HA mole-cules grouped in different sizes from 4–8 to >950 kDa. In-terestingly, only the high-molecular weight HA substratumwas able to cause the up-regulation of LOX transcription(∼120-fold), and it did so to levels comparable with the levelstriggered by admixed MSCs (Fig. 4B), a phenotype that paral-leled the activation of the LOX promoter by HA (Fig. 4C).These observations indicated that HA was sufficient and nec-essary for the ability of MSCs to trigger LOX expression in theadmixed cancer cells.

HA acts through multiple receptors, with CD44 being its majorpartner in cancer (30). We, therefore, proceeded to explorewhether it mediated the actions of MSCs on the LOX–Twist axis.Indeed, inhibition of CD44 expression in the cancer cells (Fig. 4D,Inset) abolished the ability of MSCs to trigger LOX expressionin the MDA-MB-231MSC (Fig. 4D). Furthermore, expression ofCD44 shRNAs in the cancer cells impaired MSC-induced Twistexpression (Fig. 4D), further corroborating the link betweenLOX and Twist and placing CD44 as a critical upstream moleculethat mediates the actions of MSCs on the LOX–Twist pathway inthe cancer cells.How HA–CD44 interactions caused de novo transcription of

LOX, however, was undetermined. Previous work by Tammiet al. (31) indicated that the association of HA with CD44 causesinternalization of CD44 fragments, and some translocate to thenucleus (32) and modulate gene transcription by direct associationand activation of gene promoters (33). To investigate whethersuch a mechanism operated in our cocultures, we probed for thelocalization of CD44 in MDA-MB-231 and MDA-MB-231MSC.Strikingly, Western blot and immunofluorescence analyses in-dicated that the interaction of MSCs with cancer cells caused amarked increase in CD44 localization to the nucleus of the cancercells (Fig. 4 E and F). Particularly, we observed a preferentialenrichment (more than fivefold) of an ∼50-kDa fragment of CD44in the nucleus of the MSC-stimulated cancer cells, consistent witha potential role for this fragment in modulating gene transcription.These results suggested that CD44 might play a more proximal rolein modulating LOX transcription by affecting its promoter activity.To test this possibility, we conducted ChIP on the nuclear prepa-rations derived from MSC-stimulated cancer cells or their controlcounterparts using a polyclonal anti-CD44 or isotype controlantibodies followed by PCR using primers specific to the LOXpromoter or control human satellite 2 (SAT2). We found thatCD44 associated specifically with the LOX promoter, a couplingthat seems to be markedly accentuated by admixed MSCs (Fig.4G). Collectively, our results are consistent with a model in whichthe interaction of MSCs with cancer cells causes translocation ofCD44 to the nucleus, which associates with and activates the LOXpromoter, leading to Twist-dependent induction of EMT pheno-types conducive to cancer metastasis (Fig. 5H).

Fibronectin

Ctrl LOX Ctrl LOXBA ED

Control shLOX-2p=0.001

250300

p=0.005

BCCMSCBCC

BCCWI38

ctin 20

100120

p=0.009 p=0.02 p=0.004

BCCBCCMSC

BCC-shLOX-2shLOX MSC

SMA

Vimentin

N-cadherin

E-cadherin

Twist

Actin

MSC - + - +0

1020304050

100150200

LOX Twist

p=0.003

C l LOX βAPN C l LOX LOX βAPN

p=0.002

CtrlLOX

FN m

RN

A /

A

p=0.0020

10

2530

020406080

70

Act

in

p=0.001

5060

p=0.009

BCC- -2

Actin

MDA-MB-231 MCF7/RasF

Twist

Actin

Control LOX +

MDA-MB-231 MCF7/Ras

28 kDa

56 kDa

Contro +

Cp=0.005

05

101520

010203040

200

2505060

SM

A m

RN

A /

ctin

p=0.0008p=0.022

p=0.0097080

BCCMSCBCC

BCCWI38

CtrlLOX G

RN

A /

Act

in

150200250300350

LOXTwist

100

150

200

250

p=0.001

p=0.0009

ctin p=0.01p=0.001

0

50

100

150

010203040

20

Vim

mR

NA

/ A

p=0.0001 p=0.11

p=0.14

p 0 083

18 mR

NA

/ A

ctin

0102030405060

MDA-MB-231

p=0.002

=0.0027080

p=0.0004p=0.0005

mR

NA

/ A

ctin

m

100

050

0

50

saR/7FCM132-BM-ADM

βAPNAlone +MSC

- + - +Alone +MSC

- + - +

MDA-MB-231 MCF7/Ras

N-c

ad m

RN

A /

A

4

8

12

16

0

p=0.0001= .

48

12

16

0

Twis

t p

MCF7/Ras

p=0.008

0102030405060

Fig. 3. LOX is essential for MSC-induced EMT of BCCs. (A) RT-qPCR for the indicated genes conducted on sorted GFP-BCC cultured alone, with MSCs or WI-38cells, or on BCCLOX or their controls. (B) Western blots on BCC controls or BCCLOX. (C) Twist RT-qPCR on samples in A. (D) RT-qPCR for LOX and Twist on MDA-MB-231 controls or MDA-MB-231-shLOX-2 cultured ± MSCs. (E) Western blots on MDA-MB-231 controls or MDA-MB-231-shLOX-2 cells cultured ± MSCs. (F)Western blots on BCC controls or BCCLOX ± βAPN (100 μM). (G) RT-qPCR for LOX and Twist on BCCs cultured ± MSCs ± βAPN (100 μM) for 3 d. Data in A, C, D,and G were normalized to β-actin and are expressed as fold induction ± SEM (n ≥ 3).

4 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1206653109 El-Haibi et al.

DiscussionOur findings uncover a pivotal role for LOX in regulating theMSC-induced EMT phenotypes of cancer cells and reveal apreviously undescribed functional link between LOX and Twist.This link extends to the clinical setting, because the expressionlevels of these two genes statistically correlated across a largegroup (n = 2,158) of human cancers (Fig. 5A), including thesubgroup (n= 333) of breast cancers within this cohort (Fig. 5B).Twist and LOX also significantly correlated across three addi-tional and commonly used sets of breast cancer array databases(Fig. 5C). Furthermore, subsets of breast cancer patients withtumors that displayed elevated expression levels of both genesalso exhibited lower chances of survival (Fig. 5D). Finally, subsetsof particularly aggressive triple negative breast tumors—called

metaplastic breast cancers–previously reported to be enrichedfor EMT markers (34, 35) exhibited preponderant expressionsof both Twist and CD44 (Fig. 5E) and displayed ∼7- to 60-foldupregulation of LOX mRNA compared with controls (Fig. 5F).These results are consistent with the premise of our experi-mental findings, and they suggest that concerted assessment ofLOX/Twist levels bears a prognostic value in human breast cancers.On the mechanistic level, this report describes a direct role for

CD44 in the transcriptional regulation of LOX and suggests analternative mechanism of LOX induction that is not mediated byhypoxia-inducible factor 1α (HIF1α) and does not seem to in-volve tensegrity. Considering that the latter two stimuli onlytrigger small increases in LOX transcription in the cancer cells(27, 36) in contrast to the multifold increases that we observedwith MSCs, we posit that the HA-CD44 trigger might be a moreefficient means to regulate LOX expression in vivo. That said,how CD44 becomes internalized as a result of BCC:MSC con-tact and the specific CD44 protein sequences that couple to theLOX promoter have not been determined and are currently underinvestigation.On the cellular level, LOX expression endowed cells with

multiple phenotypes of malignancy, including enhanced abilitiesto resist anoikis, increased motility, and enhanced ability tometastasize to mouse lungs and bones (Fig. 2 and Figs. S3 E andF and S4). These results, together with the observation that LOXwas potently induced by MSCs in different BCC lines (Fig. 1I),suggest that LOX represents a conduit for the generalized re-sponse of cancer cells to the prometastatic signals of stromalMSCs. In this respect and in light of the fact that most (but not all)of the tested prometastatic actions of MSCs were sensitive to theLOX inhibitor BAPN or shLox, we posit that anti-LOX–basedtherapeutic approaches may be effective in combating metastaticdisease, not only by targeting premetastatic niche formation (19)but also by targeting cancer cells.Previous work indicated that induction of EMT caused in-

duction of CSC phenotypes, and conversely, CSCs were found tobe enriched in EMT markers (22–24, 37). These observationsindicated that EMT and CSC phenotypes are tightly regulated bythe same framework and suggested that the signal transductionpathways that govern these two programs are intimately inter-twined. However, we found that, although LOX induction wassufficient to cause Twist up-regulation (and EMT) to levelscomparable with those levels observed in BCCMSC, it was notsufficient in driving the CSC phenotype (Fig. S7). Taken together,these data argue that the upstream stroma-initiated signalingpathways governing the propagation of CSCs may not be iden-tical to those pathways fostering EMT.Emerging concepts in cancer metastasis emphasize the role of

the tumor stroma in regulating the metastatic cascade and suggestthat cancer cells are highly responsive to the contextual signals ofnearby stromal cells (2). In this context, the induction of EMT andCSC phenotypes in the cancer cells by stromal MSCs is consistentwith the transient nature of such contextual signals, because bothEMT and CSC programs may need to be reversed to allow for thecompletion of the metastatic cascade (22). Indeed, BCCMSC

exhibited a gradual decrease in their LOX content over time (Fig.S10), which corroborates this hypothesis and presents LOX-eli-cited signaling as a prominent molecular feature of this transientstate. Because the protumorigenic and prometastatic functions ofstromal MSCs have now been recognized in multiple tumor set-tings (7, 38), it would be important to exploit such a platform tofurther characterize the molecular nature of the heterotypic cross-talk that takes place at the tumor to stroma interface.

Materials and MethodsCell lines and culture conditions, Affymetrix arrays and GSEA analyses, matrixpreparations, migration assays, constructs, reagents, primer sequences, LOXenzyme assays, ALDEFLUOR assays, mammosphere assays, immunohisto-chemical methods, animal work, ChIP, and clinical analyses are discussed in SIMaterials and Methods.

BA

60

80

100

120

/ 18S

rRN

Ap=0.0003 p=0.0006

p=0.005

140

p=0.012p=0.024 p=0.0003

p=0.001

p=0.005

p=0.0004

80100120140160

A /

18S

rRN

A HA

- - + + - - + +- + - + - + - +

0

20

40

mR

NA

Hyaluronidase

(150U/ml)

MSC

MDA-MB-231 MCF7/Ras

Medium

0

20406080

LOX

mR

NA

Ctrl+MSCU. low

Low HighCtrl

MDA-MB-231

C Dm

RN

A/ A

ctin

50

100

150

200

250 p=0.0001

p=0.0125

Act

inC

D44

LOXTWISTSMA

500

1000

1500

2000

2500

3000

3500

old

chan

ge in

RLU

HAp=0.009

HFMDA-MB-231 MDA-MB-231MSC

FITC

EMSC - + - +

CD44(50kDa)

CD44(85kDa)

MSC

m

0

- + - + - +shLuc shCD44-2 shCD44-3

AMSC HA

BCC

0500

Fo

Ctrl+MSCLow

MediumHigh

mediaMDA-MB-231LOX-p-Luc

DAPI

(50kDa)

Actin

H1

(85kDa)

Nuclear CytoplasmicG

MW Ctrl +MSC Ctrl +MSC +MSC

LOX

LOX

BCC CD44

Merged

αCD44 Isotype SAT2

Twist

EMT CSC

?

Twist

?

p=0.002

TWISTLOX

Fig. 4. MSCs induce LOX in cancer cells through HA and CD44. (A) RT-qPCRon sorted GFP-BCC cultured ± MSCs ± hyaluronidase for 72 h. (B) RT-qPCR onBCCs cultured on ultra low- (U. low; 4–8 kDa), low- (15–40 kDa), medium-(75–350 kDa), or high- (>950 kDa) molecular weight HA substrata for 72 h.BCCs cultured on plastic, alone (Ctrl), or with MSCs (+MSC) served as controls.(C) Luciferase assay on BCCs expressing LOX promoter-driven luciferase re-porter cultured as in B for 72 h. Data are expressed as fold induction ofrelative luminescence units (RLU) ± SEM of triplicates (n = 3). (D, Inset)Western blots on MDA-MB-231 cells harboring shLuc control, shCD44-2, orshCD44-3. RT-qPCR for the indicated genes on cells in Inset are in the maingraph. Data in A, B, and D are expressed as fold induction ± SEM of tripli-cates. (E) Western blots of nuclear and cytoplasmic extracts on MDA-MB-231or MDA-MB-231MSC. (F) Anti–CD44-FITC immunofluorescence staining onMDA-MB-231 or MDA-MB-231MSC cells. DAPI was used for nuclear staining.(G) ChIP of MDA-MB-231 (Ctrl) or MDA-MB-231MSC (+MSC) cells using poly-clonal anti-CD44 or isotype control antibodies. Primers for LOX promoterand the control SAT2 were used in the PCR. (H) The interaction of MSCs withBCCs causes the localization of CD44 to the nucleus, where it associates withthe LOX promoter, driving LOX transcription. LOX then promotes Twisttranscription. Although the CD44–LOX pathway is necessary and sufficientfor MSC-induced EMT, it is neither sufficient nor required for the CSC phe-notype. We hypothesize that additional MSC-triggered events are needed toenforce CSC traits.

El-Haibi et al. PNAS Early Edition | 5 of 6

CELL

BIOLO

GY

ACKNOWLEDGMENTS. We thank T. Chavarria, F. Reinhardt, and S. Malstromfor assistance in animal studies; D. Louvard for helpful discussions; andR. Weinberg for critical reading of the manuscript. BM-MSCs were providedby the Texas A&M Health Science Center through National Institutes of HealthGrant P40RR017447. This work was supported by the Packard Foundation

(S.N.B.), National Institutes of Health Grant R21CA135601 (D.I.), Beth IsraelDeaconess Medical Center (A.E.K.), and the Sidney Kimmel Cancer ResearchFoundation (A.E.K.). A.E.K. is a Kimmel Scholar, a recipient of a Career De-velopment Award from the Prostate and Breast Cancer Program of BIDMC,and of a Career Catalyst Research Award from Susan G. Komen for the Cure.

1. Bissell MJ, Radisky D (2001) Putting tumours in context. Nat Rev Cancer 1:46–54.2. Polyak K, Kalluri R (2010) The role of the microenvironment in mammary gland de-

velopment and cancer. Cold Spring Harb Perspect Biol 2:a003244.3. Bianco P, Robey PG, Simmons PJ (2008) Mesenchymal stem cells: Revisiting history,

concepts, and assays. Cell Stem Cell 2:313–319.4. Karp JM, Leng Teo GS (2009) Mesenchymal stem cell homing: The devil is in the de-

tails. Cell Stem Cell 4:206–216.5. Dvorak HF (1986) Tumors: Wounds that do not heal. Similarities between tumor

stroma generation and wound healing. N Engl J Med 315:1650–1659.6. El-Haibi CP, Karnoub AE (2010) Mesenchymal stem cells in the pathogenesis and

therapy of breast cancer. J Mammary Gland Biol Neoplasia 15:399–409.7. Karnoub AE, et al. (2007) Mesenchymal stem cells within tumour stroma promote

breast cancer metastasis. Nature 449:557–563.8. Kidd S, et al. (2009) Direct evidence of mesenchymal stem cell tropism for tumor and

wounding microenvironments using in vivo bioluminescent imaging. Stem Cells 27:2614–2623.

9. Goldstein RH, Reagan MR, Anderson K, Kaplan DL, Rosenblatt M (2010) Human bonemarrow-derived MSCs can home to orthotopic breast cancer tumors and promotebone metastasis. Cancer Res 70:10044–10050.

10. Roodhart JM, et al. (2011) Mesenchymal stem cells induce resistance to chemotherapythrough the release of platinum-induced fatty acids. Cancer Cell 20:370–383.

11. Liu S, et al. (2011) Breast cancer stem cells are regulated by mesenchymal stem cellsthrough cytokine networks. Cancer Res 71:614–624.

12. Kagan HM, Li W (2003) Lysyl oxidase: Properties, specificity, and biological roles insideand outside of the cell. J Cell Biochem 88:660–672.

13. Csiszar K (2001) Lysyl oxidases: A novel multifunctional amine oxidase family. ProgNucleic Acid Res Mol Biol 70:1–32.

14. Lucero HA, Kagan HM (2006) Lysyl oxidase: An oxidative enzyme and effector of cellfunction. Cell Mol Life Sci 63:2304–2316.

15. Mäki JM (2009) Lysyl oxidases in mammalian development and certain pathologicalconditions. Histol Histopathol 24:651–660.

16. Kirschmann DA, et al. (2002) A molecular role for lysyl oxidase in breast cancer in-vasion. Cancer Res 62:4478–4483.

17. Payne SL, et al. (2005) Lysyl oxidase regulates breast cancer cell migration and adhesionthrough a hydrogen peroxide-mediated mechanism. Cancer Res 65:11429–11436.

18. Erler JT, et al. (2006) Lysyl oxidase is essential for hypoxia-induced metastasis. Nature440:1222–1226.

19. Erler JT, et al. (2009) Hypoxia-induced lysyl oxidase is a critical mediator of bonemarrow cell recruitment to form the premetastatic niche. Cancer Cell 15:35–44.

20. Wilmarth KR, Froines JR (1992) In vitro and in vivo inhibition of lysyl oxidase byaminopropionitriles. J Toxicol Environ Health 37:411–423.

21. Yang J, Weinberg RA (2008) Epithelial-mesenchymal transition: At the crossroads ofdevelopment and tumor metastasis. Dev Cell 14:818–829.

22. Chaffer CL, Weinberg RA (2011) A perspective on cancer cell metastasis. Science 331:1559–1564.

23. Mani SA, et al. (2008) The epithelial-mesenchymal transition generates cells withproperties of stem cells. Cell 133:704–715.

24. Morel AP, et al. (2008) Generation of breast cancer stem cells through epithelial-mesenchymal transition. PLoS One 3:e2888.

25. Creighton CJ, Chang JC, Rosen JM (2010) Epithelial-mesenchymal transition (EMT) intumor-initiating cells and its clinical implications in breast cancer. J Mammary GlandBiol Neoplasia 15:253–260.

26. Ginestier C, et al. (2007) ALDH1 is a marker of normal and malignant human mam-mary stem cells and a predictor of poor clinical outcome. Cell Stem Cell 1:555–567.

27. Levental KR, et al. (2009) Matrix crosslinking forces tumor progression by enhancingintegrin signaling. Cell 139:891–906.

28. Park HW, Shin JS, Kim CW (2007) Proteome of mesenchymal stem cells. Proteomics 7:2881–2894.

29. Astachov L, Vago R, Aviv M, Nevo Z (2011) Hyaluronan and mesenchymal stem cells:From germ layer to cartilage and bone. Front Biosci 16:261–276.

30. Toole BP (2009) Hyaluronan-CD44 Interactions in Cancer: Paradoxes and Possibilities.Clin Cancer Res 15:7462–7468.

31. Tammi R, et al. (2001) Hyaluronan enters keratinocytes by a novel endocytic route forcatabolism. J Biol Chem 276:35111–35122.

32. Janiszewska M, De Vito C, Le Bitoux MA, Fusco C, Stamenkovic I (2010) Transportinregulates nuclear import of CD44. J Biol Chem 285:30548–30557.

33. Su YJ, Lai HM, Chang YW, Chen GY, Lee JL (2011) Direct reprogramming of stem cellproperties in colon cancer cells by CD44. EMBO J 30:3186–3199.

34. Lien HC, et al. (2007) Molecular signatures of metaplastic carcinoma of the breast bylarge-scale transcriptional profiling: Identification of genes potentially related toepithelial-mesenchymal transition. Oncogene 26:7859–7871.

35. Hennessy BT, et al. (2009) Characterization of a naturally occurring breast cancersubset enriched in epithelial-to-mesenchymal transition and stem cell characteristics.Cancer Res 69:4116–4124.

36. Pez F, et al. (2011) The HIF-1-inducible lysyl oxidase activates HIF-1 via the Aktpathway in a positive regulation loop and synergizes with HIF-1 in promoting tumorcell growth. Cancer Res 71:1647–1657.

37. Blick T, et al. (2010) Epithelial mesenchymal transition traits in human breast cancercell lines parallel the CD44(hi/)CD24 (lo/-) stem cell phenotype in human breast cancer.J Mammary Gland Biol Neoplasia 15:235–252.

38. McLean K, et al. (2011) Human ovarian carcinoma–associated mesenchymal stem cellsregulate cancer stem cells and tumorigenesis via altered BMP production. J Clin Invest121:3206–3219.

A B C

D

F

E

Fig. 5. The LOX–Twist axis is a powerful prognostic indicator in breast cancer. Correlation analyses of the expression intensity of Twist and LOX in GSE2109(adjustedR2 = 0.12, P = 2.2E-16) (A), in the breast cancer samples inA (adjustedR2 = 0.15, P = 2.875E-13) (B), and in three cohorts of breast cancers GSE1456 (R2 = 0.38,P = 2E-10), GSE4922 (R2 = 0.17, P = 4E-13), and GSE2034 (R2 = 0.17, P = 4E-13) (C). (D) Kaplan–Meier patient survival curves for the indicated breast cancer pools inGSE2034 (p = 0.037). (E) CD44 and Twist staining of metaplastic breast cancer tissues. (F) LOX mRNA expression in clinical metaplastic breast cancer tissues. 1–6,Normal matched controls; 7, epidermoid; 8, squamous; 9, glandular; 10, chondroid; 11–20, fusiform.

6 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1206653109 El-Haibi et al.

Supporting InformationEl-Haibi et al. 10.1073/pnas.1206653109SI Materials and MethodsCell Lines and Cell Cultures. Breast cancer cell (BCC) lines MDA-MB-231,MCF7,MCF7/Ras, T47D, andMDA-MB-435 and normalhuman embryonic lung fibroblasts (WI-38) or HEK-293T cellswere cultured inDMEMsupplementedwith 10%FBS, 100 units/mLpenicillin, and 100 μg/mL streptomycin. Bone marrow-derivedhuman mesenchymal stem cells (BM-MSCs) were obtained fromthe Institute for Regenerative Medicine at Scott and White,Texas A&M Health Science Center (Temple, TX). These cellsoriginate from bone marrow aspirates of healthy donors, andthey are purified first by density centrifugation and then by ex-pansive culture conditions conducive to the growth of MSCs.Before release to users, these cells are analyzed over threepassages for CFUs, cell growth, differentiation potential into fat,bone, and chondrocytes, and various cell marker expression byflow cytometry (negative for CD36, CD34, CD19, CD11b, CD45,CD49b, CD3, and CD184; positive for CD166, CD90, CD105,CD147, CD49c, and CD29). Adipose-derived MSCs (Ad-MSCs)were purchased from ScienceCell Research Laboratories (7510).The cells were isolated from human adipose tissue and charac-terized as being uniform for CD73, CD90, and CD105 and lipidstaining after differentiation tests. BM-MSCs and Ad-MSCs werecultured in α-MEM supplemented with 16.5% FBS, 1% gluta-mine, 100 units/mL penicillin, and 100 μg/mL streptomycin. Allcells were maintained at 37 °C under 5% CO2.

Affymetrix Arrays and Gene Set Enrichment Analyses. One micro-gram total RNA was used to prepare cDNA using the GenechipHT One-Cycle cDNA Synthesis Kit (900687; Affymetrix) and theGeneChip HT IVT Labeling Kit (900688; Affymetrix). cRNAtargets were cleaned up, fragmented, and hybridized to Affy-metrix HT-HG U133 2.0 Plus Arrays (900751; Affymetrix). Thehybridization and subsequent washing and staining were per-formed on the Affymetrix GeneChip Array Station automationplatform. Samples that present with the range of 40–60% ofgenes and GAPDH and β-actin ratios less than three passedand were incorporated in the analyses. Gene set enrichmentanalyses (Broad Institute of Harvard and Massachusetts Instituteof Technology; www.broadinstitute.org/gsea) were performedon the genes found up-regulated in the array sets of BCCMSC

vs. BCC alone.

Direct Coculture, Sorting, and Long-Term Reculture. BM-MSCs, Ad-MSCs, or WI-38 cells were seeded at a density of 3 × 106 cells ina 15-cm tissue culture plate and allowed to adhere overnight.GFP-expressing BCCs were then overlaid onto the monolayersat a density of 1 × 106 cells. All cell types were cultured in-dividually in parallel as controls. Cells were maintained in MSC-specific media and incubated for 3 d. Based on the experiment,cells were treated with β-aminopropionitrile (βAPN; 100 μM,A3134; Sigma Aldrich), hypoxia-inducible factor 1α (HIF1α)inhibitor LW6 (10 μM; EMD Chemicals), Rho-associated kinase(ROCK) inhibitor (10 μM, Y27632; Calbiochem), or hyaluroni-dase (150 U/mL; EMD Chemicals) for the duration of the incu-bations. After 3 d incubation, cocultured cells were trypsinized,filtered to obtain single-cell suspensions, and separated by flowcytometry. GFP-expressing BCCs were recovered, and the sam-ples were divided into portions processed for RNA and/or pro-tein extraction using PARIS kit (Ambion); where applicable, theywere recultured for additional analysis.

Indirect Coculture. BM-MSCs were cultured in α-MEM with2% FBS, 1% glutamine, 100 units/mL penicillin, and 100 μg/mLstreptomycin for 3 d. Media were then collected and transferredto BCCs to determine the effects of MSC-secreted soluble fac-tors on BCC gene expression. After culture in MSC-conditionedmedia for 3 d, BCCs were lysed, and RNA was extracted andused in quantitative RT-PCR (RT-qPCR) gene expressionanalyses. BCCs grown in α-MEM media served as controls. Al-ternatively, MSCs and BCCs were grown in a Boyden chamber,where the two populations were separated by a 0.4-μm porousmembrane that allows for soluble factor exchange. MSCs werecultured alone or together with BCCs in the top insert in α-MEMwith 2% FBS, 1% glutamine, 100 units/mL penicillin, and 100 μg/mL streptomycin, whereas BCCs were cultured in the bottomwell for 3 d. After incubation, BCCs seeded in the bottomchamber were lysed, and RNA was extracted and used in RT-qPCR gene expression analyses.

Conditioned ECM Assay. BM-MSCs were cultured in α-MEM with2% FBS, 1% glutamine, 100 units/mL penicillin, and 100 μg/mLstreptomycin for 3 d with or without hyaluronidase (150 U/mL)treatment. Cells were then washed three times with PBS andincubated with 25 mM EDTA in PBS (pH 7.4) at 4 °C overnightto allow for gentle cell detachment with minimal effect on theMSC-deposited matrix. After cell detachment, the solution wasaspirated, and the plate was washed repeatedly with PBS. BCCswere then seeded onto that plate and incubated for 3 d in MSCmedia. After incubation, BCCs were lysed, and RNA was ex-tracted to determine the effect of MSC ECM proteins on BCCgene expression. For preparation of the hyaluronic acid-coatedculture surfaces, hyaluronan produced by microbial fermentationof Streptococcus pyogenes was purchased from R&D Systems atdifferent molecular masses [GLR003, ultra low (4–8 kDa);GLR001, low (15–40 kDa); GLR004, medium (75–350 kDa);GLR002, high (>950 kDa)] and reconstituted in water as per themanufacturer’s recommendations. Products had >99% purityand were used at a concentration of 100 μg/mL as described inthe work by Martchenko et al. (1). Tissue culture plates werepreincubated with 100 μg/mL hyaluronic acid overnight at 4 °C.The next day, 1 × 106 MDA-MB-231 cells were overlaid on topof the hyaluronic acid layer and incubated for 3 d at 37 °C. Afterincubation, cells were lysed, and their RNA was extracted andused in RT-qPCR analysis.

Micromechanical Reconfigurable Culture (Combs). This system con-sists of micromachined silicon substrates suitable for tissue cul-ture. It has two parts that can be repositioned to allow for dynamicmanipulation of cell–cell interactions. Using tweezers, the deviceparts can be locked in two different configurations: gap, wherecell populations are separated by a narrow 80-μm distance, orcontact, where direct intimate contact is allowed between twocell populations. MSCs and BCCs were seeded onto combs ata density of 500,000 cells/mL in a 12-well plate and the cell-coated combs were incubated in gap or contact mode for 3 d.After incubation, cells were scraped, lysed, and RNA-extractedfor gene expression analysis by RT-qPCR.

Hypoxia Cultures. To generate a positive control for HIF1αWestern blot experiments and validate the effectiveness of theHIF1α chemical inhibitor (LW6; 10 μM; Calbiochem), cells wereplaced in a reduced oxygen chamber for 2 d and immediatelylysed for RNA and protein isolation followed by RT-qPCR and

El-Haibi et al. www.pnas.org/cgi/content/short/1206653109 1 of 12

Western blot analysis of HIF1α levels (610958; BD Trans-duction). Hypoxic conditions were achieved using INVIVO2 400from Ruskinn Technology.

Rho Kinase (ROCK) Inhibition and Focal Adhesion Kinase (FAK)Immunoblot. The effectiveness of the ROCK chemical inhibitor(Y27632, 10 μM; EMD Chemicals) was validated by Westernblot analysis of p-FAK levels in treated and untreated cells (p-FAK-Tyr397, 3283; Cell Signaling).

pBabe-puro-LOX Construct. pCR4-TOPO-LOX was purchasedfrom Open Biosystems (MHS1010-98052123). Primers for PCRamplification were purchased from Eurofins MWG Operon andincluded BamHI/EcoRI restriction sites. Primer sequences for full-length human LOX are hLOX-5PRIME-BamHI-CGCGGAT-CCATGCGCTTCGCCTGGACCGTG and hLOX-3PRIME-EcoRI-CCGGAATTCCTAATACGGTGAAATTGTGC. PCRproducts were subcloned into pBabe-puro vector and verifiedby sequencing. HEK-293T cells were transfected with pBabe-LOX using retroviral packaging and Fugene HD (Roche). Theviral supernatants were collected at 48 h, passed through a 0.45-μm filter, and used to infect BCCs (MDA-MB-231 and MCF7/Ras) in the presence of polybrene (10 μg/mL). Stable cell lineswere generated after puromycin selection (2 μg/mL). LOXmRNAexpression and protein levels were verified by RT-qPCR andWestern blot analysis, respectively. LOX promoter-driven lu-ciferase reporter was purchased from SwitchGear Genomics(S703087).

shRNA Constructs. LOX and Twist shRNA constructs were obtainedfrom the Dana-Farber Cancer Institute (Boston, MA) through theRNAi Consortium. The CD44 shRNA constructs were obtainedfrom S. Godar (University of Cincinnati Cancer Institute, Cin-cinnati, OH) (2). HEK-293T cells were transfected with shRNAplasmids using viral packaging and Fugene HD (Roche). The viralsupernatants were collected at 48 h, and BCCs were infected in thepresence of polybrene (10 μg/mL). Stable cell lines were generatedafter puromycin selection (2 μg/mL), and gene silencing was ver-ified by RT-qPCR and Western blot analysis.

Amplite LOX Enzymatic Activity Assay.BCC lysates and conditionedmedia from direct cocultures with BM-MSCs or BCCLOX andcontrol cells were collected to measure H2O2 production in-dicative of LOX enzymatic activity using the Amplite Fluori-metric Lysyl Oxidase Assay Kit (AAT; Bioquest) according to themanufacturer’s instructions. βAPN (100 μM) control was addedto ensure assay specificity.

Transwell Migration Assays. Transwell migration assays were per-formed as standard. Briefly, 50,000 BCCs were seeded in the topchamber of a Boyden chamber apparatus and allowed to migrateovernight across 8-μm porous membranes toward media with10% FBS placed in the bottom chamber. The membranes weresubsequently fixed with 4% paraformaldehyde (PFA). The innerpart of the membrane was cleaned with a cotton swab, and thebottom surface was stained with DAPI and mounted on a glassslide. The number of cells that migrated to the lower surface of themembrane was determined by nuclei counts under the microscope.

Microfluidic Migration on Collagens I and IV. Microfluidic devicescomposed of microchannels coated with collagen I and collagenIV were used to test the ability of BCCLOX and control cells toinvade and migrate in the presence or absence of βAPN (100μM). Devices were placed in a multiwell plate, and 2 μL cellsuspension were loaded at a density of 106 cells/mL The platewas mounted on the automated stage of a Zeiss Axiovert mi-croscope equipped with an environmental chamber set at 37 °Cand 5% CO2. Cells were imaged using a 10× objective, and phase

contrast with individual frames was acquired from eight differentlocations of each device every 10 min for 24 h. The resultingmovies were analyzed by manual tracking of 50 cells per frameusing ImageJ (National Institutes of Health).

Anoikis Assays. Cancer cells (30,000) were starved in 2% FBS/DMEM for 24 h and then trypsinized and suspended in serum-free media. Tubes were continuously rotated at 37 °C for theduration of the experiments. Viable cells were counted using thetrypan blue exclusion assay.

Real-Time PCR Analysis.RT-qPCR on cultured cells was performedas standard. Most primers were purchased from Eurofin MWGOperon. LOX primers were purchased from Qiagen. Primersequences used are as follows: Human Lysyl Oxidase: QiagenQuantitect primer assay hs LOX_1_SG QT00017311, Lot Number91332315. LOX-like 1: (R) 5′-AGCACCCGGGAGCTACTCT-3′and (F) 5′-GCTGATCCAGTGGGAGAACAA-3′. LOX-like 2:(R) 5′-TCCCTGTCTTCGGGCTGAT-3′ and (F) 5′-GGATCC-CTGAAACCAAAAAGC-3′. LOX-3: (R) 5′-CCCACCAAAA-GGCTAGAGGTA-3′ and (F) 5′-GAGGTCAGGATGGTCA-GCTCCAGT-3′. LOX-like 4: (R) 5′-GATCTCTGTGGCAA-GATTCAG-3′ and (F) 5′-GTGGCCTATGGAATATGTCC-3′.Human Twist: (R) 5′-TCTGGAGGACCTGGTAGAGG-3′ and(F) 5′-GGAGTCCGCAGTCTTACGAG-3′. Snail: (R) 5′-CCAG-GCTGAGGTATTCCTG-3′ and (F) 5′-CCTCCCTGTCAGAT-GAGGAC-3′. Slug: (R) 5′-TCCTCATGTTTGTGCAGGAG-3′and (F) 5′-GGGGAGAAGCCTTTTTCTTG-3′. Zeb1: (R) 5′-CCCTGGTAACACTGTCTGGTC-3′ and (F) 5′-AAGAAAGT-GTTACAGATGCAGCTG-3′. Vimentin: (R) 5′-GCTTCCTGTA-GGTGGCAATC-3′ and (F) 5′-GAGAACTTTGCCGTTGAAGC-3′. Fibronectin: (R) 5′-TCCCTCGGAACATCAGAAAC-3′ and (F)5′-CAGTGGGAGACCTCGAGAAG-3′. E-cadherin: (R) 5′-CAT-TCTGATCGGTTACCGTGATC-3′ and (F) 5′-AGAACGCATT-GCCACATACACTC-3′. N-cadherin: (R) 5′-CCGAGATGGGGT-TGATAATG-3′ and (F) 5′-ACAGTGGCCACCTACAAAGG-3′.Smooth Muscle Actin: (R) 5′-GCGTCCAGAGGCATAGAGAG-3′and (F) 5′-ACTGGGACGACATGGAAAAG-3′. CD44: (R)5′-ACTGGGGTGGAATGTGTCTTGGTC-3′ and (F) 5′AG-GAGCAGCACTTCAGGAGGTTAC-3′. HIF-1α: (R) 5′-ACA-GTCACCTGGTTGCTGCA-3′ and (F) 5′-GAATGCTCAGAG-GAAGCGAAA-3′. GAPDH: (R) 5′-GCCCAATACGACCAA-ATCC-3′ and (F) 5′-AGCCACATCGCTCAGACAC-3′. β-Actin:(R) 5′-CACGCAGCTCATTGTAG-3′ and (F) 5′-AGGTATCCT-GACCCTGAAG-3′. 18S: (R) 5′-CCATCCAATCGGTAGTAG-CG-3′ and (F) 5′-GTAACCCGTTGAACCCCATT-3′. For theclinical samples, total RNA was extracted from cancer cell-enrichedmacrodissected tumors using miRNeasy (Qiagen).

ALDEFLUOR Assay and Analysis. The ALDEFLUOR kit (StemCell)was used to measure aldehyde dehydrogenase 1 (ALDH1) en-zymatic activity in BCCs from direct coculture with GFP-labeledMSCs. Sorted MDA-MB-231 and MCF7/Ras cells previouslyexposed to MSCs were suspended in ALDEFLUOR assay buffercontaining ALDH1 substrate and incubated for 1 h at 37 °C.Treatment with diethylaminobenzaldehyde (DEAB), an ALDHinhibitor, served as negative control. Stained cells were analyzedusing the FITC channel on the BD-LSRII flow cytometer.

Mammosphere Assays. BCCs were suspended at a density of 1,000cells/well in ultra low attachment six-well plates. Cells were in-cubated for 5 d and imaged every day, and the mammosphereswere quantified. For secondary mammospheres, the primaryspheres were dissociated with Trypsin, counted, and reseeded ata density of 1,000 cells/well for another 5 d.

Western Blots.Cells were lysed using 150 mM sodium chloride, 1%Triton X-100, 1% sodium deoxycholate, 0.1% SDS, 50 mM

El-Haibi et al. www.pnas.org/cgi/content/short/1206653109 2 of 12

Tris·HCl, pH 7.5, and 2 mM EDTA supplemented with ProteaseInhibitor Mixture kit (Roche) and phosphatase inhibitor mixture(Roche). Wells were loaded with 50 μg protein. After separationby denaturing gel electrophoresis, proteins were transferred toImmobilon-P transfer membranes (IPVH00010; Millipore) andprobed overnight at 4 °C using primary antibodies against LOX(ab31238; Abcam), Twist1 (sc-15393; Santa Cruz), fibronectin(610078; BD Transduction), E-cadherin (610182; BD Trans-duction), N-cadherin (610921; BD Transduction), smooth muscleactin (ab5694; Abcam), vimentin (NCL-VIM-V9; NOVOCAS-TRA), CD44 (3578; Cell Signaling), GFP (A6455; Invitrogen),histone H1 (MA1-35403; Pierce), and β-actin (Cell Signaling).Primary antibody incubations were followed by incubation withHRP-labeled secondary antibodies to detect the immune com-plexes. Signals were developed using Immobilon-Western chemi-luminescent-HRP substrate from Millipore.

Immunofluorescent Staining. Cells cultured alone or admixed withMSCs were seeded into four-well chamber glass slides at a 1:3ratio and incubated for 3 d. Cultures were then washed two timeswith PBS, fixed with 2% PFA for 15 min, permeabilized (0.1%Triton X-100, PBS, 0.5% albumin, 0.02% NaN3) for 10 min, andblocked in normal goat serum for 30 min at room temperature.After blocking, cells were incubated in primary antibody solutionagainst Twist1 (ab50887; Abcam) and CD44 (3570; Cell Signaling)overnight at 4 °C. The next day, the chambers were washed withPBS, and cells were incubated in FITC- or tetramethyl rhodamineisothiocyanate (TRITC)-labeled secondary antibody solution for1 h at room temperature. Slides were mounted in Vectashieldmounting solution containing DAPI and covered with a cover-slip. Pictures were taken under fluorescent microscopy (Nikon).

Animal Experiments. Nude mice were injected s.c. with BCCs (106

cells) overexpressing LOX or containing vector controls. An-other set of mice was injected with BCCs harboring LOX-shRNA cells alone or admixed with MSCs. Cells with Luc-shRNA were used as controls. Tumor growth was monitoredover weeks and assessed by caliper measurements. At the end ofthe study, tumors, lungs, and ribs were harvested. Metastasis tolung was assayed by counting the number of GFP-positive nod-ules under fluorescence microscopy. Metastasis to bone was as-sessed using immunohistochemistry and counting the number ofcolonies that stained positive for GFP. The number of GFP-positive colonies was averaged for each group and reported asfold change relative to control. Where indicated, tumors weredissociated with collagenase, sorted based on GFP, and analyzedby RT-qPCR to probe for LOX mRNA expression. Lungs andribs were formalin-fixed and processed for immunohistochemicalanalysis as described below.

Immunohistochemistry. Tumors, lungs, and ribs were fixed over-night in 10% formalin. Ribs were additionally softened indecalcifying solution for 2 d. Fixed tissues were embedded inparaffin and sectioned. Antigen retrieval was done in a pressurecooker by boiling sections immersed in sodium citrate for 1 h.Sections were blocked in normal goat serum for 1 h and incubatedin 1:500 dilution of anti-LOX (ab31238; Abcam) or anti-GFPprimary antibody (A6455; Invitrogen) overnight at 4 °C followedby secondary anti-rabbit biotinylated antibody (Vector Laborato-ries). Avidin–biotin solution was added on the slides followed bydetection with DAB and then counterstaining with hematoxylin.For staining of snap-frozen tumors obtained from patients with

metaplastic breast cancer, sections were fixed in freshly prepared4% PFA solution for 10 min and then dipped in hot citrate buffersolution (10 mM) for 12 min. Sections were chilled on ice, washedtwo times with PBS, and blocked in 7% horse serum for 30 min atroom temperature. Primary antibody incubation with Twist1(1:100, ab49254; Abcam) and CD44 (1:100, ab6124; Abcam) was

done overnight at 4 °C. This incubation was followed by washingand the addition of secondary antibody solution using AlexaFluor-labeled donkey anti-rabbit 594 or goat anti-mouse 488 antibodies(1:300; Molecular Probes) for 1 h at room temperature. Hoechstsolution was used for nuclear staining (1:10,000) for 3 min. Slideswere then rinsed with distilled water and covered with polyvinylalcohol mounting medium (Fluka).

ChIP Analyses. GFP-labeled MDA-MB-231 cells were coculturedwithMSCs for 72 h and then sorted, lysed, and processed for ChIPusing the MAGnify ChIP System (Invitrogen). Briefly, 106 cellswere treated with formaldehyde to cross-link chromatin, lysedusing buffer provided in the kit, and incubated with polyclonalanti-CD44 antibody (3578; Cell Signaling) overnight at 4 °C. Thechromatin was then washed, the cross-linking was reversed, andthe DNA was purified before PCR analysis with LOX promoterprimers: full-length forward 5′-CGCGAATTCTGTGAAGATG-CCTTCTGAGTG-3′ and reverse 5′-CGCGGATCCGGGAAG-GAGAAATCTTCAACC-3′; SAT2 primers forward 5′-CTGCA-ATCATCCAATGGTCG-3′ and reverse 5′-GATTCCATTCGG-GTCCATTC-3′ served as controls.

Statistical Analysis.All in vitro and in vivo data were analyzed usingStudent’s t test. Significance interval was set at P < 0.05.

Clinical Correlation Studies. Raw datasets of GSE2109 with 2,158various human cancer samples were downloaded from the Na-tional Center for Biotechnology Information Gene ExpressionOmnibus (http://www.ncbi.nlm.nih.gov/projects/geo/query/acc.cgi?acc=GSE2109). There are 333 breast cancer samples in thisdataset. Raw array intensity CEL files were normalized with Af-fymetrix package from bioconductor (http://www.bioconductor.org) software. Normalized intensity value for both TWIST1 (probeset ID 213943_at) and LOX (probeset ID 204298_s_at and215446_s_at) were pulled out for correlation analysis with all 2,158cancer samples as well as only breast cancer samples. The expres-sion level of LOX was calculated as an average of the two probesets. The correlation plot with r and P values between log2 TWIST1and log2 LOX expression intensity was calculated by EXCEL.Raw datasets of GSE1456, GSE2034, and GSE4922 were also

downloaded from the National Center for Biotechnology In-formation Gene Expression Omnibus website, log-transformed ifnot already done so, and median-centered. Pearson correlationwas calculated between LOX (probeset 215446_s_at) andTWIST1 (probeset 213943_at), and a linear model was fit. The rvalues for the three studies were GSE1456 = 0.48, GSE2034 =0.41, and GSE4922 = 0.41.

Patient Survival Studies. Using median-centered log ratios fromGSE2034, samples were partitioned into groups of high or lowlevels of LOX and TWIST1 by comparison with the medianof all samples. Samples with high or low levels of both LOXand TWIST1 were selected for additional analysis. Togetherwith clinical parameters from the National Center for Bio-technology Information Gene Expression Omnibus, survivalanalysis was performed using Kaplan–Meier estimates usingthe log-rank test and the proportional hazard model to com-pare survival curves.

SI ResultsThe LOX promoter harbors HIF1α-response elements (3), andhypoxic conditions were shown to stimulate LOX transcriptionalactivation (4). Interestingly, de novo stimulation of LOX mRNAby MSCs was not accompanied by HIF1α up-regulation (Fig.S9A). In addition, treatment of the MDA-MB-231:MSC co-cultures with 10 μM of a cell-permeable amidophenolic com-pound that prevents HIF1α accumulation (5, 6) (Fig. S9 B andC) was not sufficient in blocking the ability of MSCs to trigger

El-Haibi et al. www.pnas.org/cgi/content/short/1206653109 3 of 12

LOX expression in the cancer cells (Fig. S9A). These resultsindicated that MSC-triggered LOX up-regulation in the cancercells may be independent of HIF1α.Most recently, the work by Levental et al. (7) described the in-

duction of LOX protein expression in mammary epithelial cellsthat invaded firm matrices and suggested that stiff stroma may beregulating LOX in the cancer cells by some form of mechanicalsignaling (7). Furthermore, previous reports indicated thatmechano-transduction signaling critically relies on the activity ofthe Rho effector ROCK and that the ROCK inhibitor Y27632 wassufficient to disrupt the ability of cells to sense mechanical tensionin the tumormicroenvironment (8). However, treatment ofMSC+BCC cultures with Y27632 (10 μM) (Fig. S9D andE) did not affectLOXmRNA levels inMSC-stimulated cancer cells, suggesting thatLOX induction in our BCCMSC model does not result from tensionpotentially exerted on the cancer cells by juxtaposed MSCs.

We next explored whether MSC-derived soluble factors wereresponsible for LOX up-regulation in the cancer cells. In thiscontext, we found that neither the conditioned media derived fromMSC cultures (Fig. S9F) nor culture of cancer cells and MSCs onopposite sides of membranes in 0.4-μm transwell chambers (Fig.S9G) elicited any LOX overexpression in the cancer cells. To in-vestigate the possibility that the contact of MSCs with cancer cellsenables the release of soluble mediators that then prompt cancercells to overexpress LOX, we allowed BCCs to exchange factorswith BCC:MSC cocultures plated across 0.4-μm membranes (Fig.S9G). However, RT-qPCR on the BCCs stimulated by BCC:MSCmedia (Fig. S9G, Bottom Left) revealed no stimulation of LOXexpression. Similarly, the culture of BCCs with MSCs across combsseparated by 80-μm gap (9) (Fig. S9H) did not result in the acti-vation of LOX transcription in the BCCs (Fig. S9I), stronglyshowing that the MSC-derived signal responsible for robust LOXstimulation in cancer cells is not soluble in nature.

1. Martchenko M, Jeong SY, Cohen SN (2010) Heterodimeric integrin complexescontaining beta1-integrin promote internalization and lethality of anthrax toxin. ProcNatl Acad Sci USA 107:15583–15588.

2. Godar S, et al. (2008) Growth-inhibitory and tumor- suppressive functions of p53depend on its repression of CD44 expression. Cell 134:62–73.

3. Sahlgren C, Gustafsson MV, Jin S, Poellinger L, Lendahl U (2008) Notch signaling mediateshypoxia-induced tumor cell migration and invasion. Proc Natl Acad Sci USA 105:6392–6397.

4. Erler JT, et al. (2006) Lysyl oxidase is essential for hypoxia-induced metastasis. Nature440:1222–1226.

5. Lee K, et al. (2007) (Aryloxyacetylamino)benzoic acid analogues: A new class ofhypoxia-inducible factor-1 inhibitors. J Med Chem 50:1675–1684.

6. Lee K, et al. (2010) LW6, a novel HIF-1 inhibitor, promotes proteasomal degradation ofHIF-1alpha via upregulation of VHL in a colon cancer cell line. Biochem Pharmacol 80:982–989.

7. Levental KR, et al. (2009) Matrix crosslinking forces tumor progression by enhancingintegrin signaling. Cell 139:891–906.

8. Ghosh K, et al. (2008) Tumor-derived endothelial cells exhibit aberrant Rho-mediatedmechanosensing and abnormal angiogenesis in vitro. Proc Natl Acad Sci USA 105:11305–11310.

9. Hui EE, Bhatia SN (2007) Micromechanical control of cell-cell interactions. Proc NatlAcad Sci USA 104:5722–5726.

El-Haibi et al. www.pnas.org/cgi/content/short/1206653109 4 of 12

CA CSM-MBCSM-MB GFP-MCF7/RasGFP-MDA-MB-231

Co-culture for 3 days

Fluorescence-activated cell sorting p53 signaling pathway

Pathways in cancer p=0.032

g

RNA extraction

GFP-MCF7/RasMSC

Affymetrix U133 gene expression array

GFP-MDA-MB-231MSC

Focal adhesion

TGF-beta signaling pathway

p=2.13E-9

p=0.008

g p y

6 genes 2 genes0 5 10 15 20

ECM-receptor interaction

Fold enrichment

p=1.76E-10

46 genes74 genes 7 genes

MDA-MB-231MSC - MDA-MB-231 MCF7/RasMSC - MCF7/RasBNCBI gene ID and symbol

1281 COL3A1

1278; COL1A2

4015; LOX

11167; FSTL1

Log2

7.2

6.9

5.4

5 4

Gene name

Collagen, type III, α1

Collagen, type I, α2

Lysyl oxidase

Follistatin like 1

p.value

7.78E-06

1.33E-06

8.48E-06

8 71E 06

p.value.adj

0.006

0.002

0.006

0 006

Log2

7.8

9.2

5.8

5 3

p.value

1.33E-05

2.80E-07

1.70E-05

4 09E 05

p.value.adj

0.01

0.0009

0.01

0 0211167; FSTL1

3486; IGFBP3

10631; POSTN

151887; CCDC80

5.4

5.2

5.1

3.7

Follistatin-like 1

Insulin-like growth factor binding protein 3

Periostin

Coiled-coil domain containing 80

8.71E-06

2.31E-05

9.96E-07

5.93E-05

0.006

0.01

0.002

0.021

5.3

5.9

8.2

5.1

4.09E-05

2.57E-05

3.02E-08

1.20E-05

0.02

0.01

0.0001

0.01

p=0.025

Fig. S1. Identification of MSC-modulated genes in BCCs. (A) GFP-labeled MDA-MB-231 and MCF7/Ras cells were cultured alone or on a monolayer of MSCs for3 d (BCC:MSC ratio was 1:3). Cells were then sorted based on GFP, and their RNA was purified and processed for gene expression analysis using Affymetrix-based arrays. (B) When all BCCMSC samples were analyzed together, seven genes were found to be increased in both MDA-MB-231MSC and MCF7/RasMSC. (C)Gene set enrichment analysis categorizing the overall modulated genes into the indicated canonical pathways.

El-Haibi et al. www.pnas.org/cgi/content/short/1206653109 5 of 12

ALOX250

p=0.001

LOX-L1LOX-L2LOX-L3LOX-L4

100

150

200

RN

A / G

APD

H

p=0.0001

B

0

10

MDA-MB-231 MDA-MB-231MSC MCF7/Ras MCF7/RasMSC

m

0

2

34

5

OX

mR

NA

/ GA

PDH

10LO

BCCs - +

T47D

- +

MCF7

- +

MCF7/Ras

- +

MDA-MB-435

- +

MDA-MB-231

p=0.0142C

4

6

8

10

OX

mR

NA

/ GA

PDH

p=0.0127

-2

0

2

MSCMDA-MB-231MSC

MCF7/RasMSC

2^-∆

Ct L

O

Fig. S2. Induction of LOX is specific and occurs in MSC-activated cancer cells and not the cancer-activated MSCs. (A) GFP-labeled BCCs were cultured alone oron a monolayer of MSCs for 3 d (BCC:MSC ratio was 1:3). Cancer cells were then sorted, lysed, and analyzed for LOX, LOX-L1, LOX-L2, LOX-L3, and LOX-L4mRNA expression by RT-qPCR. Results were normalized to GAPDH. Data are expressed as fold induction ± SEM (n = 3). (B) MSCs were sorted out of cocultureswith the indicated GFP-labeled BCCs, and their RNA was isolated and analyzed for LOX mRNA expression by RT-qPCR. Results were normalized to GAPDH. Dataare expressed as fold induction ± SEM (n = 3). (C) RT-qPCR data showing the difference (ΔCt) between LOX amplification cycles in cDNA derived from theindicated groups.

El-Haibi et al. www.pnas.org/cgi/content/short/1206653109 6 of 12

BA25

30

AP

DH

p=0.005

p=0.001

0

5

10

15

20

Control LOX Control LOX

LO

X m

RN

A / G

A

LOX

GAPDH

Control LOX Control LOX

32 kDa

36 kDa

MDA-MB-231 MCF7/Ras

40

50

60

on

ed

m

ed

ia

p p

p=0 00006 p=0 0001

DC

20

25

30

l lys

ate

p=3.425E-07

MDA-MB-231 MCF7/Ras

0

10

20

30

40

RF

U fro

m C

on

ditio

(F

old

)

Ctrl Ctrl LOX LOX Ctrl Ctrl LOX LOX

bAPN (100 μM) - + - + - + - +

p=0.024 p=0.22

p=0.00006 p 0.0001

0

5

10

15

20

RF

U f

ro

m C

ell

(F

old

)

p=0.0005 p=0.0064

p=6.543E-07

p=1.332E-06

Ctrl Ctrl LOX LOX Ctrl Ctrl LOX LOX

bAPN (100 μM) - + - + - + - +

MDA-MB-231 MCF7/Ras

+ + + +

MDA-MB-231 MCF7/Ras

+ + + +

FE

Co

ntro

l

Baseline 0 hr 12 hrs Baseline 0 hr 12 hrs

100

120

/m

in

)

Collagen IV

30

35Collagen IV

p=0.00032

p=0.0004C

LO

XA

PN

5x 5x

20

40

60

80

Ave

ra

ge

V

elo

city (µ

m/

5

10

15

20

25

Control LOX LOX+βAPN

LO

X +

saR/7FCM132-BM-ADM

0 0Control LOX LOX+βAPN

saR/7FCM132-BM-ADM

=0.0003=0.00005p=9.72E-06

Fig. S3. βAPN reduces the ability of exogenous LOX to promote cellular migration. (A) LOX mRNA expression levels detected by RT-qPCR on RNA derived fromMDA-MB-231 and MCF7/Ras cells constitutively expressing LOX. Results were normalized to GAPDH. Data are expressed as fold induction ± SEM (n = 3). (B)Western blot analyses of the total lysates of MDA-MB-231 and MCF7/Ras cells overexpressing LOX. GAPDH was used as a loading control. The (C) culture mediaand (D) lysates of control (Ctrl) MDA-MB-231 and MCF7/Ras cells or cells overexpressing LOX (LOX) were harvested, and the enzymatic activity of LOX wasmeasured using Amplite-based fluorometric analysis. Data are expressed as fold change in relative fluorescensce units (RFU) ± SEM (n = 3). (E) Cells over-expressing LOX with or without βAPN (100 μM) were allowed to migrate on collagen types I and IV across microchannels and recorded for 24 h. Movies wereanalyzed using manual tracking (ImageJ; National Institutes of Health), and average cell velocities (in micrometers per min) were calculated. Data are rep-resentative of 50 cells/frame. (F) Wound healing assays initiated by a straight streak in a confluent monolayer of controls or MDA-MB-231LOX and MCF7/RasLOX

cells treated with or without βAPN (100 μM). Cells were imaged at 0 and 12 h.

MDA MB 231 ControlBA - - -C

60

80

100

MDA-MB-231LOX

MDA-MB-231-Control

e (x

100

mm

3 )

1.5

2

2.5

or w

eigh

t (g)

p=0.029

6

8

10

tasi

s (F

old)

p=0.0012p=0.133

MDA-MB-231LOX

0

20

40

1 2 3 4 5 6 7 8 9

Tum

or v

olum

e

Weeks0

0.5

1

Control LOX

Ave

rage

tum

o

0

2

4

Control LOX

Lung

met

ast

MDA-MB-231 MDA-MB-231

Fig. S4. Exogenous LOX is sufficient in promoting metastasis. (A) A total of 106 MDA-MB-231-control and MDA-MB-231LOX cells were injected s.c. into nudemice. Tumor kinetics were followed for 9 wk. Data represent mean ± SEM (n > 7 for each group). (B) Average tumor weight of xenografts in A ± SEM. (C)Representative pictures of GFP-positive nodules in the lungs of mice described in A and B. There was a 5.132-fold increase in the number of GFP-positivenodules observed in the lungs of the MDA-MB-231LOX

–bearing mice compared with their controls.

El-Haibi et al. www.pnas.org/cgi/content/short/1206653109 7 of 12

A50 MDA-MB-231

MDA-MB-231MSCp=0.002

CBTWIST DAPI Merged

10

20

30

40 MCF7/RasMCF7/RasMSC

mR

NA

/ 1

8S

rR

NA

p=0.0097

p=0.627p=0.087

Twist

H1

Nuclear fraction

28 kDa

33 kDa

- βAPN

D E

160180200

shLucshLOX-2

MDA-MB-231MSC

DH

0 0004

0TWIST SNAIL SLUG ZEB1

Nuclear fraction+ βAPN

/

MDA-MB-231MSC

120140 p=0.022

0 0002

020406080

100120140

LOX TWIST E-Cadherin SMA

mR

NA

/ G

AP p=0.0004

p=0.006p=0.217

p=0.34

LO

X m

RN

AG

AP

DH

shLuc shLOX-1 shLOX-20

20406080

100

shLOX-3 shLOX-4 shLOX-5

F

406080

100120140160180200

shLucshLOX-2

MCF7/RasMSC

mR

NA

/

GA

PD

H

p=0.0002

p=0.0005

p=0.041p=0.086

MCF7/RasMSC

20406080

100120140

LO

X m

RN

A /

GA

PD

H p=0.0106

020

LOX TWIST E-Cadherin SMA

G MCF7/RasMSCMCF7Ras H MDA-MB-231 MCF7/Ras

Ctrl LOX LOXCtrl LOX LOX

shLuc shLOX-1 shLOX-2 shLOX-3 shLOX-4 shLOX-50

28 kDa

56 kDa

Twist

Actin

βAPN (100 μM) - + - +

SMA

Vimentin

Actin

βAPN (100 μM) - - +

42 kDa

56 kDa

60 kDa

- - +

Fig. S5. LOX activity is required for MSC-triggered and Twist-dependent epithelial-to-mesenchymal transition (EMT). (A) The expression levels of EMTtranscription factors were analyzed by RT-qPCR on RNA derived from MDA-MB-231MSC and MCF7/RasMSC and compared with controls. Results were normalizedto 18S and are expressed as fold ± SEM. (B) Anti–Twist-TRITC immunofluorescence analyses of MDA-MB-231 control and MDA-MB-231MSC cells treated with orwithout βAPN (100 μM). DAPI is used to stain the nuclei. (C) Western blot analysis of nuclear extracts of MDA-MB-231 or MDA-MB-231MSC probed with anti-Twist and anti-histone H1 antibodies. (D) MDA-MB-231 and MCF7/Ras cells stably expressing the indicated LOX-shRNA plasmids (shLOX) were stimulated withMSCs for 3 d, after which LOX silencing was analyzed by RT-qPCR on sorted BCCs. shRNA against luciferase (shLuc) served as the control. Data were normalizedto β-actin and are represented as fold induction in LOXmRNA expression ± SEM. (E and F) GFP-labeled BCCs harboring shLuc or shLOX-2 were cultured alone orwith MSCs for 3 d. Sorted BCCs were subjected to RT-qPCR for the indicated genes. Results were normalized to GAPDH. Data are expressed as fold induction ±SEM (n = 3). (G) GFP-MCF7/Ras cells cultured alone or with MSCs were treated with vehicle or βAPN (100 μM). BCCs were sorted as described earlier, lysed, andsubjected to Western blot analysis for Twist and Actin. (H) Western blots on control (Ctrl) or LOX-overexpressing cells (LOX) using antibodies for the indicatedproteins. Actin was used as loading control.

El-Haibi et al. www.pnas.org/cgi/content/short/1206653109 8 of 12

MDA-MB-231A BA

/G

AP

DH

-

Twist

MDA-MB-231

3035404550 p=0.0012

p=0.0005

Tw

ist

mR

N

hT i t 4hT t 3hT i t 1hLMSC - + - + - + - + - +

Actin

shTwist 4shTwist 2shTwist 1shLuc- + - + - + - +MSC

05

10152025

p=0.351

s w s -s wis -hT i t 2s w s --s w ss uc i ---

C MDA-MB-231

p=0.0034

150

200

250

A/

GA

PD

H

p=0.0025

0

50

100

LO

Xm

RN

A

p=0.162

shTwist-4shTwist-2shLuc

- + - + - +MSC

Fig. S6. Twist is required for MSC-triggered LOX induction in the cancer cells. (A) RT-qPCR on sorted cancer cells stably expressing the indicated shRNAconstructs after culture with or without MSCs for 3 d. Data were normalized to GAPDH and are presented as fold of Twist mRNA expression ± SEM. (B) Westernblot results on select groups in A. Actin was used as a loading control. (C) GFP-MDA-MB-231 control and GFP-MDA-MB-231-shTwist cells sorted out from MSCcocultures were analyzed by RT-qPCR for LOX mRNA expression levels. Data were normalized to GAPDH and are presented as fold induction ± SEM.

El-Haibi et al. www.pnas.org/cgi/content/short/1206653109 9 of 12

BAMDA-MB-231-Ctrl MDA-MB-231LOX

BAED+BAED+0% 0.34% 0.06% 0.54%

D

2 98% 68% 49 5%

Alone +MSCs +MSC+βAPN

MD

Control LOX

MDA-

MB-2

31

10 10x

CMCF7/Ras-Ctrl MCF7/RasLOX

+DEAB +DEAB0.16% 1% 0.19% 0.54%

2.98% 68% 49.5%

1.08% 41.5% 71.9%

DA

-MB

-231M

CF

MCF7

/Ras

x

10x 10x

osph

eres

8101214

FE + MSC H

F7/Ras

Total

Mam

m

02468

Ctrl LOX Ctrl LOXMDA-MB-231 MCF7/Ras

60 MDA MB 231Alone

MD

A-M

B-231

BAED+BAED+

MC

F7/BAED+BAED+

0.14% 0.94% 0.04% 0.15%

0.51% 0.74% 0.44% 0.75%10

20

30

40

50

60

ota

l M

am

mo

sp

he

res MDA-MB-231

MCF7/Ras

JMDA-MB-231-shLOX-2

Ras

I Ctrl BCCMSC BCCMSC+βAPN

MB

-231 1 2 3

G +MSC+βAPNAlone +MSC

1

0T

Ctrl shLOX-2

MSCβAPN

- + -+ ++- - - -

Ctrl shLOX-2

- + -+ ++- - - -

60708090

100

500 µ

m)

MDA-MB-231-shLOX-2MSC

1.93%

shLOX-2MSC

MD

A-M

CF7

/Ras

Ctrl LOX

7 8 9

10X

231

MD

A-M

B-2

3

4x 4x 4x

10x 10x 10x

8090

100

µm

)

0102030405060

1 2 3 4 5 6

2°

sp

here

s (

<

7 8 9 10 11 12

4 5 662.8%

10 11 12

10X

MD

A-M

B-2

MC

F7/R

asMC

F7/R

as

4x 4x 4x

10x 10x 10x 01020304050607080

1 2 3 4 5 6

2°

sp

here

s (

> 5

00

7 8 9 1 11 12 ALDH-1

SS

C-A

SS

C-A

ALDH-1

SS

C-A

ALDH1

ALDH-1

MSC

0.4 µm

filter

Cancer

cells

Fig. S7. Induction of the cancer stem cell phenotype by MSCs is not dependent on LOX. (A) ALDH1 activity determined on control cells and BCCLOX. Panels arerepresentative of three independent repeats and compared with DEAB-treated samples. (B) Mammosphere formation of control cells (Ctrl) and BCCLOX (LOX)seeded in low-attachment conditions. (C) Total numbers of spheroids in B were quantified on day 5 after culture and averaged from three independent repeats,each performed in triplicates (mean ± SEM). (D) ALDH1 activity assays on sorted cancer cells cultured alone, from MSCs, or stimulated by MSCs in the presence ofβAPN (100 μM) for 3 d. Samples were normalized to DEAB-treated controls. (E) MSCs were cultured with cancer cells in Boyden chambers across 0.4-μm filters,allowing for exchange of soluble factors. (F) Cancer cells in E were recovered and subjected to ALDEFLUOR assays to determine ALDH1 activity. Panels arerepresentatives of three independent repeats and compared with DEAB-treated controls. (G) Primary mammosphere formation of BCC controls or BCC sortedfrom 3-d cultures with MSCs or 3-d cultures with MSCs in the presence of βAPN (100 μM). (H) Spheroids of low-attachment cultures in the indicated groups werequantified on day 5 after culture and averaged from three independent repeats, each performed in triplicates (mean ± SEM). (I) Secondary mammospheres weregenerated from primary spheres of the same groups that have been dissociated and reseeded at a density of 1,000 cells/well in low attachment plates. Bar graphsrepresent the number of spheres per group categorized by size (<500 or >500 μm) and counted at day 5 after culture. (J) ALDH1 activity determined byALDEFLUOR assays on MDA-MB-231 controls or MDA-MB-231-shLOX-2 sorted out of cocultures with MSCs. Results were normalized to DEAB-treated controls.

El-Haibi et al. www.pnas.org/cgi/content/short/1206653109 10 of 12

1600

MDA-MB-231-shLOX-2MDA-MB-231-shLOX-2 + MSCMDA-MB-231-shLucMDA-MB-231-shLuc + MSC

1000

1200

1400

(m

m3)

ule

s p

er

lun

g)

25 p=0.036

30 p=0.039

400

600

800

Tu

mo

r vo

lu

me

(nu

mb

er

of

no

du

10

15

20

p=0.0179

0

200

1 2 3Weeks

Meta

stasi

s

0

5

Fig. S8. MSC-induced metastasis depends, in part, on LOX activity. Nude mice were injected with 0.7 × 106 MDA-MB-231 cells expressing shLuc or MDA-MB-231 cells expressing shLOX-2 alone or together with 2.1 × 106 MSCs. Tumor kinetics were measured for 3 wk, after which the mouse lungs were excised andanalyzed for GFP-positive nodules. Indicated values represent the average number of GFP-positive clusters per lung section. n = 10 mice per group.

El-Haibi et al. www.pnas.org/cgi/content/short/1206653109 11 of 12

A CB

Actin

HIF1α

Normoxia Hypoxia

Ctrl+MSC

+MSC+InhCtrl

Ctrl+Inh

56 kDa

120 kDa

p=0.0003 LOXHIF1

p=0.00009

Dp=0.0001

p=0.00005

p=0.0001

p=0.0003

E F GMDA-MB-231+231-CMMCF7/Ras+Ras-CMMDA-MB-231+MSC-CMMCF7/Ras+MSC-CMMDA-MB-231+(231+MSC)-CMMCF7/Ras+(Ras+MSC)-CMMDA-MB-231 aloneMCF7/Ras aloneMDA-MB-231+MSCMCF7/Ras+MSC

BCC

BCC

BCC

0.4 μm insert

MSC

MSC + BCC

BCC

MSC+MDA-MB-231/MDA-MB-231

MSC+MCF7/Ras/MCF7/Ras

MDA-MB-231 aloneMDA-MB-231/MDA-MB-231MSC/MDA-MB-231

MCF7/Ras aloneMCF7/Ras/MCF7/RasMSC/MCF7/Ras

MDA-MB-231MDA-MB-231+MSCMCF7/Ras aloneMCF7Ras+MSC

0

10

100

150

200

250

300 p=0.00002

p=0.000001

0

10

100

150

200

250 p=0.000001

p=0.000009

Conditioned ECM from

monolayer of MSCs

Cells and medium

discarded

Cancer cells seeded on ECM from

MSCs

J

p=0.018

p=0.0071

H Day 1Day 3

p=0.004p=0.005

I K

p=0.107

SEPARATED GAP CONTACT

Fig. S9. Induction of LOX by MSCs is mediated by ECM factors. (A) RT-qPCR of LOX and HIF1α on sorted MDA-MB-231 cells cultured alone or on a monolayer ofMSCs for 3 d (BCC:MSC ratio was 1:3) in the presence or absence of HIF1α inhibitor (LW6; 10 μM). Results were normalized to β-actin. Data are expressed as foldinduction ± SEM (n = 3). (B) RT-qPCR of HIF1α on MDA-MB-231 cells cultured under hypoxia in the presence or absence of HIF1α inhibitor (C) Western blotanalysis of cell lysates of MDA-MB-231 cells cultured alone or sorted out of cocultures with MSCs with or without HIF1α inhibitor under normoxia or hypoxia.(D) RT-qPCR for LOX on sorted BCCs cultured alone or along with MSCs in the presence or absence of the ROCK inhibitor Y27632 (10 μM). Data are expressed asfold induction ± SEM (n = 3). (E) Western blot analysis of cell lysates of MDA-MB-231 or MCF7/Ras cells cultured alone or sorted out of cocultures with MSCswith or without the ROCK inhibitor. Antiphospho-FAK antibody was used as a measure of ROCK activation. (F) RT-qPCR analyses on MDA-MB-231 and MCF7/Ras cells treated for 3 d with the conditioned media (CM) derived from the indicated cultures. Data represent mean ± SEM. Data were normalized to GAPDH.(G) RT-qPCR analyses on MDA-MB-231 and MCF7/Ras cells cultured in the Boyden chamber set up (Left, bottom wells) for 3 d with the indicated culture groupson the membrane top (Left, top wells). Data represent mean ± SEM and were normalized to GAPDH. (H) GFP-labeled cancer cells were cultured on combs andclipped together with partner combs carrying MSCs in two positions: gap and contact. (I) RT-qPCR analysis for LOX on cancer cells recovered from H. Data werenormalized to GAPDH and are represented as fold induction ± SEM (n = 3). (J) MSC monolayers were cultured for 3 d and then gently lifted with EDTA solution.The remaining ECM tethered to the culture dish was washed with PBS. Cancer cells were then plated on top of the conditioned ECM for 3 d. (K) MDA-MB-231controls or cells harboring shLOX-2 were recovered after 3 d of culture on MSC-ECM in J and subjected to RT-qPCR analyses on LOX. Values were normalized toGAPDH and are represented as fold induction ± SEM of duplicate experiments (n = 2).

Fig. S10. The induction of LOX in the cancer cells by MSCs is reversible. RT-qPCR analysis of LOXmRNA expression in MDA-MB-231 or MCF7/Ras cells sorted outafter 3-d coculture with MSCs and put back in culture alone for 21 d. Values were normalized to GAPDH and are represented as fold induction ± SEM (n = 2).

El-Haibi et al. www.pnas.org/cgi/content/short/1206653109 12 of 12