Cancer Letters 323 (2012) 20–28

Contents lists available at SciVerse ScienceDirect

Cancer Letters

journal homepage: www.elsevier .com/ locate/canlet

Sphere formation reverses the metastatic and cancer stem cell phenotypeof the murine mammary tumour 4T1, independentlyof the putative cancer stem cell marker Sca-1

Heli Matilainen a,1, Xiao-Wen Yu a,1,2, Ching-Wen Tang b, Michael V. Berridge c, Melanie J. McConnell a,⇑a Cell Survival Laboratory, Malaghan Institute of Medical Research, PO Box 7060, Wellington 6242, New Zealandb Vaccine Research Laboratory, Malaghan Institute of Medical Research, PO Box 7060, Wellington 6242, New Zealandc Cancer Cell Molecular Biology Laboratory, Malaghan Institute of Medical Research, PO Box 7060, Wellington 6242, New Zealand

a r t i c l e i n f o a b s t r a c t

Article history:Received 19 January 2012Received in revised form 19 March 2012Accepted 22 March 2012

Keywords:4T1MetastasisSca1Tumour spheresTGFb signalling

0304-3835/$ - see front matter � 2012 Elsevier Irelanhttp://dx.doi.org/10.1016/j.canlet.2012.03.028

⇑ Corresponding author. Tel.: +64 44996914/857; fE-mail address: mmcconnell@malaghan.org.nz (M

1 Equal contribution by these authors.2 Present address: Northwestern University Interde

gram. 320 East Superior St., Searle Building 5-474, Chi

Breast cancer stem cells (BCSCs) initiate and sustain breast cancers, and several putative markers havebeen proposed to prospectively isolate BCSC from the non-cancer stem cell population. The candidateBCSC marker Sca-1 is a GPI-linked membrane protein expressed on activated lymphocytes, hematopoieticstem cells and mammary stem cells. Sca-1+ cells were purified from the murine mammary tumour cellline 4T1. However, this did not enrich for a stem-like, tumour initiating or metastatic cell populationin vitro or in vivo. Sphere formation, which induced high levels of Sca-1, reduced BCSC gene expressionwith near complete loss of spontaneous metastasis from sphere-derived tumours. This was associatedwith decreased expression of TGFB2 and reduced activation of the TGFb signalling pathway in spheres.Both TGFB2 expression in vitro and spontaneous metastasis in vivo could be restored upon re-differentiationof sphere cells by exposure to serum, and this occurred with retention of the majority of Sca-1 expression.We conclude that while putative BCSC, including spheres, can have high Sca-1 expression, Sca-1 itself isnot a marker of BCSC in established 4T1 tumours or the cell line.

� 2012 Elsevier Ireland Ltd. All rights reserved.

1. Introduction

Breast cancers, like most tumours, consist of heterogeneouspopulations of cells with distinct morphology, characteristic re-sponse to therapy and varying metastatic ability. The cancer stemcell hypothesis posits that while the majority of cancer cells have alimited lifespan and cannot initiate new tumours, a sub-populationexhibit tumour-initiating properties due to the acquisition of self-renewal activity [1]. This is most often mediated by expression of acohort of embryonic stem cell (ESC) transcription factors, includingSox2, Oct4, Nanog and/or Msi1 [2]. These regulators, which arecritical for inducing the development of pluripotent stem cells(iPS), similarly can confer stem cell properties on otherwise non-stem cancer cells [3]. These cancer stem cells (CSCs) are thoughtto be responsible for metastasis of solid tumours, and increasinglyhave been shown to be inherently resistant to conventional che-

d Ltd. All rights reserved.

ax: +64 44996915..J. McConnell).

partmental Neuroscience Pro-cago, IL 60611-3010, USA.

motherapy and radiation treatment [4], potentially explainingthe resistance of metastatic disease to treatment.

After the first identification of CSCs in leukaemia [5], markersthat can prospectively identify CSC from other tumour types havebeen actively sought. A marker expressed specifically on the sur-face of CSCs would allow the direct purification and analysis ofthese cells. In human breast cancer, while many cell surface pro-teins have been investigated and some proposed to mark breastCSCs [6], a consensus is forming around CD44+CD24� as a cell sur-face BCSC marker [7–14]. In murine models, breast CSC activity hasalso been associated with the small GPI-linked membrane proteinSca-1 (stem cell antigen 1)/Ly6A [15]. Predominantly used as amarker of mouse hematopoietic stem cells, and of lymphocytelineage cells [16], Sca-1 is also expressed in various tissues, includ-ing mammary stem and progenitor cells [17]. Sca-1 is believed toregulate cell signalling by formation of glycosphingolipid rafts thatconcentrate or exclude specific signalling molecules. Recent dataindicate that the normal function of Sca-1 may be to block signal-ling through the TGFBR complex [18], and that this is important inthe early stages of carcinogenic transformation, where TGFb sig-nals are tumour suppressive. Sca-1 is up-regulated on murinemammary cells undergoing carcinogen-driven transformation[18], and in spheres generated from oncogene-driven murine

H. Matilainen et al. / Cancer Letters 323 (2012) 20–28 21

tumours [15]. There is no direct human homolog of the Sca-1 gene,but other members of the Ly6A superfamily of membrane proteinsmight conceivably play a similar role, and functional data suggeststhe existence of a human Sca-1 ortholog [19].

Several assays have been used in vitro to rapidly infer the iden-tity of putative CSCs, including measuring ESC gene expression toshow acquisition of a stem-like phenotype, and ability to self-renew in stem cell media over multiple passages, usually as ananchorage-independent spheroid [20]. However, the key assay isenhanced tumorigenicity in a mouse model [21]. This is most oftendefined as either (i) an increase in the frequency of tumour-initiating cells, allowing formation of tumours with lower cellnumber, or (ii) accurate recapitulation of the original phenotypeof the tumour from which CSCs were purified, i.e. phenocopying.

While these assays address initiation of primary tumours, theydo not address metastatic ability of the putative CSC, which is a keycomponent of breast cancer in particular. Further, the analysis ofhuman CSC in vivo requires the use of immune compromised mice,which necessarily prevents the analysis of immune parameters inCSC biology. This is particularly problematic with regard to metas-tasis, since immune evasion is believed to be critical for successfulmetastasis. Hence, we investigated the potential identification ofmetastatic CSC by using Sca-1, as either a naturally-expressed ortumour sphere-induced CSC marker, in the transplantable murinebreast carcinoma 4T1, in immune competent mice.

2. Methods and materials

2.1. Cell culture

4T1 cells were obtained from ATCC (Alexandria, VA) and maintained in 4.5 g/LD-glucose RPMI (Life Technologies, Auckland, NZ). Media was supplemented with10% non-heat inactivated FCS (Sigma, St. Louis, MO). At 70–80% confluency, cellswere dissociated with 1 mM EDTA (Sigma, St. Louis, MO) and replated. Cells weregrown in 37 �C and 5% CO2 incubators. All the experiments used cells no later thanpassage 15 from ATCC. 4T1 tumorspheres were grown in ultra-low attachmentplates (Corning Life Sciences, Lowell, MA) at a density of 1000 viable cells/mL in ser-um free Mammary Epithelial Basal Medium (MEBM), supplemented with MEGMSingleQuots (Lonza BioResearch, Mt Waverley, Australia), including insulin, humanepidermal growth factor (hEGF) and hydrocortisone (antibiotics and bovine pitui-tary extract were omitted), 20 ng/ml basic fibroblast growth factor (bFGF) and4 lg/ml heparin (Stem Cell Technologies, Melbourne, Australia). Non-adherenttumorspheres were collected by gentle centrifugation (1000 rpm) after reaching�200 lM (7 to 10 days), and dissociated enzymatically using 1 mM EDTA for10 min at 37 �C followed by mechanical trituration using a 26G syringe. Single cellsuspensions were then seeded at 1000 cells/mL. Alternatively, spheres were seededinto adherent T75 mm2 or T25 mm2 flasks using parental cell media for re-differentiation.

2.2. Flow cytometry and cell sorting

Parental 4T1 cells were sorted into Sca-1 positive and negative using the BDFACSVantage DiVa (Becton Dickinson, CA, USA). Flow cytometry was conductedon single cell suspensions of 1 � 106 cells in Dulbecco’s phosphate buffered saline(PBS), (Life Technologies, Auckland, NZ) containing 1% bovine serum albumin (ICP-bio, Auckland, NZ). Antibody used: TGF-bRII-APC: 25 lg/mL (R&D Systems#FAB532A); Sca1-PE: 1:200 dilution of 0.2 mg/mL (BD Biosciences, San Jose, CA#553108); CD44-APC: 1:500 dilution of 0.2 mg/mL (BD Biosciences #559250);CD24-FITC: 1:500 dilution of 0.5 mg/mL (BD Biosciences #553261). Control sampleswere stained with isotype controls: IgG-APC: (R&D Systems #IC108A); IgG-PE: (BDBiosciences #553930); IgG-APC: (BD Biosciences #553991); IgG-FITC, (BD Biosci-ences #553988). All cell were analysed using BD FACSCalibur (Becton Dickinson,San Jose, CA). FlowJo v. 9.2 (TreeStar, Ashland, OR) was used to analyse the flowcytometry data. The percent positive of each sample relative to the isotype controlswas obtained using Overton subtraction and the Kolmogorov-Smirnov test forsignificance.

2.3. Quantitative RT-PCR

Total RNA was extracted from Sca-1 positive and negative cells using the MiniRNA Isolation II kit (Zymo Research, Irvine, CA) followed by removal of genomicDNA using DNA-free kit (Ambion, Austin, TX). cDNA was synthesised using iScript(BioRad, Hercules, CA) according to the manufacturer’s instructions. QuantitativeRT-PCR of 18s rRNA, TGF-b2, Ly6A (Sca-1), Msi-1, c-myc, Sox2, Wnt-1, and ABCG2

(QuantiTect primer assay and SYBR green PCR kit, Qiagen, Valencia, CA) was carriedout using the ABI 7500 platform. Cycle threshold (Ct) was determined in the expo-nential phase of the amplification curve. Ct of TGF-b2, Ly6A (Sca-1), Msi-1, c-myc,Sox2, Wnt-1, and ABCG2 were normalised to the Ct of 18s ribosomal RNA (DCt).Amplification efficiency of QuantiTect primers are equivalent (http://www.qiagen.-com/Products/PCR/QuantiTect/PrimerAssays) so the DDCt method was used todetermine fold changes. All experiments were carried out in triplicate.

2.4. Western blotting

Whole cells were extracted into lysis buffer (140 mM NaCl, 50 mM Tris pH 7.5–8, 1% triton and protease inhibitor cocktail (Complete EDTA free, Roche, New Zea-land)) and soluble protein collected by centrifugation. Nuclear protein was ex-tracted as follows. One 106 cells were incubated in 400 uL of buffer A (10 mMHEPES-KOH pH7.9, 1.5 mM MgCl2, 10 mM KCl, 0.5 mM DTT) for 10 min on ice, vor-texed for 10 sec and nuclei collected by centrifugation at 13,000g for 10 s. Nucleiwere washed in buffer A with 0.1% Tween-20 for 5 min on ice, centrifuged, thenthe soluble nuclear protein extracted by 20 min incubation on ice in buffer B(20 mM HEPES-KOH ph7.9, 25% glycerol, 420 mM NaCl, 1.5 mM MgCL2, 0.2 mMEDTA, 0.5 mM DTT) followed by centrifugation at 13,000g for 5 min. All superna-tants were quantified using the DC protein assay according to manufacturersinstructions (Bio-Rad, Hercules, CA). 30ug of protein was loaded onto a 10% SDSpolyacrylamide gel, electrophoresed and transferred to PVDF membrane (Bio-Rad,Hercules, CA). The membrane was stained with 0.1% Amido black and digitallyscanned to capture actual protein loading. After blocking in 3% skim milk at roomtemperature, the membrane was incubated with primary antibody (anti-mouseSmad2/3, rabbit polyclonal, #07-408 Millipore, Temecula, CA; at 1:1000 dilution;anti-b-actin, mouse monoclonal, #A5441 Sigma, St. Louis, MO at 1:100,000 dilution;or anti-a-tubulin, rabbit polyclonal, #ab18251 Abcam, Cambridge, UK at 1:4000dilution). Secondary antibodies (goat anti-mouse IgG-HRP #sc-2005, or goat anti-rabbit IgG-HRP #sc-2004 Santa Cruz Biotechnology, Santa Cruz, CA) were used at1:7000 in 3% skim milk. Detection of protein was performed by enhanced chemilu-minescence (Supersignal ECL kit, Pierce) and autoradiography (Kodak X-Omat, Ko-dak, NY).

2.5. Tumour model and lung metastasis quantification

To generate tumours, 4T1 cells suspended in PBS were injected into the mam-mary fat pad of female BALB/c mice, minimum age of 8 weeks were used in allexperiments. Tumour development was monitored twice weekly, using a caliperto measure tumour width and length. After 3 weeks or when tumours reached max-imum allowable size, the mice were sacrificed, and tumours and lungs removed.Metastasis to the lungs was quantified as described by Pulaski and Ostrand-Rosenberg [22]. Briefly, lungs were first chopped mechanically, then incubated withcollagenase (2 mg/mL) and elastase (6U/mL) (Worthington, NJ, USA) in order toobtain a single cell suspension. Cells were serially diluted to 1:10, 1:100 and1:1000 in RPMI media supplemented with 10% fetal calf serum, penicillin/strepto-mycin and 0.01 mg/mL 6-thioguanine (Sigma). Each dilution was then plated andafter 10–14 days incubation in 37 �C 5% CO2, cell colonies were fixed, stained withmethylene blue and counted, each colony representing one metastatic 4T1 cell. Allanimal experiments were approved by the Victoria University Animal EthicsCommittee.

2.6. Quantification of TGFb in supernatants

Parental cells, and cells from dissociated spheres at passage 5 were each platedat 3000 cells/mL into a 6-well plate. After 48 h, conditioned media was removedand the concentration of TGFB1, TGFB2 and TGFB3 secreted into the media wasdetermined with the BioPlex Pro TGFb assay (BioRad, Hercules CA), as directed bythe manufacturer.

3. Results

Transplantable murine breast carcinoma 4T1 cells were ana-lysed for expression of the breast cancer stem cell (BCSC) surfacemarkers CD24, CD44 and Sca-1 by flow cytometry. Expression ofCD44 and CD24 was reproducibly homogenous. All cells expressedhigh levels of both CD24 and CD44 (Supp. Fig. 1A), with no distinctor specific CD44+/CD24�/lo population. In contrast, and consistentwith a putative BCSC sub-population, Sca1 was expressed on�10% (5–15%) of 4T1 cells Supp. Fig. 1B). This was more frequentthan Sca-1+ cells in normal mammary epithelium, which inBALB/c mice is �5% [23]. The distribution of CD24 and CD44 onSca-1+ cells was similar to that on the parental population (Supp.Fig. 1C), supporting the homogenous nature of CD44/24 expression

22 H. Matilainen et al. / Cancer Letters 323 (2012) 20–28

in 4T1. Hence, Sca-1 was chosen as a potential 4T1 BCSC marker forfurther analysis.

3.1. Sca-1 positive cells are not cancer stem cells

The Sca-1+ and Sca-1� 4T1 populations were isolated by flowcytometry to an average of 92% purity (range 89–95%) and 100%purity respectively (Fig. 1A) after which the BCSC activity wasdetermined. Expression of a panel of candidate BCSC genes wasanalysed by quantitative RT-PCR in Sca-1+ and Sca-1� 4T1 cells.SOX2 and Msi1 are embryonic stem cell genes that confer self-

Rel

ativ

e G

ene

Expr

essi

on

A

B

50K

100K

150K

SSC-A

4T1 pre-sort

0 102 103 104 1050

50K

100K

150K

200K

250K

Sca1 PE

SSC

-A

0 102 10

50K

100K

150K

200K

250K

4T1

SSC

-A

Sc

150K

5%

Sca10

20

40

60

80Sca1 +Sca1 -

Rel

ativ

e G

ene

Expr

essi

on

C

D

Fig. 1. Sca-1+ cells have increased self-renewal activity but only modest changes in CSphycoerythrin conjugated antibody (left panel) and flow sorted into Sca-1+ (centre pcytometry and given in the top left hand corner of each graph. Data are from a represnormalised to 18s rRNA and expression level in Sca-1+ cells (grey) expressed relative to th±SD. D. Spheres from Sca-1+ (upper) and Sca-1� (lower) at passage 1 (p1), p4 and p6. S

renewal activity to cells [2,3,24], while TGFB2, Wnt1, ABCG2 (alsoknown as BCRP) and c-myc have previously been described asmarkers of BCSC [25–28]. Sca-1 transcripts were routinely 60-foldhigher in sorted cells with surface Sca-1 protein than in cells with-out (Fig. 1B). However, the majority of the BCSC candidate geneswere only 2–3 fold more highly expressed in Sca-1+ cells relativeto Sca-1� cells, while ABCG2 was weakly or not significantly differ-entially expressed, and Sox2 was not expressed at all (Fig. 1C anddata not shown).

Formation of spheres by tumour cells is often directly correlatedwith a CSC phenotype [4,20]. Sca-1+ and Sca-1� 4T1 cells were

0

1

2

3

TGFB2 Msi1 c-myc Wnt1 AbcG2

Sca1 +Sca1 -

03 104 105

Sca1+

a1 PE

95.2%

4T1 Sca1-

0 102 103 104 1050

50K

100K

150K

200K

250K100%

Sca1 PE

SSC

-A

C gene expression compared to Sca-1� cells. A. Sca-1+ cells were labelled with aanel) or Sca-1� (right panel) populations. Purity was determined by repeat flowentative experiment. B and C. Gene expression was determined by real-time PCR,at of Sca1- cells (white). Data are averaged from 3–5 replicate flow cytometry sorts,

phere score at p1, average ±SEM. Scale bar, 200 lm.

Sca1+ Sca1 -

12/12 (100%)10 000 16/16 (100%)5 000 10/14 (71%) 17/19 (89%)1 000 5/8 (63%)5/8 (63%)

A

0

100

200

300

400

500

0 4 8 12 16 20 24 28 32

Tum

or v

olum

e (m

m3 )

Time (days)

Sca1 +

Sca1 -C D

0

1000

2000

3000

4000

5000

6000

7000

Sca1- Sca1+

Lung

met

asta

tic c

ells

0 5 10 15 20 25 30 350

25

50

75

100

Time (days)

B

Perc

ent

tum

or fr

ee

cell no.Sca1 +Sca1 -

Fig. 2. Sca-1+ and Sca-1� cells behave similarly in tumour assays. A. Initiation of orthotopic tumour formation in the mammary fat pad by cell dose, 3–5 mice per group. Datafrom a total of 5 independent sorts are combined. B. Initiation of tumours over time with 5000 cell dose. Black, Sca-1+ cells; grey, Sca-1� cells, data are sum of severalindependent experiments. C. Average mammary tumour volume using 5000 cell dose, 5–8 mice per group, data represent average ±SD of 3 independent sorts. D.Quantification of spontaneous metastasis to lung from mammary tumours in C. Average colony number per lung, ±SD from 3 independent sorts.

H. Matilainen et al. / Cancer Letters 323 (2012) 20–28 23

cultured in standard serum-free conditions to induce growth ofanchorage-independent spheres. While both Sca-1+ andSca-1� cells formed spheres at the same frequency (�150/1000 cells), the Sca-1� cells predominantly formed tight 3Dspheres with indistinct individual cells (scored as 3). The Sca1+cells did not, instead forming loose spheres with clearly visiblecells (scored as 2), or clusters of cells (scored as 1) (Fig. 1D). At least50 spheres from Sca-1+ and Sca-1� cultures were scored to quan-tify the poor quality of Sca-1+ spheres with an average score of1.73 ± 0.09, compared to the Sca-1� average of 2.65 ± 0.07. Self-renewal, a key stem cell activity, can be measured in vitro by theability to maintain sphere formation over multiple passage, soSca-1+ and Sca-1� spheres were grown for at least 6 passages.Neither the frequency of sphere initiating cells, nor the ‘spherescore’ changed for either population until passage 5, when thenumber of Sca-1� spheres decreased from 150/1000 cells to30/1000 cells. The Sca-1� spheres that did re-form at p5 did notsurvive another passage, and the culture collapsed into a mass ofcellular debris. In contrast, the Sca-1+ cells continued to grow be-yond passage 7 at the same frequency of sphere initiation. Thesedata implied that Sca-1+ cells had greater self-renewal activityin vitro than Sca-1� cells, consistent with the known expressionof Sca-1 on mammary stem and progenitor cells.

Because the gene expression and sphere-forming data in vitroonly weakly supported a BCSC phenotype in Sca-1+ cells, the CSCphenotype of Sca-1+ and Sca-1� cells was determined in vivo byorthotopic transplantation into the mammary fat pad of BALB/cmice. The frequency of tumour initiation, a key CSC assay, wasdetermined in Sca-1+ and Sca-1� cells by titrating the tumourinoculum. With either 10,000, 5000 or 1000 cells (Fig. 2A), or100 cells (no tumours, data not shown), both Sca-1+ and Sca-1�cells initiated tumours in the same proportion of animals, indicat-ing that tumour-initiating cells were not enriched in the Sca-1+population. The time-course of tumour onset demonstrated thatthe rate of initiation was also not significantly different betweenSca-1+ and Sca-1� cells (Fig. 2B). There was a tendency for theSca-1� tumours to initiate slightly, but not significantly, faster

than the Sca-1+ tumours, particularly with larger inoculum sizes(data not shown). Similarly, the growth rate of the Sca-1+ andSca-1� tumours was equivalent similar (Fig. 2C), with the Sca-1�cells growing slightly but not significantly faster. These data indi-cate that Sca-1+ selection did not enrich for tumour-initiating cellsin the 4T1 model.

Metastasis is a key BCSC characteristic. Therefore, spontaneousmetastasis from the mammary fat pad to the lung was quantifiedin a colony-forming assay using 6-thioguanine resistance of 4T1cells as a selective marker [29]. Lungs derived from animals bear-ing both Sca-1+ and Sca-1� tumours contained equal numbers ofmetastatic cells, again with slightly but not significantly more met-astatic cells from the Sca-1� tumours (Fig. 2D).

3.2. Sca-1 is induced in putative cancer stem cells

As determined by gene expression analysis, in vitro and in vivoassays, basal expression of Sca-1+ was not sufficient to confer aBCSC phenotype on 4T1 cells. To test whether a BCSC phenotypecould be induced by Sca-1 up-regulation, 4T1 cells were subjectedto various treatments that induce a stem cell phenotype, and theeffect on Sca-1 expression determined.

Cells that are intrinsically resistant to chemotherapy have CSCcharacteristics [30–32], so Sca-1 expression was examined ondoxorubicin-resistant cells. Intrinsically drug-resistant 4T1 cellswere selected by 16 h exposure to 1 lM doxorubicin, then replatedinto fresh media without drug. This eliminated the majority ofcells, leaving a residual population of intrinsically drug-resistantcells. After surviving cells had re-entered cell cycle and gonethrough one round of division (approximately 5 days), cells wereharvested, RNA extracted and gene expression analysed. Sca-1expression was increased 16–20 fold by this treatment (Fig. 3).Expression of other BCSC-associated genes TgfB2 and AbcG2 wereincreased �5-fold, while expression of the embryonic stem cellgene Sox2 was also induced (Supp. Fig. 2).

Metastatic cells capable of initiating new tumour formation, areby definition CSC, so spontaneously metastatic cells were obtained

Sca10

20

40

60

80

100parentadoxRlung metastassphere

Rel

ativ

e G

ene

Expr

essi

on

Fig. 3. Sca-1 was induced in all putative 4T1 BCSC. Expression of BCSC genes in eachinduced cell population was determined by real-time PCR and expressed relative tothat in parental cells. Data are averaged from 3 independent experiments, ±SD.

24 H. Matilainen et al. / Cancer Letters 323 (2012) 20–28

from the lungs of tumour-bearing mice by short-term culture.Gene expression analysis after 4 days selection indicated thatSca-1 was 35–40 fold higher in cells that had metastasised to lung,than in parental cells (Fig 3) or in cells selected from the primarytumour (similar to parental cells, data not shown). Unlikechemo-resistant cells, metastatic cells did not express Sox2 (Supp.Fig. 2).

Third, the effect of sphere formation on Sca-1 expression wasdetermined. Spheres were passaged at least 5 times to ensure thatsphere formation was stably maintained before analysis of Sca-1expression, which was induced approximately 100-fold. Well-formed, tight spheres made up the majority of the 4T1 sphere cul-ture, with only small numbers of slowly dividing non-spheroidcells present (Fig. 4A). Sca-1 transcript increased 60–100-fold inspheres at passage 4, 6, and 8 (Fig. 4B), with high expression con-sistently maintained over 15 passages (data not shown). To con-firm induction of Sca-1 in spheres, protein level was determinedby flow cytometry. Results indicated that the number of Sca-1+cells increased from�5% in the parental cells, to 75–95% in spheres(Fig. 4C).

3.3. Sca-1 positive sphere cells are not metastatic cancer stem cells

Despite robust induction and continued expression of Sca-1,4T1 spheres showed no other evidence of having acquired a BCSC

Tgfb20.0

0.5

1.0

1.5

2.0

50

100

150

200

Rel

ativ

e G

ene

Expr

essi

on

A B

C

0 102 103 104 105

Sca1-PE

0

20

40

60

80

100

% o

f Max

Fig. 4. Sphere formation induced Sca-1 but decreased BCSC gene expression. A. Sphenormalised to 18s rRNA and expressed relative to the level in parental cells at equivalentat least 3 independent replicate experiments. C. Sca1 expression determined by flow cyparental cells; grey line, expression in sphere cells. Data are representative of at least 3

phenotype. Expression of c-myc and Wnt1 were essentially un-changed in spheres regardless of passage, while Msi1 transcript de-creased to half the parental level over 8 passages, and TGFb2 andAbcG2 transcripts rapidly decreased to less than 1% of the levelin parental cells (Fig. 4B). While surprising, these data wereabsolutely consistent with behaviour of these cells in vivo. Titrationof cell dose indicated no observable difference in the frequency oftumour initiating cells (Fig. 5A and data not shown), nor in thegrowth rate of 4T1 sphere-derived tumours (Fig. 5B) compared toparental cells, at any inoculum. However, there was a dramatic97% (32-fold) loss of metastasis to the lung from sphere-derivedtumours (Fig. 5C). Together these data implied that a BCSC pheno-type was actually lost in spheres, rather than induced, despite thehigh expression of Sca-1.

3.4. TGFb signalling is altered in spheres

Loss of the BCSC phenotype and metastasis correlated with lossof transcript of TGFB2. Because the TGFb signalling pathway is crit-ical for breast cancer metastasis [33], components of the TGFb/SMAD pathway were compared between spheres and parental4T1 cells. Secretion of TGFB1, 2 and 3 was measured in superna-tants from sphere and parental cells (Fig. 6A). TGFB1 was producedat high concentration by parental 4T1 cells (average 6600 pg/mL),and was reduced by 80% in spheres (average 1380 pg/mL). TGFB2and TGFB3 were present at much lower levels in parental cellsupernatants, but were also reduced in sphere supernatants, par-ticularly TGFB3 which was reduced by 50%.

Expression of the ligand-binding type II receptor, TGFBR2, wasassessed by flow cytometry. Levels were variable between experi-ments, but essentially equivalent between parental 4T1 cells andspheres, with on average 32% of parental and 31% of spheresexpressing TGFBR2 on the cell surface (Fig. 6B). Activation of theTGFBR complex and signal transduction results in the phosphory-lation of SMAD2, and/or SMAD3, dimerisation with SMAD4 andtranslocation into the nucleus. The nuclear fraction was collectedfrom 4T1 spheres and parental cells, and SMAD2/3 content andlocalisation analysed by Western blot (Fig. 6C). In parental cells,SMAD2 and SMAD3 were expressed at similar levels in whole celllysates, and both were present in the nucleus, where SMAD3 wasmore abundant than SMAD2. In contrast, spheres had lost themajority of SMAD3 in the whole cell lysate compared to parentalcells while SMAD2 (upper band, asterisk) was undetectable. Therewas also extensive loss of nuclear SMAD2 in spheres, consistent

c-myc Wnt1 Msi1 AbcG2 Sca1

parental

sphere p4

sphere p6

sphere p8

re formation at passage 6. B. Gene expression was determined by real-time PCR,passage, either p4 (white), p6 (light grey) or p8 (dark grey). Data are average ±SD oftometry at passage 6. Grey shaded curve, isotype control. Black line, expression inindependent sphere lines. all showed similar Sca-1 expression.

0 5 10 15 20 250

25

50

75

100

sphere parental

Perc

ent t

umor

free

0

100

200

300

400

500

600

0 5 10 15 20 25Time (days)

parentalsphere

Tum

our v

olum

e (m

m3 )

0

1

2

3

4

5

6

aver

age

met

asta

tic lu

ng c

olon

ies

(103 )

parental sphere

Time (days)

A

B

C

Fig. 5. Sphere formation reversed spontaneous metastasis of 4T1 cells. A. Initiationof orthotopic mammary tumours, B. growth of tumours in mammary fat pad, and C.metastasis from mammary tumours. 6–8 mice per group were inoculated with10,000 parental (black) or sphere (grey) cells. Data are average ±SD within a group,and representative of 3 independent experiments.

H. Matilainen et al. / Cancer Letters 323 (2012) 20–28 25

with the overall SMAD2 reduction. Despite the lower basal level ofSMAD3 in spheres, a similar amount of nuclear SMAD3 was ob-served in spheres as in parental cells, implying that some TGFb sig-nalling was retained in spheres.

As Sca-1 has been found to be associated with the TGFb receptorcomplex [18], the potential for co-expression of TGFBR2 and Sca-1on spheres was determined by flow cytometry (Fig. 6D). Incubationwith both antibodies reduced the intensity of signal for each pro-tein (data not shown), suggesting the two proteins are in closephysical proximity. Under these conditions, few parental cellscould be identified as positive for both Sca-1 and TGFBR2. Withspheres, cells with little or no Sca-1 were negative for TGFBR2. Incontrast, most TGFbR2 + cells were in the Sca-1+ population, sug-

gesting that the physical and functional interaction between theseproteins could be conserved.

3.5. Restoration of TGFB2 correlated with restoration of metastasis

Given that spheres retained expression of TGFBR2, P5 sphereswere re-differentiated by exposure to serum, a major source ofTGFb ligand. This rapidly led to adherent 2D growth and re-expression of TGFB2, which returned to the same or higher levelas parental cells. Sca-1 transcripts decreased a little upon re-differentiation, but were maintained at a level �20-fold higherthan in parental cells (Fig. 7A). The growth of re-differentiated cellsin vivo was also examined. Primary tumour initiation and growthwere not statistically different between parental and re-differenti-ated cells (data not shown). However, and consistent with the res-toration of TGFB2 expression, re-differentiated cells had regainedmetastatic ability, with an average 60-fold increase in metastasiscompared to the spheres from which they were derived (Fig. 7B).

4. Discussion

Murine mammary cells that express the stem cell antigen Sca-1have been shown to exhibit properties of mammary stem cells[34]. Therefore the question of whether Sca-1-expressing 4T1 cellssimilarly exhibit breast cancer stem cell properties was deter-mined. Sca-1 was expressed on a subset of �10% of unstimulated4T1 cells, which was higher than expression on normal mammaryepithelial cells, and isolated Sca-1+ cells had greater self-renewalactivity than Sca-1� cells, consistent with the role of Sca-1 as astem cell marker. However, Sca-1+ cells did not have a BCSC phe-notype, as determined by stem cell gene expression, tumour initi-ation and spontaneous metastasis assays. When Sca-1 was highlyinduced on 95% of 4T1 cells by culture as tumour spheres, spherecells did not assume a BCSC phenotype but instead showed lossof expression of BCSC genes associated with stem cell self-renewal(Msi1), drug resistance (ABCG2/BCRP) and metastasis (TGFB2). Fur-ther, the loss of BCSC gene expression in 4T1 spheres correlatedwith the near-complete loss of spontaneous metastasis fromsphere-derived tumours. Loss of metastasis in spheres was associ-ated with alterations in the TGFb signalling pathway in spheres,specifically decreased TGFb ligand expression and decreased nucle-ar localisation of SMAD2. Both TGFB2 expression and spontaneousmetastasis could be restored by re-differentiation of spheres byexposure to serum, while retaining significant Sca-1 expression,thus dissociating Sca-1 expression from the metastatic properties.

Given that Sca-1 expression did not enrich for any BCSC charac-teristic, we conclude that is not an effective marker of a BCSC in4T1 cells. This conflicts with a recent report from Kim et al thatSca-1+ 4T1 cells form better spheres in vitro, and larger tumoursin vivo, than Sca-1� cells [23]. It is difficult to directly comparethe two studies. The choice of media used for sphere formationcan impact on the phenotype of the resulting spheres. We noteddifferences in both the efficiency of sphere formation and theappearance of spheres with different media formulations. The opti-mal media was chosen for our study based on appearance andnumber of spheres (data not shown), which was not the samemedia as Kim et al [23]. More significantly, the 4T1 cell line usedby Kim et al appears to be different. The authors indicate that therewas no migration to other sites from the primary tumours, whilesignificant metastasis to lung and liver was routinely observed inthis study. This difference could be due to the inadvertent choiceof a line of 4T1 cells with less metastatic potential than one usedin this study, or drifting of the phenotype of the 4T1 cells over ex-tended passage. Phenotype drift was avoided in this study byrestricting cells to less than 15 passages from the original aliquot

TGFB1 TGFB2 TGFB30

204060

2000

4000

6000

8000

100 101 102 103 104

TGFBR2-APC

0

20

40

60

80

100

% o

f Max

parental

100 101 102 103 104

TGFBR2-APC

0

20

40

60

80

100

% o

f Max

sphere

A

WCL N

αSMAD2/3

parental sphere

αTUBA1

40 50

60 80

WCL N

*

40 50

60 80

C

100 101 102 103 104

TGFBR2-APC

100

101

102

103

104

Sca1

-PE

1.67 0.161

2.8595.3

100 101 102 103 104

TGFBR2-APC

100

101

102

103

104

Sca1

-PE

(Y58

5/42

)

62.4 6.96

0.96429.7

sphereparental

B

D

TGFB

(pg/

mL)

parentalsphere

Fig. 6. Spheres have alterations in the TGFb signalling pathway. A. Secretion of TGFB1, 2 and 3 from parental (black) or sphere (white) cells was measured in triplicate andnormalised to the media only control. Data are average ±SD. B. Expression of TGFBR2 on the surface of parental (left) and sphere (right) cells compared to isotype control(shaded grey curve). C. Western blot in whole cell lysate (WCL) and nuclear extracts (N) for SMAD2/3 (top panel) and alpha-tubulin (bottom panel). SMAD2 indicated byasterisk, MW 56–60 kDa. SMAD3 MW 48–52 kDa. Position of MW markers shown on right. D. Dual staining for Sca-1 (y-axis) and TGFBR2 (x-axis) in parental (left panel) andsphere (right panel) cells. Percent of cells in each quadrant is indicated.

26 H. Matilainen et al. / Cancer Letters 323 (2012) 20–28

obtained from ATCC, and maintaining cells at or below optimal 70%confluency at all times. Regardless of how the difference arose, thedifference in metastatic ability makes it difficult to directly com-pare the two studies.

Based on recently published data in normal murine mammarycells, Sca-1 function is most likely conserved in 4T1 cells. In normalmammary cells, Sca-1 is physically associated with TGFBRI, onehalf of the TGFb receptor complex, blocking TGFb signal transduc-tion. TGFb signalling is tumour suppressive in normal cells, butonce tumours are established TGFb promotes invasion and metas-tasis [35]. Sca-1 was induced during carcinogen treatment, whereit blocked the tumour-suppressive function of TGFb signalling,contributing to tumour development in response to carcinogen[18]. Since TGFb signalling mediates metastasis in 4T1 tumours[36], our data are consistent with the probability that Sca-1 expres-sion blocks TGFb signalling in 4T1 spheres in vitro and in vivo.While Sca-1 probably marks a BCSC during the initial phases of tu-mour genesis, the present data do not support the hypothesis thatSca-1 marks a BCSC in established tumours, where TGFb signallingpromotes tumour progression.

The role of Sca-1 in tumorigenesis and metastasis is not yetclear. In general, Sca-1 up-regulation is associated with a moreaggressive tumour phenotype [37]. Similarly, in mouse modelsSca-1 has been shown to be associated with greater tumourigenicpotential [38]. However, there are reports that Sca-1 can act to re-strict cell growth in a mammary tumour model [39]. Moreover, thelevel of Sca-1 protein on the cell surface may be important for reg-ulating function [34]. It is possible that Sca-1 has different effectson tumour initiation and progression that vary in different tumoursystems. This could potentially be related to the range of TGFb li-gands expressed in each system, and their precise role, althoughthis has not been explored.

While the physical and functional interactions between Sca-1and TGFb in spheres were not explicitly examined in this study,

our data show that the Sca-1+ and TGFBR2+cells in 4T1 spheresare largely concordant, suggesting that the proteins could be co-localised and possibly interact. There is also potential cross-regulation between the two pathways – Sca-1 on the surface ofspheres might block TGFBR2 function [18], but the reduction inTGFb signalling might have up-regulated Sca-1 expression inspheres, as has been observed in normal mammary glands withloss of TGFb [40].

One interesting possibility raised by these data concerns the dif-ferential role of SMAD2 and SMAD3 in 4T1 metastasis. TGFb ligandbinding to the TGFb receptor results in phosphorylation of SMAD2and/or SMAD3, which then dimerise with SMAD4 and translocateto the nucleus. Spheres had lost a large proportion of SMAD2, withvery little nuclear SMAD2 in evidence. While SMAD3 signalling ap-peared to be retained in spheres, based on nuclear translocation ofSMAD3, it was apparently not sufficient to mediate spontaneousmetastasis from sphere-derived tumours. With this distinctexpression pattern, our spheres provide an ideal model for furtherdifferential analysis of SMAD2 and SMAD3 in 4T1 metastasis.

Many researchers use spheres as a direct correlate of a cancerstem cell phenotype, and serial sphere formation has commonlybeen invoked, by ourselves [31] and many others, as evidence ofa self-renewing stem-like cell. Many cancer cell lines, both humanand murine, have been grown as spheres and exhibit induction ofstem cell gene expression, chemotherapy resistance and invasivetumour phenotype, including the breast cancer cell lines MCF7,MDA361 and BT474 [41]. However, our present data clearly dem-onstrate that self-renewal activity does not necessarily equate withsphere formation or cancer stem cell activity. A separation be-tween cancer stem cell phenotype and sphere formation has beenrecently reported in the brain tumour glioblastoma multiforme[42], and similarly we have observed in melanoma that stem cellactivity is not a pre-requisite for tumour initiation or maintenance(Grasso et al, manuscript in preparation). Self-renewal has been a

TGFB20.0

0.5

1.0

1.5

2.0

20

40

60

80R

elat

ive

Gen

e Ex

pres

sion

rediff p4

parentalsphere p8

Sca1

0.38

23.1

0102030405060708090

100

Met

asta

sis

(per

cent

of p

aren

tal)

parental sphere p8 rediff p4

A

B

Fig. 7. Restoration of TGFB2 expression correlates with restoration of metastasis. A.Gene expression in spheres (white) or re-differentiated spheres (hatched) wasdetermined by real-time PCR, normalised to 18s rRNA and expressed relative toparental cells. Data are average of 2 replicate experiments, ±SD. B. Mammarytumours were established with parental (black), sphere (white) or re-differentiated(grey) cells, 8 mice per group. Spontaneous metastasis to lung was quantified, andmetastasis of each experimental normalised to that of parental cells (100%),percentage given above bar. Average of 2 independent experiments.

H. Matilainen et al. / Cancer Letters 323 (2012) 20–28 27

central tenet of the cancer stem cell hypothesis, and serial sphereformation a key assay of self-renewal, so we suggest caution ininterpretation of sphere experiments while this finding is validatedby others.

Our initial aim was to identify a BCSC marker in 4T1 cells thatcould be used to model immune responses against metastatic can-cer stem cells. While we have demonstrated that neither sphereformation, CD44/24 nor Sca-1 expression are suitable to establisha 4T1 cancer stem cell model, the question remains. While expres-sion of a marker on the surface of a cell would allow a simple can-cer stem cell selection process, functional assays such as intrinsicchemoresistance [31], aldehyde dehydrogenase expression [43],and selection of quiescent cells by label retention [44] have shownpromise for breast cancer stem cell studies, as well as other tumourtypes. Since intrinsically drug resistant 4T1 cells expressed Sox2,the transcription factor in embryonic stem cells responsible forself-renewal, we are currently pursuing chemo-resistant 4T1 cellsas a functional marker of BCSC that can be used in immune compe-tent mouse models of immunotherapy.

Acknowledgements

We would like to thank Kate Broadley and Kathryn Farrand fortechnical assistance, and the Cancer Society of New Zealand forsupport (XY). The work was funded by a grant from the Breast Can-cer Research Trust (BCRT) to MB and MJM. BCRT had no involve-ment in study design, data interpretation or preparation of themanuscript.

Appendix A. Supplementary data

Supplementary data associated with this article can be found, inthe online version, at http://dx.doi.org/10.1016/j.canlet.2012.03.028.

References

[1] S. Bomken, K. Fiser, O. Heidenreich, J. Vormoor, Understanding the cancer stemcell, Br. J. Cancer 103 (2010) 439–445.

[2] V. Kashyap, N.C. Rezende, K.B. Scotland, S.M. Shaffer, J.L. Persson, L.J. Gudas,N.P. Mongan, Regulation of stem cell pluripotency and differentiation involvesa mutual regulatory circuit of the NANOG, OCT4, and SOX2 pluripotencytranscription factors with polycomb repressive complexes and stem cellmicroRNAs, Stem Cells Dev. 18 (2009) 1093–1108.

[3] O. Leis, A. Eguiara, E. Lopez-Arribillaga, M.J. Alberdi, S. Hernandez-Garcia, K.Elorriaga, A. Pandiella, R. Rezola, A.G. Martin, Sox2 expression in breasttumours and activation in breast cancer stem cells, Oncogene (2011).

[4] M.A. Velasco-Velazquez, V.M. Popov, M.P. Lisanti, R.G. Pestell, The role ofbreast cancer stem cells in metastasis and therapeutic implications, Am. J.Pathol. 179 (2011) 2–11.

[5] T. Lapidot, C. Sirard, J. Vormoor, B. Murdoch, T. Hoang, J. Caceres-Cortes, M.Minden, B. Paterson, M.A. Caligiuri, J.E. Dick, A cell initiating human acutemyeloid leukaemia after transplantation into SCID mice, Nature 367 (1994)645–648.

[6] H. Nakshatri, E.F. Srour, S. Badve, Breast cancer stem cells and intrinsicsubtypes: controversies rage on, Curr. Stem Cell Res. Ther. 4 (2009) 50–60.

[7] M.O. Idowu, M. Kmieciak, C. Dumur, R.S. Burton, M.M. Grimes, C.N. Powers,M.H. Manjili, CD44(+)/CD24(�/low) cancer stem/progenitor cells are moreabundant in triple-negative invasive breast carcinoma phenotype and areassociated with poor outcome, Hum. Pathol. (2011).

[8] Z. Wang, Q. Shi, Y. Gu, Y. Shen, M. Sun, M. Deng, H. Zhang, J. Fang, S. Zhang, F.Xie, Clinicopathologic correlation of cancer stem cell markers CD44, CD24,VEGF and HIF-1alpha in ductal carcinoma in situ and invasive ductalcarcinoma of breast: an immunohistochemistry-based pilot study, Pathol.Res. Pract. 207 (2011) 505–513.

[9] S. Ricardo, A.F. Vieira, R. Gerhard, D. Leitao, R. Pinto, J.F. Cameselle-Teijeiro, F.Milanezi, F. Schmitt, J. Paredes, Breast cancer stem cell markers CD44, CD24and ALDH1: expression distribution within intrinsic molecular subtype, J. Clin.Pathol. 64 (2011) 937–946.

[10] K.H. Wang, A.P. Kao, C.C. Chang, J.N. Lee, M.F. Hou, C.Y. Long, H.S. Chen, E.M.Tsai, Increasing CD44+/CD24(-) tumor stem cells, and upregulation of COX-2and HDAC6, as major functions of HER2 in breast tumorigenesis, Mol Cancer 9(2010) 288.

[11] H.J. Kim, M.J. Kim, S.H. Ahn, B.H. Son, S.B. Kim, J.H. Ahn, W.C. Noh, G. Gong,Different prognostic significance of CD24 and CD44 expression in breastcancer according to hormone receptor status, Breast 20 (2011) 78–85.

[12] M. Huang, Y. Li, H. Zhang, F. Nan, Breast cancer stromal fibroblasts promote thegeneration of CD44+CD24- cells through SDF-1/CXCR4 interaction, J. Exp. Clin.Cancer Res. 29 (2010) 80.

[13] T. Blick, H. Hugo, E. Widodo, M. Waltham, C. Pinto, S.A. Mani, R.A. Weinberg,R.M. Neve, M.E. Lenburg, E.W. Thompson, Epithelial mesenchymal transitiontraits in human breast cancer cell lines parallel the CD44(hi/)CD24 (lo/-) stemcell phenotype in human breast cancer, J. Mammary Gland Biol. Neoplasia 15(2010) 235–252.

[14] M. Al-Hajj, M.S. Wicha, A. Benito-Hernandez, S.J. Morrison, M.F. Clarke,Prospective identification of tumorigenic breast cancer cells, Proc. Natl. Acad.Sci. USA 100 (2003) 3983–3988.

[15] C. Grange, S. Lanzardo, F. Cavallo, G. Camussi, B. Bussolati, Sca-1 identifies thetumor-initiating cells in mammary tumors of BALB-neuT transgenic mice,Neoplasia 10 (2008) 1433–1443.

[16] B.J. Classon, L. Coverdale, Mouse stem cell antigen Sca-2 is a member of the Ly-6 family of cell surface proteins, Proc. Natl. Acad. Sci. USA 91 (1994) 5296–5300.

[17] B.E. Welm, S.B. Tepera, T. Venezia, T.A. Graubert, J.M. Rosen, M.A. Goodell, Sca-1(pos) cells in the mouse mammary gland represent an enriched progenitorcell population, Dev. Biol. 245 (2002) 42–56.

[18] G. Upadhyay, Y. Yin, H. Yuan, X. Li, R. Derynck, R.I. Glazer, Stem cell antigen-1enhances tumorigenicity by disruption of growth differentiation factor-10(GDF10)-dependent TGF-beta signaling, Proc. Natl. Acad. Sci. USA 108 (2011)7820–7825.

[19] S.B. Bradfute, T.A. Graubert, M.A. Goodell, Roles of Sca-1 in hematopoieticstem/progenitor cell function, Exp. Hematol. 33 (2005) 836–843.

[20] F. Hirschhaeuser, H. Menne, C. Dittfeld, J. West, W. Mueller-Klieser, L.A. Kunz-Schughart, Multicellular tumor spheroids: an underestimated tool is catchingup again, J. Biotechnol. 148 (2010) 3–15.

[21] R.J. Ward, P.B. Dirks, Cancer stem cells: at the headwaters of tumordevelopment, Annu. Rev. Pathol. 2 (2007) 175–189.

[22] B.A. Pulaski, S. Ostrand-Rosenberg, Mouse 4T1 Breast Tumor Model, JohnWiley & Sons, Inc., 2001.

[23] R.J. Kim, S.R. Kim, K.J. Roh, S.B. Park, J.R. Park, K.S. Kang, G. Kong, B. Tang, Y.A.Yang, E.A. Kohn, L.M. Wakefield, J.S. Nam, Ras activation contributes to themaintenance and expansion of Sca-1pos cells in a mouse model of breastcancer, Cancer Lett. 287 (2010) 172–181.

28 H. Matilainen et al. / Cancer Letters 323 (2012) 20–28

[24] R.I. Glazer, X. Wang, H. Yuan, Y. Yin, Mammary stem and progenitor cellregulation, Cancer Biomark. 3 (2007) 171–181.

[25] F. Staud, P. Pavek, Breast cancer resistance protein (BCRP/ABCG2), Int. J.Biochem. Cell Biol. 37 (2005) 720–725.

[26] A.P. Morel, M. Lievre, C. Thomas, G. Hinkal, S. Ansieau, A. Puisieux, Generationof breast cancer stem cells through epithelial-mesenchymal transition, PLoSONE 3 (2008) e2888.

[27] M. Santisteban, J.M. Reiman, M.K. Asiedu, M.D. Behrens, A. Nassar, K.R. Kalli, P.Haluska, J.N. Ingle, L.C. Hartmann, M.H. Manjili, D.C. Radisky, S. Ferrone, K.L.Knutson, Immune-induced epithelial to mesenchymal transition in vivogenerates breast cancer stem cells, Cancer Res. 69 (2009) 2887–2895.

[28] Y. Li, B. Welm, K. Podsypanina, S. Huang, M. Chamorro, X. Zhang, T. Rowlands,M. Egeblad, P. Cowin, Z. Werb, L.K. Tan, J.M. Rosen, H.E. Varmus, Evidence thattransgenes encoding components of the Wnt signaling pathway preferentiallyinduce mammary cancers from progenitor cells, Proc. Natl. Acad. Sci. USA 100(2003) 15853–15858.

[29] B.A. Pulaski, S. Ostrand-Rosenberg, Mouse 4T1 breast tumor model, Curr.Protoc. Immunol. Chapter 20 (2001) (Unit 20 22).

[30] M. Dean, T. Fojo, S. Bates, Tumour stem cells and drug resistance, Nat. Rev.Cancer 5 (2005) 275–284.

[31] K.W. Broadley, M.K. Hunn, K.J. Farrand, K.M. Price, C. Grasso, R.J. Miller, I.F.Hermans, M.J. McConnell, Side population is not necessary or sufficient for acancer stem cell phenotype in glioblastoma multiforme, Stem. Cells 29 (2011)452–461.

[32] S.L. Cooke, J. Temple, S. Macarthur, M.A. Zahra, L.T. Tan, R.A. Crawford, C.K. Ng,M. Jimenez-Linan, E. Sala, J.D. Brenton, Intra-tumour genetic heterogeneity andpoor chemoradiotherapy response in cervical cancer, Br. J. Cancer 104 (2011)361–368.

[33] J. Massague, TGFbeta in Cancer, Cell 134 (2008) 215–230.[34] C. Holmes, W.L. Stanford, Concise review: stem cell antigen-1: expression,

function, and enigma, Stem Cells 25 (2007) 1339–1347.

[35] H. Ikushima, K. Miyazono, TGFbeta signalling: a complex web in cancerprogression, Nat. Rev. Cancer 10 (2010) 415–424.

[36] R.S. Muraoka, N. Dumont, C.A. Ritter, T.C. Dugger, D.M. Brantley, J. Chen, E.Easterly, L.R. Roebuck, S. Ryan, P.J. Gotwals, V. Koteliansky, C.L. Arteaga,Blockade of TGF-beta inhibits mammary tumor cell viability, migration, andmetastases, J. Clin. Invest. 109 (2002) 1551–1559.

[37] I.P. Witz, Differential expression of genes by tumor cells of a low or a highmalignancy phenotype: the case of murine and human Ly-6 proteins, J. Cell.Biochem. Suppl. 34 (2000) 61–66.

[38] H. Yuan, G. Upadhyay, Y. Yin, L. Kopelovich, R.I. Glazer, Stem cell antigen-1deficiency enhances the chemopreventive effect of peroxisome proliferator-activated receptorgamma activation, Cancer Prev. Res. (Phila.) 5 (2012) 51–60.

[39] T.D. Batts, H.L. Machado, Y. Zhang, C.J. Creighton, Y. Li, J.M. Rosen, Stem cellantigen-1 (sca-1) regulates mammary tumor development and cell migration,PLoS ONE 6 (2011) e27841.

[40] K. Roarty, S.E. Baxley, M.R. Crowley, A.R. Frost, R. Serra, Loss of TGF-beta orWnt5a results in an increase in Wnt/beta-catenin activity and redirectsmammary tumour phenotype, Breast Cancer Res. 11 (2009) R19.

[41] Y. Wang, Y. Yu, A. Tsuyada, X. Ren, X. Wu, K. Stubblefield, E.K. Rankin-Gee, S.E.Wang, Transforming growth factor-beta regulates the sphere-initiating stemcell-like feature in breast cancer through miRNA-181 and ATM, Oncogene 30(2011) 1470–1480.

[42] L.E. Barrett, Z. Granot, C. Coker, A. Iavarone, D. Hambardzumyan, E.C. Holland,H.S. Nam, R. Benezra, Self-renewal does not predict tumor growth potential inmouse models of high-grade glioma, Cancer Cell 21 (2012) 11–24.

[43] M.R. Alison, N.J. Guppy, S.M. Lim, L.J. Nicholson, Finding cancer stem cells: arealdehyde dehydrogenases fit for purpose?, J Pathol. 222 (2010) 335–344.

[44] N. Moore, S. Lyle, Quiescent slow-cycling stem cell populations in cancer: areview of the evidence and discussion of significance, J. Oncol. 2011(2011).