A breast cancer stem cell niche supported by juxtacrine signalling ...

ART ICLES

A breast cancer stem cell niche supported byjuxtacrine signalling from monocytes and macrophagesHaihui Lu1,2, Karl R. Clauser3, Wai Leong Tam1,2,4, Julia Fröse1,5,6, Xin Ye1, Elinor Ng Eaton1, Ferenc Reinhardt1,Vera S. Donnenberg7,8, Rohit Bhargava9, Steven A. Carr3 and Robert A. Weinberg1,2,10,11

The cell-biological program termed the epithelial–mesenchymal transition (EMT) confers on cancer cells mesenchymal traits andan ability to enter the cancer stem cell (CSC) state. However, the interactions between CSCs and their surroundingmicroenvironment are poorly understood. Here we show that tumour-associated monocytes and macrophages (TAMs) create a CSCniche through juxtacrine signalling with CSCs. We performed quantitative proteomic profiling and found that the EMT programupregulates the expression of CD90, also known as Thy1, and EphA4, which mediate the physical interactions of CSCs with TAMsby directly binding with their respective counter-receptors on these cells. In response, the EphA4 receptor on the carcinoma cellsactivates Src and NF-κB. In turn, NF-κB in the CSCs induces the secretion of a variety of cytokines that serve to sustain the stemcell state. Indeed, admixed macrophages enhance the CSC activities of carcinoma cells. These findings underscore thesignificance of TAMs as important components of the CSC niche.

Despite improvements in diagnosis and treatment, breast-cancer-associated mortality remains high owing to clinical relapseassociated with metastasis to distant organs. During primarytumour progression, breast carcinoma cells may pass through anEMT, thereby acquiring traits associated with high-grade malignancy,including motility, invasiveness and an increased resistance toapoptosis1,2. Furthermore, passage of both normal and neoplasticmammary epithelial cells through an EMT confers on the cellsmany of the properties associated with normal mammary stem cells(MaSCs) and cancer stem cells (CSCs) respectively3–5. Importantly,carcinoma cells that have passed through an EMT exhibit heightenedresistance to conventional chemotherapeutic agents and hence mayregenerate tumour growth after initial drug treatment is halted6–8.

Various types of stromal cell have been found to influence the CSCstate through paracrine signalling. As an example, mesenchymal stemcells contribute to the formation ofCSCs by secreting prostaglandin E2(PGE2), IL-6, IL-8 and Gro-α, which help to trigger activation of thepreviously latent EMT program in nearby carcinoma cells9. Periostin(POSTN) released by lung stromal fibroblasts fosters creation of ametastatic niche for breast cancerCSCs (ref. 10). Somehave speculatedthat a physical CSCnichemaintained by both paracrine and juxtacrinesignalling exists in colon cancers similar to the normal colon stem

cell niche supported by myofibroblasts11. Nevertheless, a juxtacrine-mediated CSC niche that maintains these cells in their stem cell statehas not been described.

Tumour-associated monocytes and derived macrophages (TAMs)have been shown to be involved in many aspects of tumour initiationand progression12,13. These observations have not, however, shedlight on how CSCs interact with TAMs in a manner different fromnon-CSCs, and whether heterotypic interactions contribute to theformation and maintenance of the stem cell niche.

Some have reported that normal MaSC function requires thecontinuing presence of macrophage-derived factors14, but the natureof these factors has not been explored. Moreover, it remains unclearwhether the interactions betweenmacrophages and normalMaSCs arerelevant to carcinoma pathogenesis. In the present study, we describea key mechanism by which already-formed mammary CSCs interactwith niche-formingTAMs in a contact-dependentmanner tomaintaintheir residence in the mesenchymal/stem-like state.

RESULTSQuantitative proteomic profiling of cell-surface proteinsWe and others have found extensive similarities between the stem cellprogram of normal MaSCs and that of mammary CSCs (refs 3–5).

1Whitehead Institute for Biomedical Research, Cambridge, Massachusetts 02142, USA. 2MIT Ludwig Center for Molecular Oncology, Cambridge, Massachusetts02139, USA. 3Broad Institute of Massachusetts Institute of Technology and Harvard, Cambridge, Massachusetts 02142, USA. 4Genome Institute of Singapore,60 Biopolis Street, Singapore 138672, Singapore. 5German Cancer Research Center (DKFZ), 69120 Heidelberg, Germany. 6University of Heidelberg, 69120Heidelberg, Germany. 7Hillman Cancer Center, University of Pittsburgh Cancer Institute, Pittsburgh, Pennsylvania 15213, USA. 8Department of Cardiothoracic Surgery,University of Pittsburgh, School of Medicine, Pittsburgh, Pennsylvania 15213, USA. 9Magee-Womens Hospital, University of Pittsburgh, School of Medicine,Pittsburgh, Pennsylvania 15213, USA. 10Department of Biology, Massachusetts Institute of Technology, Cambridge, Massachusetts 02142, USA.11Correspondence should be addressed to R.A.W. (e-mail: [email protected])

Received 1 July 2014; accepted 21 August 2014; published online 28 September 2014; DOI: 10.1038/ncb3041

NATURE CELL BIOLOGY VOLUME 16 | NUMBER 11 | NOVEMBER 2014 1105

© 2014 Macmillan Publishers Limited. All rights reserved.

cgrantha

Text Box

9/22/2015 JOURNAL CLUB GREGG FOX

ART ICLES

−6 −4 −2 0 2 4 6

−6

−4

−2

0

2

4

6

Replicate 2 log2 (heavy/light)

Mam

mos

phe

res

200

cells

per

wel

l

Rep

licat

e 1

log 2

(lig

ht/h

eavy

)N

umb

er o

f pep

tides

0

100

200

300

Mean = 0.02s.d. = 1

Mean = 0.02s.d. = 1

0 100 200 300 Mam

mos

phe

res

per

8 w

ells

Twist

Snail 9

0hi

Snail

HMLE

90h

i

HMLE

90m

ed

HMLE

0

20

40

60

∗∗∗

∗∗∗∗

∗∗∗∗

105

105

104

104

103

103

102

102

101

0A

PC

-A

PE-A

Q2Q1

Q3 Q4

–14

Specimen 001-HMLE_Twist_CD24+90-

Q2

Q3 Q4

Q1

CD24

CD

90

HMLE HMLE-Snail HMLE-Twist

0.1 99 18

b

c

a

d

Specimen 001-HMLE_Snail_CD24+90-

Q2

Q4Q3

Q1

Specimen 001-HMLE_JY_CD24+90-P4

0

105

105

104

104

103

103

102

102

101

0A

PC

-A

PE-A–40

0

105

105

104

104

103

103

102

102

0

–40

AP

C-A

PE-A–14

0

CDH1

EPHA4

VIMFN1

CDH2PDGFRb THY1

ROR1

Twist Snail90hi24– Snail HMLE90hi24–

Figure 1 Quantitative proteomic profiling identified EMT-induced membraneprotein changes in HMECs. (a) Expression ratios of proteins detected in themesenchymal HMLE-Twist cells over the epithelial HMLE-vector cells plottedfor two biological replicate experiments. Ratios highlighted in red were foundto be significantly differentially expressed (P<0.05) after standardizing andapplying a moderated t-test corrected by the Benjamini–Hochberg method17.Gene names are labelled for known EMT marker proteins and candidates thatwere followed up in this study. (b) The HMLE, HMLE-Twist and HMLE-Snail

cells were stained with anti-CD90 and anti-CD24 antibodies and analysed byflow cytometry. The percentage of the CD90hiCD24− population is shown oneach plot. (c,d) The cells were FACS sorted based on CD90 and CD24 levelsand seeded into 3D mammosphere medium. The numbers of mammospheresare plotted (c), and representative pictures of mammospheres formed by theindicated cell populations are shown (d). n=3 independent experiments, datarepresent s.e.m. ∗∗∗P<0.001; ∗∗∗∗P<0.0001 (Student’s t-test). Source dataare provided in Supplementary Table 7.

To identify cell-surface proteins that enable normal and neoplasticmammary stem cells to interact with nearby stromal cells, weperformed quantitative proteomic profiling of membrane-associatedproteins before and after immortalized human mammary epithelialcells15 (HMLEs) were forced experimentally to undergo an EMT (andthereby acquiremesenchymal and stemcell traits).We employed stableisotope labelling by amino acids in cell culture16 (SILAC) followedbymembrane-associated protein fractionation andmass spectrometryanalysis (MS; Supplementary Fig. 1).

Plasma-membrane- and extracellular-matrix-associated proteinswere represented by the most abundant peptides detected by MS(Supplementary Table 1). We identified 2,607 proteins reproduciblywith at least two peptide SILAC ratiosmeasured in each replicate, with460 proteins found to be either upregulated (277) or downregulated(183) significantly (P < 0.05) in the EMT/stem-like cells afterapplying a moderated t-test17. Several known EMT marker proteins,E-cadherin (CDH1), N-cadherin (CDH2), vimentin (VIM) andfibronectin (FN1), showed significant expression level changes inthe expected directions, confirming the specificity of the proteomicprofiling (Fig. 1a).

We focused subsequent analyses on the CD90 protein, also knownas Thy1, which was among the most strongly upregulated plasmamembrane proteins (12.4-fold) following passage through an EMT.CD90 is a GPI-anchored glycoprotein and has been shown to interactwith integrins displayed by adjacent cells18,19. Its messenger RNA wasalso substantially upregulated during the EMT programs induced inHMLE cells by expression of the Twist, Snail, Slug or Zeb1 EMTtranscription factors (TFs; Supplementary Fig. 2a).

Presence of stem-like cells in the CD90hi cell populationAs human mammary epithelial cells that have undergone an EMTcan acquire stem-like properties4, we postulated that CD90 expressionwas upregulated in MaSCs and breast CSCs. We analysed the gene-expression data from two published reports and found a significantenrichment of CD90 expression in freshly isolated human mammarySC and CSC fractions in primary patient tissues across multiplespecimens20,21 (Supplementary Fig. 2b,c).

We then examined the cell-surface antigen profile of HMLE-derived cells using anti-CD90 and -CD24 antibodies. CD24 has beenfound to be mostly expressed in differentiated mammary epithelial

1106 NATURE CELL BIOLOGY VOLUME 16 | NUMBER 11 | NOVEMBER 2014


ART ICLES

Cell number

Tumour incidence from sorted HMLE-Ras cells

CD90hiCD24– CD90lo Bulk

1 × 104 cells

8 × 104 cells

2 × 105 cells

3 × 105 cells

1 × 106 cells

3/10 N.D. N.D.

10/10 N.D. 0/10

10/10 1/10 0/10

10/10 0/10 3/10

5/5 N.D. 6/10

CSC frequency 1/21,394 N.D. 1/1,328,927

0

5

10

15Lung metastasis

0.5 mmHMLER CD90hi24– Bulk HMLER

HMLER CD90hi24– Bulk HMLER

0.5 mm

Mac

ro m

ets

per

lung ∗∗∗

00 7 14 21 28 36 43 50

50

100

150

200

250 CD90 (also known as Thy1) knockdown

DaysTu

mou

r si

ze (m

m2 )

Thy1 shRNA HMLER90hi

∗

∗

∗∗

∗∗

1 mm

0

0

0.2

0.4

0.6

Pro

bab

ility

0.8

1.0ER– breast cancer ER+ breast cancer

5

Low

HR = 1.99 (1.44 – 2.73)Logrank P = 1.7 × 10–5

HR = 1.31 (1.06 – 1.61)Logrank P = 0.013

High

10Time (yr)

15

1 mm 1 mm 1 mm

CD90hiCD24– Bulk

0

0

0.2

0.4

0.6

Pro

bab

ility

0.8

1.0

5 10Time (yr)

15

LowHigh

a b

d

c

e f g

Figure 2 Enrichment of CD90 protein on the cell surface of EMT/stem-like cells. (a) HMLE-Ras cells were FACS-separated and then injectedorthotopically into NOD-SCID mice in a limiting dilution assay. The tumourinitiation rates are shown, with the estimated CSC-frequency shown for thebulk and the CD90hiCD24− populations. N.D., not determined. (b,c) H&Estaining comparing xenograft tumours (b) and lungs (c) of mice receivinginjections from the bulk and CD90hiCD24− HMLER cells. Black arrows (c)indicate macro-metastasis from the CD90hiCD24− tumours. (d) The numbersof macro-metastases per lung (n=5 mice). (e,f) Correlation of CD90 level

in patient tumours with relapse-free survival. High CD90 expression (redline) was significantly correlated with poor prognosis in ER− (n= 457,P=1.7×10−5) and ER+ (n= 1413, P = 0.013) breast cancer patients.The plots were generated using http://kmplot.com (probe: 213869_x_at;ref. 58). (g) HMLER90hi cells with or without CD90 shRNA were injectedorthotopically into mice and the tumour sizes were measured weekly duringa course of 50 days. n=10 mice. Knockdown of CD90 by the shRNA wasconfirmed by flow cytometry analysis (Supplementary Fig. 2h). All error barsshown indicate s.e.m. ∗P<0.05; ∗∗P<0.01; ∗∗∗P<0.001 (Student’s t-test).

cells21. Flow cytometry analysis revealed that 99% of HMLE-Twistcells and 18% of HMLE-Snail cells were CD90hiCD24−, whereas theHMLE-vector control cells contained only 0.1% cells in the samecompartment (Fig. 1b), indicating that forced expression of either theTwist or Snail EMT-TFs led to substantial increases in CD90hiCD24−

cells. The CD90hiCD24− cells from the three cell lines formedmammospheres22 with a far higher efficiency than cell populationswith lower CD90 expression (Fig. 1c,d), providing further support forthe association of CD90 expression with the MaSC state.

We subsequently undertook to measure the association of tumour-initiating ability with the CD90hiCD24− state in HMLE cells thathad been transformed with the H-RasV12 oncogene (HMLER); thisability represents the operational definition of CSCs. Sorted HMLERcells were implanted orthotopically at limiting dilutions into NOD-SCID mice. The CD90hiCD24− HMLER cells initiated tumours at a

60-fold higher efficiency than the bulk HMLER cells and the tumourswere more invasive and metastatic; the CD90lo populations rarelyformed tumours (Fig. 2a–d). Similar results were observed when thesorted cell populations were implanted orthotopically into Nude mice(Supplementary Fig. 2d). CD90hi cells from theMB-MDA-435 humanbreast cancer cell line also formed larger and more aggressive tu-mours when compared with CD90lo cell populations (SupplementaryFig. 2e–g). Consistently, the expression level of CD90 is negatively cor-relatedwith patient survival, especially in ER− breast cancer (Fig. 2e,f).Taken together, these data demonstrated that the CD90marker servedto enrich for CSCs in carcinoma cells of mammary origin.

Physical interaction between CD90hi CSCs and TAMsReducedCD90 expression by short hairpinRNA (shRNA) knockdownresulted in delayed tumour onset and decreased tumour size (Fig. 2g)



ART ICLES

CTK CD90 CD68

CTK CD90 CD68

HMLER

HMLER

CD90hi CSCs

Monocytes

Stroma

Tumour

Stroma

DAPI CD90+ CSCs F4/80+ TAMs

1.5 ± 0.31; n = 6ratio

CD90med (CD90 shRNA)

0.45 ± 0.23; n = 3

CD90lo non-CSCs

0.071 ± 0.03; n = 4

0.5 mm

10 µm

10 µm

10 µm 10 µm

10 µm 10 µm

0.5 mm 0.5 mm

CD90hi CSCsa

b

e

f

i

g

h

c d

Distance fromedge (a.u.)

Sig

nal i

nten

sity

Sig

nal i

nten

sity

1 2 3 4

Distance fromedge (a.u.)

1 2 3 4

25

30

35

40

F4/80+ TAMs

10

15

20

25

1–11–22–12–23

CTK+ CD90– non-CSCs and TAMsR = –0.1204, P = 0.603

CTK– CD90+ CSCs and TAMsR = 0.4386, P = 0.002

CTK– CD90+ CSCs

CTK+ CD90+

CTK+ CD90– non-CSCs

Linear (CTK– CD90+

CSCs)

Patient tumour sections

Cel

l num

ber

cou

nted

22 32 41 420

00.5

0.5

1.0

1.0

1.5

1.5

Log10 CD68+ per field

Log 1

0 ca

ncer

cel

ls p

er fi

eld

2.0

2.0

2.5

2.5

3.0

3.0

20

40

60

80

100

In contact with TAMs

No adjacent TAMs

24% 50% 51% 36%

CD90hi CSCs

CD68+ TAMs CD90+ CSCs DAPI

41

Core 5

Core 7

Core 1

80

2

4

6

8

10 CD90– non-CSCs

Patient tumour sections

Per

cent

age

of c

ells

adja

cent

to

TAM

s

HMLER +thp1-90hi HMLER +thp1-shTHY1 HMLER +thp1-THP1 +90lo

36%

54%

64%

30%

90%

6%

GFP

BrCA032 (ER+) BrCA042 (ER–)

RFP

Tumour

105

105

104

104

103

103

102

0

102–103

PE

–Tex

as R

ed-A

GFP-A

105

105

104

104

103

103

102

0

102–73

PE

–Tex

as R

ed-A

GFP-A

105

105

104

104

103

103

102

0

102–63

PE

–Tex

as R

ed-A

GFP-A

Figure 3 CD90 serves as an anchor for the adherence of monocytes andmacrophages. (a,b) THP1–RFP monocytes were co-cultured with monolayerHMLER–GFP cells. Monocytes that did not adhere to the HMLER monolayerwere washed away after 1 h and the adherent cells were imaged andthen quantified by flow cytometry based on GFP and RFP expression.Representative images (a) and flow cytometry profiles (b) are shown. Theratios of the adhered monocytes to the HMLER cells were calculated fromthe flow-cytometry quantification in b; data are presented as mean ± s.e.m.(c) Immunofluorescence staining of human CD90 (red) and F4/80 (green)in HMLER90hi xenograft tumours. The dotted lines delineate the borderbetween the tumour cells and stroma. Scale bar, 0.1mm. (d) Quantificationof CD90 signal and F4/80 signal intensity from tumour edge to centre. a.u.,arbitrary unit. n=5 tumours. (e) ×60 confocal images of primary patient

breast tumour sections co-stained for CD90+ CSCs (red) and CD68+ TAMs(green) showed juxtaposed CSC–TAM pairs. (f) Numbers of CD90high CSCswith or without adjacent TAMs were counted from 3 to 6 representative fieldsof ×20 or ×25 images (n=4 patients). (g) The percentages of CD90−CTK+

cancer cells with adjacent TAMs. Error bars indicate s.e.m. Source dataare provided in Supplementary Table 7. (h) The numbers of CD90+CTK+,CD90+CTK−, CD90−CTK+ cancer cells and CD68+ macrophages in the samemicroscopic field (0.6mm2) were quantified from 25 core biopsy sections.The number of CD90+CTK CSCs correlates with the number of CD68+ TAMs.The CD90+ endothelial cells and fibroblasts were excluded from the cancercell counts on the basis of morphology. (i) Typical images showing infiltrationof CD68+ TAMs into the CD90+ CSC-rich areas but not the CD90−CTK+

cancer cell islands.



ART ICLES

by the HMLER90hi cells, suggesting that CD90 has a functional rolein tumour growth. CD90 has been shown to be expressed on activatedendothelial cells, where itmediates the adhesion ofmonocytes throughits counter-receptor MAC1 (integrin alphaM/betaV; ref. 19). Wespeculated that CD90 also serves as a physical anchor used bycarcinoma stem cells to tether monocytes and derived macrophages.To test this notion in cell culture, we employed a human monocyticcell line THP1 (ref. 23). GFP-expressing HMLE-Ras cells were sortedinto CD90hiCD24− (HMLER90hi) and CD90loCD24+ (HMLER90lo)populations (Supplementary Fig. 2h) and expanded in monolayerculture. RFP-expressing THP1 monocytes, normally propagated insuspension, were then added above the cell monolayers. We foundthat the RFP+ monocytes adhered in large numbers to the GFP+

monolayer of HMLER90hi cells, but rarely stayed attached to theHMLER90lo cells (a 22-fold difference; Fig. 3a,b). shRNA-mediatedknockdown of CD90 in HMLER90hi cells reduced the adherentmonocytes by threefold, confirming the requirement of CD90expression on the surface of the stem-like cells for efficient physicalinteraction with the monocytes (Fig. 3a,b middle panels).

Association of TAMs with CD90hi carcinoma cells in livingtissuesIn the xenograft tumours formed by HMLER90hi cells, CD90expression was most prominent at the outer edge of tumour sections,where a high density of infiltrating TAMs was observed intermingledwith CD90hi cancer cells (Fig. 3c,d), suggesting that CD90-mediatedadhesion of TAMs to CSCs also occurs in vivo.

CD90 expression has been detected in primary human breastcarcinomas on invading single cells or cell clusters24; in metastasesthe CD44+CD90+ subpopulation represents a major phenotype25.Although a high TAM density in primary breast cancer has beencorrelated with poor prognosis26–28, a CSC niche involving TAMs hasnot been demonstrated microscopically in patient tumour tissues. Weundertook to verify direct contact between TAMs and CD90hi CSCsin patient breast cancer tissue by immunofluorescence analyses withanti-CD90 and -CD68 (a cell-surface protein expressed largely bymacrophages) antibodies. Juxtaposition of CD68+ TAMs with single-invading CD90hi CSCs was readily detected at the tumour–stromainterface (Fig. 3e; Supplementary Fig. 3a). Indeed, 24–51% of observedCD90hi cancer cells showed the presence of TAMs in close contact;in contrast, very few (0–5%) CD90− cytokeratin+ (CTK+) cells werelocalized adjacent to TAMs (Fig. 3f,g; Supplementary Fig. 3b). Thisdirect visualization of the CSC–TAM juxtaposition in patient samplesprovided further support for the notion that CSCs and TAMs engagein direct contact in human breast tumours, where the latter cells helpto form a niche for the CSCs.

We then expanded our analysis to include 25 core biopsysamples from patient tissues with invasive ductal carcinoma obtainedduring initial diagnostic core biopsies. Most CD90+ carcinoma cellsshowed reduced levels of cytokeratin expression, consistent with theirpassage through an EMT (Supplementary Fig. 3c). The number ofCD68+ TAMs per microscopic field correlated with the number ofCD90+CTK− CSCs (r = 0.44, P= 0.002) but not with CD90−CTK+

non-CSCs (r =−0.12, P = 0.60; Fig. 3h). This indicated that TAMsoften infiltrate CD90+ CSC-rich areas, allowing the juxtacrinesignalling to occur. Accordingly, we could observe intermingling

of CD68+ TAMs with CD90+ CSCs, but the macrophages rarelyinfiltrated into islands of CD90−CTK+ non-CSCs (Fig. 3i). Finally,consistent with this, analysis of the TCGA invasive breast carcinomadata set revealed that CD90 expression in breast tumours positivelycorrelated with CD68 expression (Supplementary Fig. 3e).

Monocytes/macrophages facilitate tumour outgrowth by theCSCsMonocytes and macrophages in tumours have great functional plas-ticity12,29. To test whether the CSC–TAM interaction that we ob-served is important to tumour formation, we depleted endogenousmacrophages in the mammary fat pads at the time of implantationby co-injecting the HMLER90hi cells with clodronate liposomes14.Nude mice were used in this and subsequent experiments instead ofNOD-SCID hosts because of the documented defect in macrophageactivation in the latter. Strikingly, tumour initiation was almost com-pletely blocked by macrophage depletion, suggesting that endoge-nous macrophages in the tumour microenvironment are critical forthe survival and proliferation of implanted CSCs (Fig. 4a). To ex-amine the opposite effect, we co-injected HMLER90hi cells witheither primary human monocytes or mouse TAMs isolated fromHMLER90hi xenografts (Supplementary Fig. 4a). The admixture ofeither human monocytes or mouse TAMs resulted in higher tumourincidence and size, indicating that they promoted tumour outgrowthby CSCs (Fig. 4b).

The TAMs within tumours arising from CSCs showed lowexpression of Nos2 (M1 polarization), and high expression of Arg1,MRC1, FIZZ1 and VEGF (M2 polarization30), which resembled theexpression patterns of tumour-promoting TAMs (ref. 31) isolatedfrom mammary tumours in the MMTV-PyMT mice (SupplementaryFig. 4b). Moreover, when analysing the CD68+ TAMs from the patientsamples described above, we found that these cells were predominantlyCD163+HLADR−/dim (Supplementary Fig. 4c), indicative of an M2,tumour-promoting phenotype32.

Various types of tumour-infiltrating immune cells, includinglymphocytes, have been shown to modulate the function of TAMs(ref. 33). Therefore, we examined the effect of TAMs on CSC activitiesin immunocompetent mice. The MMTV-PyMT mice are knownto develop mammary tumours that harbour CD90+ CSCs, whichefficiently metastasize to the lungs when injected intravenously10. Wesorted freshly isolated mouse mammary carcinoma cells based onCD90 expression from the MMTV-PyMT-driven metastatic tumoursand then injected the resulting CD90+ and CD90− populations,with or without TAMs, orthotopically into wild-type syngeneic hosts.All groups developed tumours of similar weight (SupplementaryFig. 4d); lymphocyte infiltration was readily detected within thetumours, resembling spontaneously arising MMTV-PyMT tumours(Supplementary Fig. 4e,f). However, tumours that arose from CD90−

non-CSCs (with or without co-injected TAMs) were largely necroticand contained cysts filled with fluid (Supplementary Fig. 4g,h), withlow numbers of proliferating cells (Supplementary Fig. 4i,j) and nometastasis in the lungs (Fig. 4c,d). CD90+ CSCs injected on their ownresulted in tumours with less necrosis and more proliferating cells butyielded very few lung metastases. In contrast, the CD90+ CSCs withco-injected TAMs resulted in tumours that contained little necrosis,weremore proliferative, andmetastasized to lungwith a 22-fold higher



ART ICLES

1 mm

Tumour incidence from secondary transplant

200,000 100,000

0

1

2

3Macrophage depletion

a

d

e g

ji

f

h

b c

Cells injected

CD90– non-CSCs CD90– non-CSCs + TAMs CD90+ CSCs CD90+ CSCs + TAMs

Tum

our

wei

ght

(g) PBS

liposome

Clodronateliposome

∗∗∗ ∗

Tum

our

wei

ght

(g)

0.0

0.5

1.0

1.5P < 10–4

P = 0.02

CSC+huMono

CSC+TAM

CSC CD90– CD90–

+TAMsCD90+ CD90+

+TAMs

0

5

10

Lung metastases

Met

asta

tic in

dex

NS

∗∗∗∗

NS

0/13 0/14 2/14 10/14

0 10 20 30 40 500

50

100CD90hi CSC only

Days

Tum

our-

free

sur

viva

l (%

)

0 10 20 30 40 500

50

100

Days

Tum

our-

free

sur

viva

l (%

)

0 10 20 30 40 500

50

100

Days

Tum

our-

free

sur

viva

l (%

)

400,000200,000100,00050,00010,000

400,000200,000100,00050,00010,000

CD90hi CSC + TAM co-injection

CD90– non-CSCs

400,000400,000 + TAM

First injection CSCs alone CSCs + TAMs

Second injection Tumour cells Tumour cells

3 × 104 3/8 10/10

1 × 104 1/8 8/10

3 × 103 0/8 6/8

CSC frequency 1/72,700 1/4,231

200,000200,000 + TAM

0400,000 200,000 100,000 50,000 10,000

0.5

1.0

1.5

Numbers of CSCs injected

Tum

our

wei

ght

(g)

CSCs onlyCSCs + TAMs

∗

∗

∗

Tum

our

sphe

res

per

8 w

ells

CSC tum

ours

CSC + T

AM tu

mou

rs–10

0

10

20

30

40

50

400 cells per well100 cells per well

1 mm 1 mm 1 mm 1 mm

Figure 4 Macrophages enhance tumour initiation by the CD90hi CSC-likecells. (a) CD90hi CSCs were mixed with PBS liposomes or clodronateliposomes and orthotopically implanted into Nude mice (n=10 per group).Data were pooled from 2 independent experiments, each assessing 5 miceper group. The tumours were collected 50 days later and weighed. (b) 105

HMLER90hi cells were injected either alone or co-mixed with 5×104 humanmonocytes (huMono) or mice TAMs, into the mammary fat pads of femaleNude mice. n=9, 12, 12 mice (left to right). A reduced amount of Matrigelwas used for the injection mix. The tumour weight was measured after5 weeks. Data from 3 independent experiments were pooled and plotted.(c,d) Naturally arising MMTV-PyMT tumours were dissociated and tumourcells were sorted based on CD90 expression, whereas TAMs were isolated byF4/80 expression. 3×105 CD90+ or CD90− tumour cells were orthotopicallytransplanted into syngeneic wild-type Fvb mice with or without co-mixed1.5×105 TAMs. (c) The numbers of lung macro-metastases from tumour-bearing mice were quantified and normalized to primary tumour weight (pergram), shown as metastatic index. Data were pooled from 3 independentexperiments. n= 13, 14, 14, 14 mice from left to right (15 mice were

injected for each group, but not all mice developed tumours; SupplementaryFig. 4d). (d) H&E staining of the lungs of tumour-bearing mice. (e–g) CD90hi

CSCs were injected alone (e) or with admixed TAMs (f) at limiting dilutionorthotopically (n=10 mice per group). Tumour incidence was determinedby weekly palpation, and tumours were collected and weighed at 50 dayspost injection (g). Data were pooled from 2 independent experiments, eachassessing 5 mice per group. (h) CD90lo non-CSCs were injected alone or withadmixed TAMs orthotopically but no tumour formed. Data were pooled from 2independent experiments, each assessing 5 mice per group. (i,j) HMLER90hicells injected alone or together with TAMs were allowed to grow in themammary fat pads of NOD-SCID mice for 3 weeks; the tumour cells werethen collected and seeded for tumorsphere forming assay (i presents datafrom 4 technical replicates) or injected as secondary tumour transplants intorecipient NOD-SCID mice (j) to measure tumour-initiating capability. Datashown are from one experiment out of two biological replicates; source dataare provided in Supplementary Table 7 (i,j). Error bars indicate s.e.m. NS,not significant; ∗P<0.05; ∗∗∗P<0.001; ∗∗∗∗P<0.0001 (Student’s t-test fora,b,g and one-way ANOVA for c).



ART ICLES

efficiency than CD90+ CSCs injected alone. These data supportedthe notion that the co-injected macrophages promoted more robustprimary tumour growth and enhanced the metastatic CSC activity ofthe CD90+ carcinoma cells in the context of an intact immune system.

We noted that both the CD90+ and CD90− MMTV-PyMTcarcinoma cells initiated primary tumours efficiently. This isprobably due to the fact that the tumour cells were isolated fromspontaneously arising MMTV-PyMT tumours at the stage ofmetastatic dissemination, when most of the carcinoma cells hadevolved to be rapidly proliferating following weeks of robusttumour growth.

Effects of admixed macrophages on tumour initiationWe reasoned that the direct contact of macrophages with theCSCs might allow the TAMs to provide signals important for themaintenance of their CSC state. To determine whether TAMs enhancetumour initiation rates and thus enrich for CSCs we performedtumour xenograft experiments by implanting the carcinoma cells inlimiting dilutions. Co-injecting macrophages with the CD90hi CSCsresulted in earlier tumour onset (Fig. 4e,f) as well as higher tumourincidence and burden compared with CSCs injected alone (Fig. 4g);in contrast, CD90lo non-CSCs did not initiate tumours either with orwithout admixed TAMs (Fig. 4h).

Tomeasuremore precisely the frequency of CSCs from the tumourswith admixed TAMs, we injected CSCs in the presence or absenceof TAMs into mouse mammary fat pads and collected the tumoursthree weeks later when they were no larger than 0.2 g (SupplementaryFig. 4k). The cancer cells from the TAMs+CSCs group formed moretumorspheres and initiated tumours with a 17-fold higher efficiencyin a secondary transplantation experiment than those from the CSCs-only injected tumours (Fig. 4i,j). Taken together, these data confirmeddirectly the ability of TAMs to boost the representation of CSCsin these carcinoma cell populations, ostensibly by enabling CSCs tomaintain their residence in the stem cell state.

Monocyte/macrophage-stimulated cytokine production in thestem-like cellsWe then sought to dissect themolecularmechanism for the interactionbetween CSCs and monocytes/macrophages. As cytokines are knownto mediate the interactions between cancer cells and TAMs (ref. 29),we measured the expression of various cytokine mRNAs in bothHMLER90hi cells and THP1 monocytes that had been separatedby fluorescence-activated cell sorting (FACS) after 3 h of co-culture(Supplementary Fig. 5a). Widespread changes in cytokine expressionwere already apparent in both cell populations within this time period(Supplementary Tables 2 and 3). Strikingly, we observed rapid androbust induction of mRNAs encoding IL-6, IL-8 and GM-CSF in theHMLER90hi cells, whose levels far exceeded the cytokines producedby the co-cultured monocytes.

The mRNA levels of IL-6, IL-8 and GM-CSF rose abruptlyfollowing monocyte co-culture and persisted for at least 6 h (Fig. 5a).HMLER90lo cells, used as controls, showed only a modest andtransient increase in these cytokines in response to monocyteco-culture. Importantly, the induction of these cytokines in theHMLER90hi cells was completely blocked when they were separatedfrom the monocytes by a cell-impermeable insert, indicating that

the observed signalling between the two cell types required directcell–cell contact, that is, juxtracrine signalling. Virtually no cytokineswere detected in the conditionedmedium by enzyme-linked immune-sorbent assay (ELISA) from mock-treated HMLER90hi cells incontrast with the robust increases of cytokine levels after 4 h of co-culture (Fig. 5b); membrane-separated indirect co-cultures yieldedonly marginal if any increases in the three cytokines. Knockdownof CD90 by shRNA resulted in a twofold to threefold reductionin cytokine mRNA induction in the HMLER90hi cells (Fig. 5c),probably due to the reduced number of monocytes adhering tothese cells.

We also noted that the induction of IL-6, IL-8 and GM-CSF inthe stem-like cells did not require the actions of H-RasV12, becausethe CD90hiCD24− cells from the non-transformed HMLE-Twist andHMLE-Slug stem-like cells also responded to monocyte co-cultureby producing these cytokines, whereas the parental epithelial HMLEcells, used here as controls, showed only a marginal upregulation ofthe cytokines (Fig. 5c). In addition, cytokine induction in the CD90hi

cells was also observed when primary mouse macrophages or primaryhuman monocytes were used instead of THP1 monocytes (Fig. 5d,e),which indicated that the above observations applied to a varietyof monocytes/macrophages, and that the signalling between mousemacrophages and the human HMLE-derived CD90hiCD24− cells wasnot compromised by inter-species signalling incompatibilities.

IL-6 and IL-8 have been shown to induce and maintain theCSC state in autocrine and paracrine fashions9,34,35. In HMLER90hicells, individual knockdown of IL-6 or IL-8 led to cessation of cellproliferation in culture, whereas the proliferation of epithelial HMLE-Ras cells was unaffected by the knockdowns (Supplementary Fig. 5b,c).In addition, IL-8 overexpression or treatment with recombinant IL-6and/or IL-8 resulted in more efficient tumorsphere formation by theHMLER90hi cells, indicating that IL-6 and IL-8 can indeed act on theHMLER90hi cells to promote the expansion of CSCs (SupplementaryFig. 5d–f). GM-CSF is known to promote the differentiation of TAMsinto tumour-supporting phenotypes12,33, and may act in a paracrinefashion to perpetuate the CSC–TAM interaction. The three cytokinestherefore may further reinforce and sustain the stem cell state of theHMLER90hi CSCs.

Role of EphA4 in mediating signalling activation in thestem-like cellsWe undertook to determine the molecular mechanisms that wereresponsible for the marked induction of cytokines following exposureof normal and neoplastic MaSCs to monocytes. CD90 lacks anintracellular domain and hence is poorly equipped to transduce signalsacross the plasma membrane18. As an alternative, we consideredthe well-documented CSF1–EGF paracrine loop between TAMs andbreast cancer cells31,36; however, the expression of EGFR is stronglydownregulated during passage through an EMT (refs 6,37,38),rendering it unlikely to mediate the response of CSCs to TAMs.

We therefore searched for other cell-surface receptorcandidates that could mediate the juxtacrine signalling fromour proteomics analysis. Three receptor tyrosine kinases (RTKs)—ROR1 (a pseudokinase39), PDGFRβ and EphA4—were stronglyupregulated in the stem-like cells (Fig. 1a, Supplementary Table 1and Supplementary Fig. 6a); among these, only EphA4 is known



ART ICLES

0

10

20

30

40

0 2 4

HMLE-Ras 90lo HMLE-Ras 90hi HMLER 90hi-memb

Hours in co-culture Hours in co-culture Hours in co-culture

6 0 2 4 6 0 2 4 6

IL6a

c d e

b

0

30

60

90

120

150 IL8

0

20

40

60

80

100 GM-CSF

IL-6 IL-8 GM-CSF0

20

40

60

80

500600700

Cytokine secretion (4 h)

Con

cent

ratio

n (p

g m

l–1)

HMLER90hi HMLER90hi+monocytesHMLER90hi-membrane-monocytes

Cytokine mRNA induction (3 h)

mR

NA

fold

-cha

nge

(nor

mal

ized

to

cont

rol)

HMLE

HMLE

-Slug

HMLE

-Slug

90hi

HMLE

-Twist

HMLE

R90hi

HMLE

R90hi-

shCD90

HMLE

HMLE

-Twist

HMLE

R90hi

HMLE

R90lo

HMLE

R90hi

+lkB-m

ut

HMLE

HMLE

-Twist

HMLE

R90hi

HMLE

R90lo

HMLE

R90hi

+lkB-m

ut0

20

40

60

80

100 IL-6IL-8GM-CSF

THP1

IL-6IL-8GM-CSF

IL-6IL-8GM-CSF


Mouse intraperitoneal-exudemacrophages

mR

NA

fold

-cha

nge

(nor

mal

ized

to

cont

rol)

0

10

20

30

40Cytokine mRNA induction (3 h)

Human peripheral blood monocytes

mR

NA

fold

-cha

nge

(nor

mal

ized

to

cont

rol)

0

10

20

30

40

50

Figure 5 Monocytes induce cytokine production in the CSCs on contact.(a) Induction of cytokine mRNA in the HMLER90hi or HMLER90lo cellswhen co-cultured with the THP1 monocytes for the indicated times.HMLER90hi-memb, an indirect co-culture where the two cell types wereseparated by a cell-impermeable insert. Data from one experiment areshown as averages of three technical replicates. Two biological replicateswere performed. (b) ELISA measurement of the secreted cytokinesfrom the co-culture. n= 3 independent experiments. Error bars indicates.e.m. (c) Cytokine mRNA fold-change stimulated by THP1 monocytes

(relative to corresponding no-monocytes control) in the epithelial HMLE,mesenchymal HMLE-Twist, HMLE-Slug CD90hiCD24−, HMLER90hi cellsand HMLER90hi with CD90 shRNA. (d,e) Cytokine mRNA fold-changes inthe HMLE-derived cells stimulated by mouse primary macrophages (d), orhuman blood monocytes (e). The HMLER90hi-IκB-mut cells lacking NF-κBactivation showed no cytokine production. Typical data are shown from oneexperiment out of 3 (a,b) and 2 (c,e) biological replicates; only technicalreplicate data are plotted. Source data are provided in SupplementaryTable 7.

to mediate contact-dependent signalling following binding withplasma-membrane-associated Ephrin ligands displayed on closelyapposed cells40. We noted that whereas peptides shared by severalEph RTKs showed little change in levels, MS analysis of the peptidesunique to EphA4 revealed its upregulation by over 20-fold followingpassage through an EMT (Methods).

Wemeasured the activation of theRTKs on contact withmonocytesby immunoblotting after removal of monocytes following co-culture(Supplementary Fig. 6b). Only EphA4 showed a robust increase inactivation in the stem-like HMLER90hi cells and HMLE-Twist cellsfollowing direct monocyte co-culture (Fig. 6a,b and SupplementaryFig. 6c).

A KYL peptide has been shown to specifically inhibit EphA4activation41. Treatment of co-cultures with the KYL peptide,but not the KYL-P7A control peptide41, blocked 90% of EphA4phosphorylation in the HMLER90hi cells and significantly reducedthe secretion of the three cytokines (Fig. 6c,d). Consistently, shRNAknockdown of EphA4 in the HMLER90hi cells hindered cytokinemRNA induction by monocytes (Supplementary Fig. 6d,e). Taken

together, these data demonstrated the involvement of EphA4 inmonocyte-stimulated cytokine production in the CSCs. This wasfurther supported by our observation that knockdown of EphA4 inthe CSCs led to a fourfold reduction in the size of resulting tumours(Fig. 6e).

As EphA4 activation requires direct engagement of its liganddisplayed on the surface of monocytes, it is likely that CD90 physicallyanchors the monocytes to the HMLER90hi cells, thereby providingthe physical association that is essential for the Ephrin ligand-receptor juxtacrine signalling. As expected, we observed that EphA4phosphorylation in the CD90 knockdown cells was 40% lower thanin the control cells when co-cultured with monocytes (SupplementaryFig. 6f).

The EphA4 receptor can promiscuously bind to both types ofhuman Ephrin ligand, including five glycosylphosphatidylinositol(GPI)-linked ephrin-A ligands and three transmembrane ephrin-Bligands40. Pretreatment of themonocyteswith recombinant PI-specificPLC before co-culture resulted in decreased phosphorylation of theEphA4 receptor, indicating that the ephrin ligand(s) displayed by the



ART ICLES

Co-culture (min)

a

d e

f g

b c

pY-PLCγ1 F4/80 DAPI

10 μm

10 μm

10 μm

52

38

10276

150225

HMLE

R90hi

Luc s

hRNA

PKCδ shR

NA1

PKCδ shR

NA2

PLCγ1

shRNA2

PLCγ1

shRNA1

Mr (K)

β-actin

PKCδ

PLCγ1

HMLER90hi

EphA4-pY779

EphA4

GFP

RFP

THP1

HMLER90hi

10 30 – 10 10 30 3030 30 – 30 30

0.33 0.09 0.03 2.09 1.97

0.67 1.22 1.34 1.04 0.95

0.79 0.68 0.44 1.59 1.57

0.59 1.16 1.39 1.55 2.12

0.44 0.04 0.00 1.81 1.87

0.48 0.43 0.29 0.77 1.13

1.80 1.31 1.05 3.39 4.28

0.83 1.15 1.39 1.66 2.59

Co-culture (min)– – – – 20 KYL-P7A (mM)

10 20 – – – KYL peptide (mM)

PLCγ1-pY783

PLCγ1

PKCδ -pY311

PKCδ

pY416-Src

Src

EphA4-pY779

EphA4

GFP

PLCγ1-pY783

PLCγ1

PKCδ -pY311

PKCδ

Src-pY416

Src

30' d

irect

30' m

emb

rane

No

mon

ocyt

es

Co-culture with KYL-P7A


Con

cent

ratio

n (p

g m

l–1)

IL-6 IL-8 GM-CSF0

20

40

60

100

300

500

Co-culture HMLE

R90hi

Vecto

r con

trol

EphA4

shRNA3

EphA4

shRNA4

0

0.5

1.0

1.5

EphA4 knockdown

Tum

our

wei

ght

(g) ∗∗∗∗

∗∗∗∗

NS


mR

NA

fold

-cha

nge

(nor

mal

ized

to

cont

rol)

Paren

tal

Lucif

eras

e shR

NA

PKCδ shR

NA1

PKCδ shR

NA2

PLCγ1

shRNA1

PLCγ1

shRNA2

0

20

40

60

IL-6IL-8GM-CSF

h

Co-culture with KYLHMLER90hi

Figure 6 EphA4 mediates signal transduction in the CSCs on stimulationby the monocytes. (a) Phospho- and total-protein levels of EphA4 as wellas downstream proteins PLCγ1, PKCδ and Src in the two co-cultured celltypes were detected by western blot. GFP and RFP were detected only in thecorresponding cell populations with minimum cross-contamination. Numbersabove each lane indicate minutes of co-culture. Phospho-EphA4 was detectedwith an antibody that non-discriminatively recognizes the phosphorylatedforms of EphA3, 4 and 5. (b) Indirect co-culture (membrane) with monocytesdid not lead to activation of EphA4 and downstream proteins in the CSCs.(c) Treatment of HMLER90hi cells with an EphA4-specific inhibitor peptideKYL before and during co-culture with monocytes resulted in decreasedphosphorylation. A scrambled peptide (KYL-P7A) served as the control. Thelargely diminished phosphorylation level with KYL treatment indicated thatthe signals detected by the phospho-EphA3/4/5 antibody were mostly fromactivated EphA4. The numbers below each panel are the intensities of thebands quantified using Fiji software and normalized to GFP (bottom panel)of the corresponding samples. (d) Cytokine secretion from co-culture treated

with KYL or KYL-P7A was measured by ELISA. Technical replicate data areplotted. Source data are provided in Supplementary Table 7. (e) 1×105

HMLER90hi parental cells or HMLER90hi with luciferase shRNA or EphA4shRNA were orthotopically implanted into NOD-SCID mice. n=8 mice (datawere pooled from 2 independent experiments, each assessing 4 mice pergroup). The tumours were collected 40 days after implantation and weighed.NS, not significant; ∗∗∗∗P<0.0001 (one-way ANOVA). Error bars indicate s.d.(f) Western blots measuring levels of knockdown with 2 pairs of shRNA eachfor PLCγ1 and PKCδ in the HMLER90hi cells. (g) Cytokine mRNA induction inthe HMLER90hi cells with PLCγ1 and PKCδ knockdown. Technical replicatedata are presented as mRNA fold-changes in the HMLER90hi cells onmonocyte co-culture. Source data are provided in Supplementary Table 7.(h) Immunofluorescence staining of xenograft tumours with antibodies againstphospho-PLCγ1 (Tyr 783) and F4/80. White arrowheads point to the phospho-PLCγ1+ cells (red) surrounded by F4/80+ TAMs (green). Areas with no TAMsshowed no phospho-PLCγ1 signals (bottom panel). Uncropped images of blotsare shown in Supplementary Fig. 7.



ART ICLES

monocytes belong to the GPI-anchored type A family (SupplementaryFig. 6g). Consistently, the EphA2 receptor, which recognizes onlyephrin-A ligands, was also activated in the CSCs on co-culture withmonocytes. Given that EphA4 can bind all five ephrin-A ligands, wedid not pursue the precise identity of its cognate ephrin-A ligand(s)displayed by the monocytes/macrophages.

Intracellular signalling activated by the EphA4 receptorIn subsequent work, we sought to characterize the downstreamsignalling pathway of EphA4 activation. EphA4 has been shown tobind and phosphorylate phospholipase Cγ1 (PLCγ1; ref. 42), whichcan activate a variety of protein kinase C (PKC) enzymes throughits production of diacylglycerol43 (DAG). In addition, Src familykinases have also been shown to mediate downstream signallingof EphA4 (ref. 44). Indeed, we observed strong phosphorylation ofPLCγ1 (Tyr 783), PKCδ (Tyr 311) and Src (Tyr 416) in the CSC-likeHMLER90hi cells in response to co-cultured monocytes (Fig. 6a,b).Activation of these proteins was greatly reduced when treated with theKYL peptide (Fig. 6c).

PKCδ is known to mediate degradation of IκB, allowing NF-κB totranslocate into the nucleus45, which can then drive the transcriptionof various genes including IL-6, IL-8 and GM-CSF (refs 46–48).Indeed, knockdown of PLCγ1 and PKCδ in the HMLER90hi cellsresulted in a 2–10-fold reduction in the induction of the three cytokinemRNAs (Fig. 6f,g). Consistently, phosphorylation of PLCγ1 couldbe readily detected in the tumour cells adjacent to TAMs in theHMLER90hi tumour xenografts (Fig. 6h).

NF-κB activation in the CSC-like HMLER90hi cellsWe also undertook to examine in more detail the involvement ofthe PKCδ-stimulated NF-κB signalling in the HMLER90hi cells. Theexpression of an NF-κB super-repressor (IκBα S32A/S36A; ref. 49)in the HMLER90hi cells blocked the cytokine mRNA inductionin these cells following co-culture with monocytes or macrophages(Figs 7a and 5d,e). Secretion of the cytokines from the co-culturedcells was also reduced by more than tenfold when the NF-κB super-repressor was expressed in the HMLER90hi cells (Fig. 7b), whichfurther demonstrated that the three cytokines were largely secretedby the HMLER90hi cells and not by the monocytes. Consistently,we observed a time-dependent nuclear accumulation of the NF-κBsubunits p65 and, more strikingly, p50, in the CD90hi cells followingco-culture with the monocytes (Fig. 7c).

More importantly, human monocytes or TAMs failed to promotetumour initiation or growth from the NF-κB super-repressor-expressing HMLER90hi cells, indicating that NF-κB activation andcytokine induction in the CSCs are functionally essential for thetumour-promoting effects of the TAMs (Fig. 7d). In addition, thetumours from co-injected CSCs and TAMs had substantially fewerapoptotic cells, consistent with the well-known roles of the NF-κBpathway and Src in conferring resistance to apoptosis (Fig. 7e).

The Twist EMT-TF has been shown to mediate IL8 transcriptionin cooperation with the RelA NF-κB subunit50. Indeed, we detectedthe association of the p50 NF-κB subunit with the endogenous Twistprotein in the HMLER90hi cells by co-immunoprecipitation (Fig. 7f).Using chromatin immunoprecipitation (ChIP), we found that Twistwasmoderately enriched close to the transcription start site of IL6, IL8

and CSF2 (encoding GM-SCF) in the absence of monocytes, whereasno binding of p50 was observed. Following monocyte co-culture, wedetected significant increases in the binding of both Twist and p50to the same promoter regions (Fig. 7g). Hence, it seems that whenstimulated by monocytes/macrophages, the EphA4 protein on theHMLER90hi cells activates the NF-κB p50 subunit, which then entersthe nucleus and cooperates with Twist to further increase its binding tothe cytokine promoters, contributing to the robust cytokine inductionin the stem-like cells (Fig. 7h).

DISCUSSIONThe EMT program is an important component of carcinomaprogression by conferring on the cancer cells invasiveness and byenriching for stem cell properties2–4. Here, we have shown that EMT-induced stem-like humanmammary epithelial cells (HMECs) interactwith and respond to monocytes and macrophages far more robustlythandodifferentiated, non-stem-likeHMECs. In the context of cancer,these cell–cell contact-dependent interactions provide evidence ofa physical CSC niche supported by the TAMs. Such direct contactcould be visualized in xenograft tumours as well as patient-derivedbreast tumour sections. Furthermore, we delineated the receptorsand signalling pathways mediating such interactions in the CSCs,which were confirmed using tumour cell transplantation models.The observed responses of the CSCs seem to sustain their residencein the stem cell state by self-reinforcing, positive-feedback loops(Supplementary Fig. 6h).

Intriguingly, co-injection of CD90+ MMTV-PyMT tumour cellswith TAMs into the mammary fat pads of wild-type syngeneicmice resulted in more efficient lung metastases than CD90+ CSCsinjected on their own. In addition to enhancing initial survival andproliferation of the CSCs, we suspect that the cohort of cytokinessecreted in response to the TAM–CSC interaction facilitated theinduction of tumour-supportive stroma, allowing tumour progressiontometastatic carcinoma. In contrast, when CD90+ CSCs were injectedalone into wild-type syngeneic hosts, in the absence of varioustumour-promoting and immune-suppressing cytokines produced bythe interacting CSCs and TAMs, the subsequently recruited immunecells may have adopted tumour-suppressive phenotypes33. The effectsof the immune cells on the CSC activities await further investigation.

During embryonic development, macrophages derive from theyolk sac and fetal liver before haematopoiesis, arising as theearliest cells of the haematopoietic lineage12. Among other functions,these tissue macrophages probably participate in embryonic organdevelopment by providing effectors necessary for the formation andfunction of various tissue stem cell niches. It has been reportedthat mouse MaSC function requires the continuing presence ofmacrophages in the developing mammary gland14. This suggests thatcomponents of the heterotypic interactions between normal MaSCsand monocytes/macrophages are recapitulated in the carcinomastem cells.

More broadly, the observed physical interactions betweenCSCs andTAMs may quite possibly operate in other carcinoma types. Increasedexpression of CD90 has been reported in CSCs of liver cancer, gastriccancer, colon cancer and glioma51–54. EphA4 overexpression has beenobserved in gastric cancer55 and liver cancer metastasis56. AdditionalEphrin receptor family members EphA2 and EphB2 have been shown



ART ICLES

HMLER90hi CSCs only CSCs + TAMs co-injection

100 μm 100 μm

Cleaved caspase-3

Ki67 DAPI

p105

p50

p65

GAPDH

SV40 LgT

Cytoplasm Nucleus

– 10 30 60 – 10 30 60MonoCo-culture(min)

IL-6 IL-8 GM-CSF0

20

0

Twis

t IP

p50

IP

+THP1

Probe:

–871

bp

+45

bp

–875

bp

+57

bp

–852

bp

+10

5 b

p

+THP1

IgG

IP

Twis

t IP

p50

IP

IgG

IP

2

4

6

8

0

Twis

t IP

p50

IP

IgG

IP

Twis

t IP

p50

IP

IgG

IP

2

4

6

8IL8 promoter

Control probe

IL6 promoter

Control probe

CSF2 promoter

Control probe

0

Twis

t IP

p50

IP

IgG

IP

Twis

t IP

p50

IP

IgG

IP

4

8

12

40

60

80350

450

550


Con

cent

ratio

n (p

g m

l–1)

Rel

ativ

e fo

ld e

nric

hmen

t (IP

/inp

ut)

HMLER90hiHMLER90hi +monocytes

IL-6 IL-8 GM-CSF0

20

4050

150

250Cytokine mRNA induction (3 h)a c d

b

g h

e f

Rel

ativ

e m

RN

A le

vel

(nor

mal

ized

to

HM

LER

90h

i )

HMLER90hiHMLER90hi +monocytes

HMLER90hi-IkB-mutIkB-mut + monocytes

HMLER90hi + IkB-mutIkB-mut + monocytes

100,000 cells 50,000 cells0

0.2

0.4

0.6

0.8

HMLER90hi-NFκB-SR

Tum

our

wei

ght

(g)

With huMono

Injected alone

With TAM

Monocyte/macrophage

CD11b

IL-6, IL-8, GM-CSF....

PLC-γ

PKC-δ

Twist

l-κB

NF-κB

NF-κB

EMT/stem-like cell

EphA4CD90

Ephrin

Src

Co-culture

52 NF-κBp50

Twist

Mr (K)

38

31

–

1% in

put

IgG

Twis

t

Twis

t

– – +

IP

+THP1

Figure 7 NF-κB activation is required for monocyte-stimulated cytokineproduction in the CSCs. (a,b) The mRNA levels (a) and secreted cytokineconcentrations (b) of IL-6, IL-8 and GM-CSF in the HMLER90hi parentalcells and the NF-κB super-repressor-expressing cells (HMLER90hi-IκB-mut)with or without monocyte co-culture were measured by qPCR and ELISA. n=3biological replicates (with 3 technical repeats each). Source data are providedin Supplementary Table 7. Error bars indicate s.e.m. (c) The HMLER90hicells, alone or in co-culture with monocytes (the latter were removed beforecell collection), were lysed and fractionated to obtain the cytosolic andnucleus extracts. The right-most lane (Mono) indicates the nuclear extractof the isolated monocytes from the co-culture at 60min. NF-κB subunitswere immunoblotted to show the nuclear translocation. GAPDH and SV40-LgT served as cytosolic and nucleus protein controls. (d) The NF-κB-super-repressor (SR)-expressing HMLER90hi cells were injected, either alone(circles) or co-mixed with human monocytes (huMono, triangles) or with miceTAMs (squares), into the mammary fat pads of female Nude mice. Tumour

weight was measured after 7 weeks and data from 2 independent experimentswere pooled. n=8, 12, 12, 10, 10 mice (left to right). Error bars indicates.e.m. (e) Immunofluorescence staining of tumours from HMLER90hiCSCs alone or CSCs+TAMs with antibodies against cleaved caspase 3(apoptosis marker) and Ki67 (proliferation marker). (f) The endogenousTwist proteins were immunoprecipitated from the HMLER90hi cells, andthen immunoblotted for the NF-κB subunit p50 to demonstrate complexformation. (g) Binding of both Twist and NF-κB at the promoters of IL6,IL8 and CSF2 (GM-CSF) in the HMLER90hi cells alone or in co-culture wasmeasured by ChIP–PCR. Data are presented as fold-enrichment of bindingto the specific segments relative to background average. The positions ofprobes are illustrated below the graphs. Data plotted are technical replicates.Source data from two biological replicates are provided in SupplementaryTable 7. (h) A depiction of the signalling transduction pathway in a CSC whenstimulated by a monocyte/macrophage. Uncropped images of blots are shownin Supplementary Fig. 7.



ART ICLES

to be upregulated in glioblastoma tumour-propagating cells and incolon CSCs (refs 11,57). Accordingly, CD90 and Eph receptor(s) maywellmediate interactions of CSCs andTAMs in a variety of carcinomasto maintain and reinforce the CSC state. Indeed, such interactionsmay one day suggest potential targets for therapeutic attackon CSCs. �

METHODSMethods and any associated references are available in the onlineversion of the paper.

Note: Supplementary Information is available in the online version of the paper

ACKNOWLEDGEMENTSWe thank L. Ma and T. Shibue for critical reviews of the manuscript, R. Bronsonfor evaluating the histopathology, D. R. Mani and P. Thiru for statistical analysesand C. Baty for assistance in imaging. We also thank A. Oberle, J. Karlsson andE. Procopio for technical assistance. This research was supported by the MITLudwig Center for Molecular Oncology and by grants from the Breast CancerResearch Foundation and National Institutes of Health (NIH; R01-CA078461, P01-CA080111 and U54-CA163109) to R.A.W. Additional support comes in part fromthe Department of Defense (BC032981 and BC044784), NIH (P30CA047904), theHillman Foundation and the Glimmer of Hope Foundation to V.S.D. This workwas also supported in part by the Broad Institute of MIT and Harvard, and bygrants from NIH (U24CA160034) to S.A.C. R.A.W. is an American Cancer Societyand D. K. Ludwig Foundation Cancer Research Professor. Postdoctoral fellowshipsupport for H.L. was from the Cancer Research Institute (New York) and the MITLudwig Center for Molecular Oncology.

AUTHOR CONTRIBUTIONSH.L. designed the project, conducted experiments and analysed data under theguidance of R.A.W. K.R.C. and S.A.C. performed the proteomics profiling anddata analysis. V.S.D. and R.B. performed patient sample staining, imaging andquantification. W.L.T. performed gene microarray, chromatin immunoprecipitationand data analysis. H.L. and X.Y. performed immunofluorescence and imaging ofmouse tumour tissues. J.F. contributed to signalling pathway analysis and E.N.E.provided technical expertise. F.R. performedmouse surgeries.H.L. andR.A.W.wrotethe manuscript. All authors critically read and contributed to the manuscript.

COMPETING FINANCIAL INTERESTSThe authors declare no competing financial interests.

Published online at www.nature.com/doifinder/10.1038/ncb3041Reprints and permissions information is available online at www.nature.com/reprints

1. Huber, M. A., Kraut, N. & Beug, H. Molecular requirements for epithelial-mesenchymal transition during tumor progression. Curr. Opin. Cell Biol. 17,548–558 (2005).

2. Thiery, J. P., Acloque, H., Huang, R. Y. & Nieto, M. A. Epithelial-mesenchymaltransitions in development and disease. Cell 139, 871–890 (2009).

3. Guo, W. et al. Slug and Sox9 cooperatively determine the mammary stem cell state.Cell 148, 1015–1028 (2012).

4. Mani, S. A. et al. The epithelial-mesenchymal transition generates cells withproperties of stem cells. Cell 133, 704–715 (2008).

5. Morel, A. P. et al. Generation of breast cancer stem cells through epithelial-mesenchymal transition. PLoS ONE 3, e2888 (2008).

6. Buck, E. et al. Loss of homotypic cell adhesion by epithelial-mesenchymal transitionor mutation limits sensitivity to epidermal growth factor receptor inhibition. Mol.Cancer Ther. 6, 532–541 (2007).

7. Creighton, C. J. et al. Residual breast cancers after conventional therapy displaymesenchymal as well as tumor-initiating features. Proc. Natl Acad. Sci. USA 106,13820–13825 (2009).

8. Gupta, P. B. et al. Identification of selective inhibitors of cancer stem cells by high-throughput screening. Cell 138, 645–659 (2009).

9. Li, H. J., Reinhardt, F., Herschman, H. R. & Weinberg, R. A. Cancer-stimulatedmesenchymal stem cells create a carcinoma stem cell niche via prostaglandin E2signaling. Cancer Discov. 2, 840–855 (2012).

10. Malanchi, I. et al. Interactions between cancer stem cells and their niche governmetastatic colonization. Nature 481, 85–89 (2011).

11. Merlos-Suarez, A. et al. The intestinal stem cell signature identifies colorectal cancerstem cells and predicts disease relapse. Cell Stem Cell 8, 511–524 (2011).

12. Wynn, T. A., Chawla, A. & Pollard, J. W. Macrophage biology in development,homeostasis and disease. Nature 496, 445–455 (2013).

13. Coussens, L. M., Zitvogel, L. & Palucka, A. K. Neutralizing tumor-promoting chronicinflammation: a magic bullet? Science 339, 286–291 (2013).

14. Gyorki, D. E., Asselin-Labat, M. L., Rooijen, N. v., Lindeman, G. J. & Visvader, J. E.Resident macrophages influence stem cell activity in the mammary gland. BreastCancer Res. 11, R62 (2009).

15. Elenbaas, B. et al. Human breast cancer cells generated by oncogenic transformationof primary mammary epithelial cells. Genes Dev. 15, 50–65 (2001).

16. Ong, S. E. et al. Stable isotope labeling by amino acids in cell culture, SILAC, asa simple and accurate approach to expression proteomics. Mol. Cell. Proteom. 1,376–386 (2002).

17. Udeshi, N. D. et al. Methods for quantification of in vivo changes in proteinubiquitination following proteasome and deubiquitinase inhibition. Mol. Cell.Proteomics 11, 148–159 (2012).

18. Barker, T. H. & Hagood, J. S. Getting a grip on Thy-1 signaling. Biochim. Biophys.Acta. 1793, 921–923 (2009).

19. Wetzel, A. et al. Human Thy-1 (CD90) on activated endothelial cells is acounterreceptor for the leukocyte integrin Mac-1 (CD11b/CD18). J. Immunol. 172,3850–3859 (2004).

20. Lim, E. et al. Transcriptome analyses of mouse and human mammary cellsubpopulations reveal multiple conserved genes and pathways. Breast Cancer Res.12, R21 (2010).

21. Shipitsin, M. et al. Molecular definition of breast tumor heterogeneity. Cancer Cell11, 259–273 (2007).

22. Dontu, G. et al. In vitro propagation and transcriptional profiling of human mammarystem/progenitor cells. Genes Dev. 17, 1253–1270 (2003).

23. Tsuchiya, S. et al. Establishment and characterization of a human acute monocyticleukemia cell line (THP-1). Int. J. Cancer 26, 171–176 (1980).

24. Donnenberg, V. S. et al. Localization of CD44 and CD90 positive cells to the invasivefront of breast tumors. Cytometry B Clin. Cytom. 78, 287–301 (2010).

25. Donnenberg, A. D., Zimmerlin, L., Landreneau, R. J., Luketich, J. D. & Donnenberg,V. S. KIT (CD117) expression in a subset of non-small cell lung carcinoma (NSCLC)patients. PLoS ONE 7, 1–11 (2012).

26. Medrek, C., Ponten, F., Jirstrom, K. & Leandersson, K. The presence of tumorassociated macrophages in tumor stroma as a prognostic marker for breast cancerpatients. BMC Cancer 12, 306 (2012).

27. Tsutsui, S. et al. Macrophage infiltration and its prognostic implications in breastcancer: the relationship with VEGF expression and microvessel density. Oncol. Rep.14, 425–431 (2005).

28. Ch’ng, E. S., Tuan Sharif, S. E. & Jaafar, H. In human invasive breast ductalcarcinoma, tumor stromal macrophages and tumor nest macrophages have distinctrelationships with clinicopathological parameters and tumor angiogenesis. VirchowsArch. 462, 257–267 (2013).

29. Qian, B. Z. & Pollard, J. W. Macrophage diversity enhances tumor progression andmetastasis. Cell 141, 39–51 (2010).

30. Murray, P. J. & Wynn, T. A. Protective and pathogenic functions of macrophagesubsets. Nat. Rev. Immunol. 11, 723–737 (2011).

31. Lin, E. Y., Nguyen, A. V., Russell, R. G. & Pollard, J. W. Colony-stimulating factor1 promotes progression of mammary tumors to malignancy. J. Exp. Med. 193,727–740 (2001).

32. Ohri, C. M., Shikotra, A., Green, R. H., Waller, D. A. & Bradding, P. The tissuemicrolocalisation and cellular expression of CD163, VEGF, HLA-DR, iNOS, and MRP8/14 is correlated to clinical outcome in NSCLC. PLoS ONE 6, e21874 (2011).

33. Gabrilovich, D. I., Ostrand-Rosenberg, S. & Bronte, V. Coordinated regulation ofmyeloid cells by tumours. Nat. Rev. Immunol. 12, 253–268 (2012).

34. Hartman, Z. C. et al. Growth of triple-negative breast cancer cells relies uponcoordinate autocrine expression of the proinflammatory cytokines IL-6 and IL-8.Cancer Res. (2013).

35. Korkaya, H., Liu, S. & Wicha, M. S. Regulation of cancer stem cells bycytokine networks: attacking cancer’s inflammatory roots. Clin. Cancer Res. 17,6125–6129 (2011).

36. Wyckoff, J. et al. A paracrine loop between tumor cells and macrophages is requiredfor tumor cell migration in mammary tumors. Cancer Res. 64, 7022–7029 (2004).

37. Tam, W. L. et al. Protein Kinase C alpha is a central signaling node and therapeutictarget for breast cancer stem cells. Cancer Cell 24, 347–364 (2013).

38. Thomson, S., Petti, F., Sujka-Kwok, I., Epstein, D. & Haley, J. D. Kinase switchingin mesenchymal-like non-small cell lung cancer lines contributes to EGFR inhibitorresistance through pathway redundancy. Clin. Exp. Metastasis 25, 843–854 (2008).

39. Gentile, A., Lazzari, L., Benvenuti, S., Trusolino, L. & Comoglio, P. M. Ror1 isa pseudokinase that is crucial for Met-driven tumorigenesis. Cancer Res. 71,3132–3141 (2011).

40. Pasquale, E. B. Eph receptors and ephrins in cancer: Bidirectional signalling andbeyond. Nat. Rev. Cancer 10, 165–180 (2010).

41. Lamberto, I. et al. Distinctive binding of three antagonistic peptides to the ephrin-binding pocket of the EphA4 receptor. Biochem. J. 445, 47–56 (2012).

42. Zhou, L. et al. EphA4 signaling regulates phospholipase Cgamma1 activation,cofilin membrane association, and dendritic spine morphology. J. Neurosci. 27,5127–5138 (2007).

43. Griner, E. M. & Kazanietz, M. G. Protein kinase C and other diacylglycerol effectorsin cancer. Nat. Rev. Cancer 7, 281–294 (2007).

44. Kalo, M. S. & Pasquale, E. B. Signal transfer by eph receptors. Cell Tissue Res. 298,1–9 (1999).



ART ICLES

45. Lu, Z. G., Liu, H., Yamaguchi, T., Miki, Y. & Yoshida, K. Protein kinase Cdelta activatesRelA/p65 and nuclear factor-kappaB signaling in response to tumor necrosis factor-alpha. Cancer Res. 69, 5927–5935 (2009).

46. Hoffmann, E., Dittrich-Breiholz, O., Holtmann, H. & Kracht, M. Multiple control ofinterleukin-8 gene expression. J. Leukoc. Biol. 72, 847–855 (2002).

47. Sasaki, C. Y., Ghosh, P. & Longo, D. L. Recruitment of RelB to the Csf2 promoterenhances RelA-mediated transcription of granulocyte-macrophage colony-stimulatingfactor. J. Biol. Chem. 286, 1093–1102 (2011).

48. Yoon, S. et al. NF-kappaB and STAT3 cooperatively induce IL6 in starved cancercells. Oncogene. 31, 3467–3481 (2012).

49. Ryan, K. M., Ernst, M. K., Rice, N. R. & Vousden, K. H. Role of NF-kappaB inp53-mediated programmed cell death. Nature 404, 892–897 (2000).

50. Li, S. et al. TWIST1 associates with NF-kappaB subunit RELA via carboxyl-terminalWR domain to promote cell autonomous invasion through IL8 production. BMC Biol.10, 73 (2012).

51. Gisina, A. M. et al. Detection of minor subpopulations of colorectal adenocarcinomacells expressing cancer stem cell markers. Bull. Exp. Biol. Med. 151,234–238 (2011).

52. He, J. et al. CD90 is identified as a candidate marker for cancer stem cells inprimary high-grade gliomas using tissue microarrays.Mol. Cell. Proteomics 11,M111010744 (2012).

53. Jiang, J. et al. Trastuzumab (herceptin) targets gastric cancer stem cells characterizedby CD90 phenotype. Oncogene. 31, 671–682 (2012).

54. Yang, Z. F. et al. Significance of CD90+ cancer stem cells in human liver cancer.Cancer Cell 13, 153–166 (2008).

55. Oki, M. et al. Overexpression of the receptor tyrosine kinase EphA4 in human gastriccancers. World J. Gastroenterol. 14, 5650–5656 (2008).

56. Oshima, T. et al. Overexpression of EphA4 gene and reduced expression of EphB2gene correlates with liver metastasis in colorectal cancer. Int. J. Oncol. 33,573–577 (2008).

57. Binda, E. et al. The EphA2 receptor drives self-renewal and tumorigenicity instem-like tumor-propagating cells from human glioblastomas. Cancer Cell 22,765–780 (2012).

58. Gyorffy, B. et al. An online survival analysis tool to rapidly assess the effect of 22,277genes on breast cancer prognosis using microarray data of 1,809 patients. BreastCancer Res. Treat. 123, 725–731 (2010).



METHODS DOI: 10.1038/ncb3041

METHODSCell culture.HMLE, HMLE-Twist, HMLE-Snail, HMLE-Slug and HMLE-Ras cellswere grown in MEGM medium (Lonza). MDA-MB-435 cells (ATCC) were grownin DMEM containing 10% inactivated fetal bovine serum (IFS). THP1 cells (ATCC)were grown in RPMI1640 containing 10% IFS with 5mM β-mercaptoethanol. Cellswere routinely tested for mycoplasma contamination.

Co-culture assays. The HMLE-derived cells were seeded into culture dishes at30% confluency, and grown into a∼70% confluent monolayer 18 h later. The THP1monocytes were pelleted by centrifugation at 300g , resuspended in the MEGMmedium at 1million cellsml−1, and laid on top of themonolayer cells (cell number ofmonocytes/monolayer cells= 3:2). The non-adherent monocytes were gently rinsedaway after 1 h by two washes with PBS, and the cells in the culture were replenishedwith fresh MEGM medium. Cells were trypsinized at indicated time points andresuspended in 2% IFS/PBS for fluorescence-activated cell sorting (FACS) sorting.For conditioned medium collection, fresh MEGM medium was added to the co-culture (without removing existing medium) after 1 h, and the conditioned mediumwas collected after filtering through a 0.2 µmAcrodisc syringe filter (Pall Life Scienceno. PN4192). For the peptide inhibitor experiments, the monolayer HMLER90hicells were treatedwith the indicated peptide or dimethylsulphoxide for 40min beforeand during the monocyte co-culture.

FACS and flow cytometry. Cells were prepared according to standard protocolsand suspended in 2% IFS/PBS. DAPI (Life Technologies) was used to exclude deadcells. Cells were sorted on BD FACSAria SORP and analysed on BD LSRII, using BDFACSDiva Software (BD Biosciences). Antibodies used were anti-CD90-APC (BDBiosciences no. 559869), anti-CD24-PE (BD Biosciences no. 555428).

Mammosphere/tumorsphere culture. Mammosphere/tumorsphere culture wasperformed as previously described22. The number of cells seeded per well waslabelled for each graph in the figures. Sphere numbers were counted between days 8to 12.

RNA preparation and qRT-PCR analysis. Total RNA was isolated using theRNeasy Plus Mini kit (Qiagen) and reverse transcription was performed withthe High Capacity RNA-to-cDNA Kit (Applied Biosystem), both according to themanufacturer’s protocol. A cDNA sample prepared from 1 µg total RNA was usedfor each PCR. The PCR reactions using SYBRGreenMix I (RocheDiagnostics), datacollection and data analysis were performed on the LightCycler 480 System (RocheDiagnostics). The thermal cycling parameters for the PCR were as follows: 95 ◦C for5min, followed by 45 cycles of 95 ◦C for 10 sec, 49 ◦C for 7 sec, and 72 ◦C for 25 sec.The relativemRNAquantity was normalized against the relative quantity of GAPDHmRNA in the same sample. Primers used are indicated in Supplementary Table 4.

Enzyme-linked immune-sorbent assay. Concentrations of cytokines IL-6, IL-8 and GM-CSF in conditioned media were detected by ELISA kits from RayBiotech (IL-6 andGM-CSF) andEBiosciences (IL-8) according to themanufacturers’protocols. The absorbance was measured on a Multiskan EX (Thermo ElectronCorporation).

Cell proliferation assay. Cells were seeded in 96-well assay plates (Corning no.3903) at 1000 cells per well, and the culture medium was replenished every 2 days.Cell proliferation was measured at the indicated days after seeding with WST1reagent (Roche 11644807001) according to the manufacturer’s protocol.

Isolation of primary mouse macrophages and human monocytes. To isolatemouse intraperitoneal-exude macrophages, mice were injected with 2% v/v Bio-GelP-100 (Bio-Rad no. 150-4174) intraperitoneally (1ml per mouse). Three days later,the mice were euthanized; RPMI medium was injected intraperitoneally and thenretrieved. The cells were then treated with Red Blood Cell Lysing Buffer (Sigma-Aldrich) andwashedwith PBS. The purity of the intraperitoneal-exudemacrophageswas analysed by flow cytometry after staining with anti-CD11b-PE (BD Biosciencesno. 557397), anti-Ly-6G-AlexaFlour700 (BD Biosciences no. 561236) and anti-F4/80-PE-Cy7 (EBiosciences 25-4801-82). Mouse tumour-associated macrophages(TAMs) were isolated from dissociated mouse tumours previously grown from theHMLE-Ras CD90hi cells. Human monocytes were isolated from human peripheralblood mononuclear cells from healthy donors. Mouse TAMs and humanmonocyteswere purified by magnetic activated cell sorting with anti-F4/80–PE–Cy7 followedby anti-PE microbeads (Miltenyi Biotec), and human CD14 microbeads (MiltenyiBiotec), according to the manufacturer’s protocol. The purity of mouse TAMs wasdirectly analysed by flow cytometry after magnetic activated cell sorting. The purityof human monocytes was analysed by flow cytometry after staining with anti-CD14–PE–Cy7 (25-0149-71), anti CD11b–APC (17-0118-71) antibodies (both fromEBiosciences).

Creation of stable cell lines. GFP-labelled HMLE, HMLE-Twist and HMLE-Slug cell lines were made by retroviral transduction using the pWZL–blast–GFPexpression vector59. The HMLE-Ras cell line (HMLER) was made by retroviraltransduction with the pMSCV–HRasV12–IRES–GFP expression vector60. ThetdTomato fluorescent protein-labelled THP1monocytes (THP1–RFP)weremade bylentiviral transduction using the pLV–tdTomato expression vector61. NF-κB super-repressor-expressing HMLER90hi cells were made by retroviral transduction usingthe pBabe–puro–IκBα-S32A/S36A expression vector (Addgene plasmid 12332;ref. 62). Short hairpin RNAs (shRNAs) targeting themRNAs encodingCD90 (Thy1),EphA4, PLCγ1, PKCδ, IL-6, IL-8 and GM-CSF were expressed from pLKO.1-puro(Open Biosystems); the target sequences of these shRNA hairpins are indicated inSupplementary Table 5. All stable cell lines were generated by retroviral infectionusing HEK293T cells, as previously described63, and followed by selection withpuromycin (2 µgml−1) or by cell sorting.

Western blotting. Phospho- and total-protein lysates were prepared with urealysis buffer (9M urea, 20mM HEPES, pH 8.0), with freshly added phosphataseinhibitors (1mM sodium orthovanadate, 5mM sodium pyrophosphate and 1mMβ-glycerol-phosphate). For the separation ofmonocytes from themonolayerHMLE-derived cells, phosphatase inhibitors (1mM sodium orthovanadate, 5mM sodiumpyrophosphate) were added to the culture medium before collection and themonocytes were repeatedly flushed off with the medium; the monolayer cells wererinsed with cold PBS and immediately lysed on the culture dish. The monocyteswere then pelleted by centrifugation at 300g for 5min and then lysed. ForFig. 7c, nuclear and cytoplasmic extracts were prepared using the NE-PER kit(Pierce). Concentrations were determined using the Dc protein assay (Bio-Rad).For Supplementary Fig. 6a, total protein was extracted with RIPA lysis buffer.Protein lysates were resolved on a 4%–12% Bis-Tris gel, transferred to PVDFmembranes, probed with primary antibodies overnight at 4 ◦C and then with HRP-linked secondary antibodies (Cell Signaling Technology, CST) and visualized withECL reagents (Thermo Scientific). The dilution, clone and catalogue numbers of theantibodies used are listed in Supplementary Table 6.

Co-immunoprecipitation and chromatin immunoprecipitation assay.Co-immunoprecipitation of Twist and NF-κB p50 was performed using theHMLER90hi cells with an endogenous high level of Twist. The monolayer cellswere rinsed with cold PBS and lysed in the IP buffer (20mM Tris, pH 7.5, 200mMNaCl, 5mMMgCl2, 0.1% NP40-substitute, 10% glycerol, with 1mM dithiothreitol),0.5mM phenylmethyl sulphonyl fluoride (PMSF) and 1× protease inhibitor(Roche) freshly added before lysis). Cell lysates were incubated at 4 ◦C withthe anti-Twist antibody (Abcam no. ab50887) or control IgG (CST, no. 5415)overnight (Supplementary Table 6). The protein–antibody mixture was thenprecipitated with the protein G magnetic beads (Invitrogen) according to themanufacturer’s protocols. NF-κB p50 was then detected with the anti-p105/p50antibody (CST, no. 3035) by western blot. The chromatin immunoprecipitationassay was performed as previously described with the same antibodies64. Thedilution, clone and catalogue numbers of the antibodies used are listed inSupplementary Table 6. The PCR primers used for confirming the proteinbindings were: promoter probe for IL6: 5′-TTCCAATCAGCCCCACCCG-3′/5′-GCTGGCAGTTCCAGGGCTAAGG-3′; control probe for IL6: 5′-TCTGGAGACTGGAGGGACAACC-3′/5′-GGACGCAGGCACGGCTCTA-3′; promoterprobe for IL8: 5′-CATCAGTTGCAAATCGTGGAAT-3′/5′-GAGTGCTCCGGTGGCTTTTTAT-3′; control probe for IL8: 5′-AAGAGCATGAAGCAACAGTGGC-3′/5′-AATAGGAGGGCTTCAATAGAGG-3′; promoter probe for CSF2: 5′-AAGTTCTCTGGAGGATGTGGCTGC-3′ /5′-CCATCTCAGCAGCAGTGTCTCTAC-3′;control probe for CSF2: 5′-TGGGGGATGCTGGGGTTATGTA-3′/5′-TAGCCCCCAGAAGACACCAGGAG-3′.

Phosphatidylinositol-specific phospholipase C protein (PI-PLC) treatment.THP1 monocytes were resuspended in PBS at 2 million cellsml−1, and incubatedwith 1 µml−1 PI-PLC (Invitrogen no. P-6466) at 37 ◦C for 1 h. Monocytes werethen washed with PBS once and resuspended in MEGM medium for co-cultureexperiments.

Animal experiments. All research involving animals complied with protocolsapproved by the MIT Committee on Animal Care. The number of cells andthe time for tumour growth are indicated in the figure legends. To measuretumour-initiating cell/CSC frequency (Fig. 2a,g, 4j and 6e and SupplementaryFig. 2d,f), serial dilutions of cancer cell suspensions in 25 µl of a 1:1 MEGMand Matrigel (BD Biosciences) mixture were injected into the fourth mammaryfat pads of 2–3-month-old female NOD-SCID mice. CSC frequencies of thesamples were determined using the ELDA webtool65. In TAMs+CSCs co-injectionexperiments (including the CSCs-only control groups; Figs 4b–h and 7d), theamount of Matrigel was reduced by sixfold to minimize any effects of growth

NATURE CELL BIOLOGY


DOI: 10.1038/ncb3041 METHODS

factors from the Matrigel. Thus, the cells were suspended in 10 µl of a 4:1 MEGMand Matrigel mixture and injected into the fourth mammary fat pads of two-month-old female Nude mice. For the MMTV-PyMT tumour cell transplantationexperiment (Fig. 4c,d; Supplementary Fig. 4d–j), naturally arising MMTV-PyMTtumours from 10- to 14-week-old females were collected (with their lungs beingchecked for visible macro-metastases to gauge the metastatic capabilities of thetumour cells). The tumour cells were dissociated into single-cell suspension,subjected to negative selection by CD45, CD31 and Ter-119 microbeads (Miltenyi)and then sorted for CD90 expression. Separate MMTV-PyMT tumours weredissociated for the isolation of TAMs. Tumour cells (3× 105) were injected withor without 1.5× 105 TAMs in a volume of 10 µl including 2 µl Matrigel into thefourth mammary fat pads of 2-month-old female wild-type Fvb mice. Tumoursand lungs were collected after two months. Tumour weight was measured afterdraining of fluid in the cyst-containing tumours. The investigator was not blindedto the group allocation for the assessment of outcome. Mice were randomlyallocated to each group by an independent person; however, no particular methodof randomization was used. No statistical method was used to predeterminesample size.

Patient samples. The tissues used in this study were collected under a Universityof Pittsburgh Internal Review Board (UPCI 04-162). All patients consentedbefore surgery or any other clinically indicated procedure. All specimens are de-identified by an honest broker system and distributed for research. Tissues used forimmunohistology were processed within 2 h of surgery.

ImmunohistostainingMouse tumours. Formalin-fixed paraffin-embedded xenograft tumour tissue slideswere stained with primary antibodies mouse anti-α-SMA (Sigma), rat anti-F4/80(Invitrogen, MF48000), rabbit anti-PLCγ1 (pY783) (Abcam, no. ab53125), rabbitanti-cleaved caspase 3 (CST no. 9664) and mouse anti-Ki67 (BD Biosciencesno. 550609), and the secondary antibodies used were goat anti-mouse Alexa594,goat anti-rat Alexa488, and donkey anti-rabbit Alexa 555 (SupplementaryTable 6). Nuclear staining was performed with DAPI (Invitrogen). Slides weremounted in Prolong Gold anti-fade reagent (Invitrogen). Immunofluorescentstaining was observed and photographed using a Nikon Eclipse E800 uprightfluorescence microscope.

Patient tumour tissues. All tissues BrCA022T, BrCA032T, BrCA041T (invasiveductal carcinomas, ER+/PR+) and BrCA042T (invasive ductal carcinoma,ER−/PR−) were formalin-fixed and paraffin-embedded. Antigen retrieval of tissuesections was performed at 125 ◦C for 20min in pH6.0-EDTA buffer (Dako);these were washed and blocked with blocking solution (PBS, 5% goat serum,0.05% Tween 20) to reduce nonspecific antibody binding. All wash steps wereperformed with Dako wash buffer. Immunofluorescence staining was performedusing the following primary antibodies overnight at 4 ◦C: monoclonal rabbitanti-human CD90 (Abcam, catalogue no. ab133350), mouse anti-human CD68(Dako catalogue no. M0876). Negative control primary antibodies were purchasedfrom Dako (N1698 and N1699). After washing, biotinylated secondary goat anti-rabbit biotin (Dako catalogue no. E0432), goat anti-mouse Alexa488 antibody(CD90/CD68 double staining, Invitrogen catalogue no. A11001) or goat anti-mouse Cy5 antibody (CD90/CD68/CTK triple staining; Abcam catalogue no.ab97037) was applied for 1 h at room temperature, followed by incubation withstreptavidin–Cy3 (Sigma, catalogue no. 6402) for 30min at room temperature.For triple staining mouse anti-CTK–FITC [AE1/AE3] (eBioscience, catalogueno. 53-9003-80) was incubated for 1 h at room temperature. For detailedinformation on the antibodies used see Supplementary Table 6. Final nuclearstaining was performed with DAPI (Invitrogen, catalogue no. D1306). Slides weremounted in Prolong Gold anti-fade reagent (Invitrogen, catalogue no. P36934).Immunofluorescent staining was observed and photographed using a Nikon 90imotorized upright epi-fluorescence microscope using a ×20 objective. Wholesections were photographed as a composite of individual ‘tiles’ captured andjoined using theNIS-Elements software. Individual CD90+/CD68+ cellular doubletswere identified from the ×20 whole-section tiled images. Confocal images of thedoublets were obtained using an Olympus Fluoview 1000 microscope (version 2.16software) and a Plan Apo N ×60 1.42 N.A. objective, ×4 zoom, and 0.45 µmsections (number varied depending on cells imaged: 13–17). Stacks were processedusing MetaMorph (version 7.7) and autocontrast was applied consistently toall wavelengths.

Statistical analysis.Data are presented as mean± s.e.m. Student’s t-test (unpaired)was used to compare two groups (P < 0.05 being considered significant). Holm–Sidak’s multiple comparisons test was performed when one-way ANOVA was usedto compare between more than two groups. Sample sizes were indicated in eachcorresponding figure legends.

Proteomic analysis ofmembranepellet samples. SILACmedia forHMLE-vectorand HMLE-Twist cells were made based on MCDB-170 (ref. 66). For the amino-acid labelling, either normal L-lysine and L-arginine or heavy-labelled 13C6-15N2lysine and 13C6-15N4 arginine were supplemented at concentrations of 40mg l−1 and120mg l−1, respectively. Labelled cells were collected after 8 cell doublings.

Membrane protein fractionation. Membrane protein isolation is described inSupplementary Fig. 1b,c. The buffers used were as follows: lysis buffer: 10mM Tris,pH 7.5, 5mM MgCl2, 0.5mM PMSF, and 0.5% PMSF and 1% protease inhibitors(Sigma) fresh before use; sucrose buffer: 10mM Tris, pH 7.5, 1.25M sucrose(molecular mass 342.3Da). Membrane proteins were pelleted in an ultracleartube (Beckman no. 344057) in a Beckman ultracentrifuge (model no. L8-M) withrotor SW50.1.

Solubilization, digestion andpeptide fractionation by off-gel electrophoresis.Membrane protein pellets (1mg) were solubilized in urea and disulphide bondswere reduced and alkylated; then proteins were deglycosylated with PNGaseFand digested with Lys-C, and trypsin. Solutions that began cloudy on initialreconstitution were clear after overnight digestion. Samples (∼50 µg) of peptidedigest were separated by off-gel electrophoresis according to their isoelectric pointinto 11 fractions using anAgilent 3100OFFGELFractionator (Agilent Technologies)and 13 cm Immobiline Drystrips pH 3–10 (GE Healthcare BioSciences AB).Experimental details have been previously described67.

Mass spectrometry. Peptide off-gel electrophoresis fractions were analysed with anautomated nano LC-MS/MS system, consisting of an Agilent 1100 nano-LC system(Agilent Technologies) coupled to either an LTQ-Orbitrap or an LTQ Orbitrap XLFourier transform mass spectrometer (Thermo Fisher Scientific) equipped witha nanoflow ionization source (James A. Hill Instrument Services). Peptides wereeluted from a 10 cm column (Picofrit 75 µm ID, New Objectives) packed in-housewith ReproSil-Pur C18-AQ 3 µm reversed-phase resin (Dr. Maisch , AmmerbuchGermany) using either a 120 or 133min gradient at a flow rate of 200 nlmin−1to yield ∼20 s peak widths as previously described67. Data-dependent LC-MS/MSspectra were acquired in ∼3 s cycles; each cycle was of the following form: one fullOrbitrap MS scan at 60,000 resolution followed by 8 MS/MS scans in the ion trapon the most abundant precursor ions using an isolation width of 3 m/z . Dynamicexclusion was enabled with a mass width of+/−25 ppm, a repeat count of 1 and anexclusion duration of 45 sec. Charge-state screening was enabled along with mono-isotopic precursor selection and non-peptide mono-isotopic recognition to preventtriggering ofMS/MS on precursor ions with unassigned charge or a charge state of 1.Normalized collision energy was set to 30 with an activationQ of 0.25 and activationtime of 30ms. Two biological replicates were performed.

Protein identification and quantification. All MS data were interpreted usingthe Spectrum Mill software package v4.1 beta (co-developed by us with AgilentTechnologies). Similar MS/MS spectra acquired on the same precursor m/z within+/−45 s were merged; MS/MS spectra with precursor charge >4 and poor-qualityMS/MS spectra, which failed the quality filter by not having a sequence taglength >0 (that is, minimum of two masses separated by the in-chain mass ofan amino acid), were excluded from searching. MS/MS spectra were searchedagainst a UniProt database containing 78,369 human sequences (including isoformsand excluding fragments) downloaded from the UniProt web site on 30 June2010, along with 73 common laboratory contaminant proteins appended. Searchparameters included: ESI linear ion-trap scoring parameters, trypsin enzymespecificity with cleavage at K-P or R-P allowed, a maximum of 4 missed cleavages,35% minimum matched peak intensity, +/−20 ppm precursor mass tolerance,+/−0.7Da product mass tolerance, and fixed/mix modifications that includedcarbamidomethylation of cysteines and either unlabelled Arg, Lys or SILAC-labelled Arg 10, Lys 8. Allowed variable modifications were oxidized methionine,deamidation of asparagine, pyro-glutamic acid modification at amino-terminalglutamine, and pyro-carbamidomethyl modification at N-terminal cysteine with aprecursor MH+ shift range of−18–64Da.

Identities interpreted for individual spectra were automatically designated asconfidently assigned using the SpectrumMill autovalidationmodule to apply target–decoy-based false-discovery rate (FDR) scoring threshold criteria using a two-stepauto threshold strategy at the spectral and protein levels. First, peptide mode wasset to allow automatic variable range precursor mass filtering with score thresholdsoptimized to yield a spectral-level FDR of 1.6% for each precursor charge state ineach LC-MS/MS run. Second, protein mode was applied to further filter all of thepeptide-level validated spectra combined from the 22 LC-MS/MS runs from bothreplicates of an experiment using a minimum protein score of 20 and a maximumprotein-level FDR of zero. As themaximum peptide score is 25, the protein level stepfilters the results so that each identified protein is comprised of multiple peptidesunless a single excellent scoring peptide was the sole match. The above criteria

NATURE CELL BIOLOGY


METHODS DOI: 10.1038/ncb3041

yielded target–decoy-based FDR estimates for each sample replicate of 0.9–1.0% atthe peptide-spectrum match level and 1.3–1.5% at the distinct peptide level.

For scores calculated at the protein level and reporting of identified proteins,redundancy is addressed as follows. The protein score is the sum of the scoresof distinct peptides. A distinct peptide is the single highest scoring instance ofa peptide identified by an MS/MS spectrum. MS/MS spectra for a particularpeptide may have been recorded multiple times (that is, both heavy and lightSILAC labels, different precursor charge states, isolated from adjacent off-gelelectrophoresis fractions, modified by deamidation at asparagine or oxidation ofmethionine) but are still counted as a single distinct peptide. When a peptidesequence >8 residues long is contained in multiple protein entries in the sequencedatabase, the proteins are grouped together and the highest scoring one and itsaccession number are reported. In some cases when the protein sequences aregrouped in this manner there are distinct peptides that uniquely represent alower scoring member of the group (isoforms, family members). Each of theseinstances spawns a subgroup and multiple subgroups are reported and countedtowards the total number of proteins with each given related protein.subgroupnumbers (for example, the ephrin receptor tyrosine kinase family shares sufficientsequence homology that the observed shared and distinct peptides resulted inone protein group with 4 subgroups corresponding to EPHA2, EPHA4, EPHB2and EPHB4 and are listed as subgroup members 214.1 to 214.4 in SupplementaryTable 1). GO annotations were then used to categorize all of the identified proteinsubgroups as being plasma membrane associated, extracellular matrix derived,and/or tyrosine kinases. EphA4 is complicated by the fact that there is over 50%sequence identity among the Ephrin family members with distinct peptides detectedfor EphA2, A4, B2, B4. In each replicate a single distinct A4 tryptic peptide wasobserved, different ones each time, and both showed >20-fold expression increase(data not shown). The peptides that are shared amongst family members and thosedistinct to the other three isoforms do not show significant expression change. Themarginal distinct A4 evidence resulted from the median ratio of all its constituentpeptides being below the significant change threshold in each replicate (Fig. 1a andSupplementary Table 1).

SILAC relative abundance ratios of proteins were determined using extractedion chromatograms for each sequenced peptide’s light and heavy labelled precursorion pair in the intervening high-resolution FT-MS scans of the LC-MS/MS runsas previously described67. Each individual protein’s SILAC ratio was calculated asthe median of all SILAC precursor ion ratios corresponding to MS/MS spectraconfidently assigned to peptides from that protein. Proteins were consideredquantifiable if they were observed in each of 2 replicate experiments with atleast 2 SILAC ratios measured in each experiment. Ratios for common laboratorycontaminant proteins were filtered out of each data set. As two biological replicateexperiments were performed with a light/heavy label-swap between experimentsthe contaminants were easily recognizable not only by identity but also by extreme

quantification ratios with nearly all signals in the light channel. To assign regulatedproteins (Fig. 1a and Supplementary Table 1), we used the Limma package in theR environment to calculate moderated t-test P values corrected by the Benjamini–Hochberg method, as described previously17. To perform the moderated t-test, theprotein SILAC ratios were standardized (distributions centred at zero and replicate1 versus 2 diagonalized). Each log2 protein ratio in a replicate experiment wasstandardized by subtracting the median and dividing by the standard deviation ofall log2 protein ratios in that replicate.

Repeatability of experiments.Representative images are shown in Figs 1–4,6 and 7and Supplementary Figs 1–4 and 6. The experiments have been repeated as follows:main figures: (1b,d) 3 repeats, (2b,c) 5 repeats, (3a,b) see fig., (3c) 3 repeats, (3e)5 patient tumours, (3i) 25 tumour biopsies, (4d) 3 repeats, (6a–c,f, 7e) 3 repeats,(6h,7c,f) 2 repeats; supplementary figures: (1c) 3 repeats, (2e,h) 4 repeats, (2g) 2repeats, (3a,c,d and 4c) 5 patient tumours, (3b) 25 tumour biopsies, (4e–g,i) 2 repeats,(6a,f,g) 2 repeats, (6c) 3 repeats.

Data deposition. The original mass spectra may be downloaded from MassIVE(http://massive.ucsd.edu) using the identifier: MSV000078799. The data areaccessible at ftp://MSV000078799:[email protected].

59. Orimo, A. et al. Stromal fibroblasts present in invasive human breast carcinomaspromote tumor growth and angiogenesis through elevated SDF-1/CXCL12 secretion.Cell 121, 335–348 (2005).

60. Yu, M. et al. A developmentally regulated inducer of EMT, LBX1, contributes to breastcancer progression. Genes Dev. 23, 1737–1742 (2009).

61. Shibue, T. & Weinberg, R. A. Integrin beta1-focal adhesion kinase signaling directsthe proliferation of metastatic cancer cells disseminated in the lungs. Proc. NatlAcad. Sci. USA 106, 10290–10295 (2009).

62. Tergaonkar, V., Pando, M., Vafa, O., Wahl, G. & Verma, I. p53 stabilization isdecreased upon NFkappaB activation: A role for NFkappaB in acquisition ofresistance to chemotherapy. Cancer Cell 1, 493–503 (2002).

63. Stewart, S. A. et al. Lentivirus-delivered stable gene silencing by RNAi in primarycells. RNA 9, 493–501 (2003).

64. Han, J. et al. Tbx3 improves the germ-line competency of induced pluripotent stemcells. Nature 463, 1096–1100 (2010).

65. Hu, Y. & Smyth, G. K. ELDA: extreme limiting dilution analysis for comparing depletedand enriched populations in stem cell and other assays. J. Immunol. Methods 347,70–78 (2009).

66. Hammond, S. L., Ham, R. G. & Stampfer, M. R. Serum-free growth of humanmammary epithelial cells: rapid clonal growth in defined medium and extended serialpassage with pituitary extract. Proc. Natl Acad. Sci. USA 81, 5435–5439 (1984).

67. Naba, A. et al. The matrisome: In silico definition and in vivo characterization byproteomics of normal and tumor extracellular matrices. Mol. Cell. Proteomics 11,M111 014647 (2012).

NATURE CELL BIOLOGY


S U P P L E M E N TA RY I N F O R M AT I O N

WWW.NATURE.COM/NATURECELLBIOLOGY 1

DOI: 10.1038/ncb3041

b

c

a Harvest 108 cells

Cell lysis in hypotonic buffer

Pellet nuclei at 900 xg

1st supernatant

Add 0.25 M sucrose to supernatant Pellet membrane vesicles at 100,000 xg

2nd supernatant

Denature and resolve protein

76

102

38

52

225

76

102

52

38

102

Supplementary Figure 1 Proteomics profiling of EMT-induced changes in membrane proteins. (a) For stable isotope labeling, the HMLE-Twist cells and control HMLE-vector cells were maintained in the same medium, except that the medium for one of the two cell types contained the C13N15 Arg and Lys amino acids (heavy) instead of the standard C12N14 Arg and Lys (light). A replicate experiment was subsequently performed in which the light/heavy labels were swapped in order to reduce noise from proteins present in the culture medium. Following culture, the HMLE-vector and HMLE-Twist cells were mixed in equal numbers; a membrane

protein fraction was prepared from the mixture (panel b) and subjected to tryptic digestion. The resulting peptides were further resolved by off-gel electrophoresis and analyzed by LC-MS/MS (liquid chromatography-tandem mass spectrometry). (b) A flow chart of the membrane protein fractionation strategy. (c) Western blot detecting protein enrichment and contamination in different fractions. E-cadherin and N-cadherin, plasma membrane proteins; Large T antigen, nuclear protein; b-actin, cytosolic and cytoskeletal protein. Numbers on the gels indicate molecular weight markers in kDa.



2 WWW.NATURE.COM/NATURECELLBIOLOGY

a b

d

CD90lo CD90hi

e g f CD90low

CD90high

2x105 cells Nude mice

Tumor incidence

Tumor detectable Weight

HMLE-RasGFP 1/10 65 days 0.08 g (65 d) CD90hi24- 7/10 22 days 0.1 -1.8 g (35-65 d)

CD90med24+/- 0/5 Undetectable at 90 d

CD90lo24+ 0/10 Undetectable at 90 d

c

h

0.2 mm

0.2 mm

Supplementary Figure 2 CD90 expression is enriched in the MaSCs and CSCs. (a) CD90 mRNA levels in HMLE-derived cell lines detected by expression array. HMLE and HMLE-vector are epithelial control cells, while Twist, Snail, Slug and Zeb1 cells have undergone EMT driven by these individual EMT-TFs. (b) Enrichment in gene expression of THY1 (CD90) and three other EMT-related genes in the human MaSC-enriched subset; plotted based on data from Lim et al. (2010) 20. (c) Expression of CD90 in CD44+/PROCR+ stem/progenitor cells or CD24+ differentiated cells purified from human breast tumors; plotted based on data from Shipitsin et al (2007) 21. Results are presented as the counts of SAGE tags for THY1 (CD90) in 10 SAGE libraries, The ratio and P value shown is between CD44+/PROCR+ (n=5; 1 normal, 4 tumors) and CD24+ (n=5; 2 normal, 3

tumors) purified cells. Student’s t-test performed. Error bars indicate S.D. (d) Tumor incidence of sorted CD90hiCD24- HMLER cells in Nude mice compared to the bulk, CD90medCD24+/- or CD90lo24+ populations. (e) MB-MDA-435 cells sorted based on CD90 expression. (f) Tumor weight and (g) H&E staining of xenograft tumors from sorted CD90low and CD90high

MB-MDA-435 cells. Data from two independent experiments were pooled (f). Error bar indicate S.E.M. **p<0.01, ***p<0.001 (Student’s t-test). n= 10,10,8,10,9,10 mice (left to right). The CD90high cells formed tumors with more aggressive histopathological characteristics, including larger nuclei and elevated stromal cell infiltration. (h) Flow cytometry profile showing CD90 and CD24 levels for the CD90hi, shCD90 in CD90hi, and CD90lo HMLER cells.




Correla'on: 0.388273, P value < 1x10-‐15

BrCA042T (ER-‐)

10 µm

10 µm

c

d

CD68+ TAMs CD90+ CSCs DAPI

BrCA032T (ER+)

a

b

BrCA032T (ER+)

BrCA042T (ER-‐)

CD90 CTK CD68 DAPI (white)

BrCA041 (ER+)

10 µm

BrCA042 (ER-‐)

10 µm

e

1

Supplementary Figure 3 CD90 expression in patient breast cancer tissue. (a) Lower magnification images of the tumors 042 (25x) and 032 (20x) as in Fig. 3G. CSC (red)-TAM (green) pairs were highlighted with red boxes in tumor 032. Tumor 042 has higher contents of CSCs and TAMs and extensive CSC-TAM contacts were observed. (b) CTK+CD90- cancer cells (green) rarely showed contact with nearby CD68+ TAMs (blue; ER+ BrCA041), whereas CD90+ CSCs (red) that have partial loss of CTK (yellow arrowheads) can be observed adjacent to TAMs (ER- BrCA042). No CD68+CTK+ cells were observed, indicating that the CD68 antibody is not staining non-specifically for cancer cells. (c) Flow cytometry profiles of breast cancer cells from freshly

excised ER- tumor BrCA042. Aneuploid cells were gated by DAPI >2n (control 2n defined using diploid lymphocytes). The CD90 and CD44 flow cytometry profiles of both cytokeratin+ (CTK+) aneuploid cells (green panel) and CTKneg aneuploid cells (blue panel) were shown. (d) Expression of CD44 and CD90 on cytokeratin+ (iCTK+) breast cancer cells in ER+ tumor BrCA032. 33.7% of hematopoietic lineage negative cells (Non-Heme) were CTK+, about a third of which were aneuploid. 8.6% of all CTK+ cells coexpressed CD90 and CD44. (e) THY1 (CD90) expression correlates with CD68 expression in patient breast cancer samples. Pearson’s correlation was obtained using TCGA breast invasive carcinoma gene expression data set downloaded from the UCSC Cancer Browser.




CD90-‐ non-‐CSCs CD90-‐ non-‐CSCs + TAMs CD90+ CSCs CD90+ CSCs + TAMs

0.5 mm 0.5 mm 0.5 mm 0.5 mm

Ne Ne

Cy

Cy Cy

Ne Ne

Ne

Ne

c

HLADR CD163 CD68 HLADR CD163 CD68

a b Tumor cells TAMs (by MACS)

F4/80

17% 97%

f

g

i 50 µm 50 µm 50 µm 50 µm Ki67 Ki67 Ki67 Ki67

j

25 µm 25 µm 25 µm 25 µm

CD3 CD3 CD3 CD3

25 µm

CD3

Spontaneously arising tumor e

d

CD90-‐ non-‐CSCs CD90-‐ non-‐CSCs + TAMs CD90+ CSCs CD90+ CSCs + TAMs h

k

25 μm

25 μm

Supplementary Figure 4 TAMs promote tumor growth. (a) TAMs were isolated by MACS from xenograft tumors derived from HMLER90hi cells. The purity of the TAMs is indicated by the percentage of F4/80+ cells. The viability of the TAMs determined by DAPI staining is typically between 75% and 87%. (b) TAMs from HMLER90hi-CSC cells (C1,C2) and CSC+TAM coinjections (CT1-3) and naturally arising tumors in MMTV-PyMT mice (P1-P7) were sorted by F4/80 positivity. Gene expression of the M1/M2 related genes were measured by Q-PCR. (c) Tumor sections from breast cancer patients were stained with CD68, CD163 and HLA-DR. Almost all the CD68+ cells show CD163+HLA-DRlo M2 macrophage phenotype (white arrowheads). Note that there are background signals which are spotted rather than homogenous staining of the cell surface and these cells are not counted as CD68+ TAMs in any of the quantifications in main Fig. 3. (d-l) Naturally arising MMTV-PyMT tumors were dissociated and tumor cells were sorted based on

CD90 expression, whereas TAMs were isolated by F4/80 expression. 3x105 CD90+ or CD90- tumor cells were orthotopically transplanted into syngeneic wildtype Fvb mice with or without comixed 1.5x105 TAMs. Typical images are shown. (d) Primary tumor weight after draining fluid; data were pooled from 3 independent experiments. n=15 mice. T lymphocyte infiltration was shown by CD3 staining of naturally arising tumor (e) and transplanted primary tumors (f). (g) H&E staining of primary tumors. Ne, necrosis; Cy, fluid-filled cyst. (h) Percentage of necrotic area (including necrosis and cysts) was quantified using Fiji software. n=9 (3 fields were randomly chosen of 3 tumors for each group). (i) Ki67 staining of primary tumors. (j) Quantification of Ki67+ proliferating cells per field at 20x magnification, n=15 (5 fields were randomly chosen of 3 tumors for each group). (k) Experimental scheme of the secondary transplantation experiment in main Figure 4g&h. Error bars indicate S.D. ** p<0.01; **** p<0.0001 (One way ANOVA).




b

c

e

d

f

a

Supplementary Figure 5 The cytokines IL-6 and IL-8 secreted by the CSC-like cells promote CSC survival. (a) The coculture experiment followed by FACS and RNA harvesting. (b) Knockdown of IL-6 and IL-8 in unsorted epithelial HMLER cells. (c) Proliferation of CD90hi or unsorted HMLER cells with IL-6 or IL-8 knockdown. (d) Doxycyclin-induced IL-8 secretion in the HMLER90hi cells with the TetO-IL8 expression vector +/- the rtTA vector,

which is required for induction. (e) Doxycyclin-induced IL-8 secretion resulted in enhanced tumorsphere formation in the HMLER90hi cells. (f) Treatment with recombinant cytokines IL-6 (0.8 ng/ml) and/or IL-8 (10 ng/ml) increased tumorsphere formation by the HMLER90hi cells. (b-f) Data plotted are average (b,c) or individual (d-f) values of 3 technical replicates from one of two independent experiments. Source data provided in Supplementary Table 7.




c a

1.00 1.08

0.39

0.0 0.2 0.4 0.6 0.8 1.0 1.2

mRN

A expressio

n

EphA4 knockdown d

g

b

e

f

h

Supplementary Figure 6 Contact with monocytes activates EphA4 and downstream proteins in the stem-like cells. (a) Western blots comparing the protein expression of EphA4 and PDGFRb in the HMLE-Twist, HMLE-Slug CD90hi and HMLER90hi mesenchymal cells with the HMLE and HMLER90lo epithelial cells. (b) Separation of HMLER90hi or HMLE-Twist cells from cocultured monocytes for western blot analysis. (c) Phospho- and total-protein levels of EphA4, PLCγ1 and PKCδ in HMLE-Twist cells (+/-monocyte-coculture) were detected by western blot. b-actin served as loading control. (d) Q-PCR verifying the knockdown of EphA4 by the shRNA. (e) Cytokine mRNA induction (fold-change relative to corresponding no-monocyte control) in the HMLER90hi cells with sh-Luciferase or sh-EphA4 after 3 hr of coculture with the monocytes. Data plotted in d,e are average values of 3 technical replicates from one out

of two biological repeats. Source data provided in Supplementary Table 7. (f) Phospho- and total-protein levels of EphA4 in HMLER90hi cells with or without shRNA knockdown of CD90 (+/-monocyte-coculture) were detected by western blot. (g) Western blots of phospho- and total EphA4, EphA2, PLCγ1, PKCδ and Src in the CSC-like cells cocultured for 30 min (30’) with monocytes. The monocytes were pre-treated with either PBS only or PBS with 1 unit/ml PI-PLC. (h) Tumor-associated monocytes and macrophages (TAMs) interact with cancer stem cells (CSCs) via CD90- and EphA4- mediated juxtacrine signaling, which activates a network of cytokine signaling. The cytokines can act in autocrine to maintain the stem-cell state, and in paracrine on the inflammatory cells to perpetuate the TAM-CSC interactions. As critical components of the CSC-niche, TAMs enhance tumor initiation and CSC expansion.




Supplementary Figure 7 Un-cropped Western Blots images. The corresponding figure numbers and panels are shown. Blue boxes highlight the lanes presented in the Figures. Blots from the same gel are grouped together, and loading controls are provided for each gel.




S6a

EphA4

PDGFRβ

β-‐ac0n 52kDa

150kDa

150kDa

102kDa

38kDa

S6c

EphA4 pY779

EphA4

β-‐Ac0n

PKCδ pY311

PKCδ

PLCγ1 pY783

PLCγ1

β-‐Ac0n

225kDa 150kDa

225kDa 150kDa

52kDa

38kDa

102kDa 76kDa

102kDa 76kDa

102kDa 150kDa

225kDa 150kDa

52kDa 38kDa

S6g

EphA4-‐pY779

EphA4

PLCγ1-‐pY783

PLCγ1

PKCδ-‐pY311

PKCδ

Src pY416

Src

GFP

EphA2-‐pY594

EphA2

225kDa 150kDa

225kDa 150kDa

102kDa

102kDa 225kDa 150kDa

225kDa 150kDa

102kDa

102kDa

225kDa

225kDa 150kDa 102kDa

76 kDa

76kDa

76kDa

52 kDa

76kDa

52 kDa

31kDa

150kDa

EphA4 pY779

EphA4

S6f

102kDa 150kDa

102kDa 150kDa

Supplementary Figure 7 continued Un-cropped Western Blots images. The corresponding figure numbers and panels are shown. Blue boxes highlight the lanes presented in the Figures. Blots from the same gel are grouped together, and loading controls are provided for each gel.




Supplementary Table Legends

Supplementary Table 1. Protein ratios between HMLE-Twist and HMLE-vector cells detected by SILAC-MS. Replicate 1 light: HMLE-Twist (EMT-inducing transcription factor), heavy: HMLE-vector. Replicate 2 heavy: HMLE-Twist (EMT-inducing transcription factor), light: HMLE-vector.To assign regulated proteins we calculated moderated T-test p-values corrected by the Benjamini Hochberg method, as described previously (Udeshi et al, 2012). In order to perform the moderated T-test, the protein SILAC ratios were standardized (distributions centered at zero and replicate 1 vs 2 diagonalized). Each log2 protein ratio in a replicate experiment was standardized by subtracting the median and dividing by the standard deviation of all log2 protein ratios in that replicate. Replicate 1 median: 1.074, standard deviation: 1.085. Replicate 2 median: 0.372, standard deviation: 1.662. GO annotations were used to categorize all of the identified protein subgroups as being plasma membrane associated (PM), extra cellular matrix derived (ECM), and/or tyrosine kinases (TK).

Supplementary Table 2. mRNA levels of cytokines expressed by the HMLER90hi cells and by the THP1 monocytes before and after coculture. mRNA levels were measured by human common cytokine Q-PCR array (SABiosciences). mRNA expression delta (Ct) values were normalized to house-keeping genes: B2M, HPRT1, RPL13A, GAPDH and ACTB. HGDC is human genomic DNA control for detection of genomic DNA contamination. The fold change values in red are up-regulation, while values in blue are down-regulation. Data were sorted according to fold-change of cytokine mRNA in the cocultured CSCs compared to CSCs alone.

Supplementary Table 3. Human common cytokine gene expression fold-change in monocytes cocultured with CSCs compared to control monocytes. N.D., not detected. Values in red are up-regulation, while values in blue are down-regulation.

Supplementary Table 4. Q-PCR primers used for measuring mRNA expression.

Supplementary Table 5. Target sequences for shRNA knockdown constructs.

Supplementary Table 6. Antibodies used in the experiments.


A breast cancer stem cell niche supported by juxtacrine signalling ...

Documents

Transcript of A breast cancer stem cell niche supported by juxtacrine signalling ...