CD200-expressing human basal cell carcinoma cellsinitiate tumor growthChantal S. Colmonta, Antisar BenKetaha, Simon H. Reedb, Nga V. Hawkc, William G. Telfordc, Manabu Ohyamad,Mark C. Udeye, Carole L. Yeee, Jonathan C. Vogele,1, and Girish K. Patela,2

aDepartment of Dermatology and Wound Healing, School of Medicine, Cardiff University, Cardiff CF14 4XN, United Kingdom; bDepartment of MedicalGenetics, Haematology and Pathology, School of Medicine, Cardiff University, Cardiff CF14 4XN, United Kingdom; cExperimental Transplantation andImmunology Branch, Center for Cancer Research, National Cancer Institute, National Institutes of Health, Bethesda, MD 20892; dDepartment of Dermatology,Keio University School of Medicine, Tokyo 160-8582, Japan; and eDermatology Branch, Center for Cancer Research, National Cancer Institute, NationalInstitutes of Health, Bethesda, MD 20892

Edited* by Douglas R. Lowy, National Cancer Institute, Bethesda, MD, and approved November 29, 2012 (received for review July 9, 2012)

Smoothened antagonists directly target the genetic basis of humanbasal cell carcinoma (BCC), the most common of all cancers. Thesedrugs inhibit BCC growth, but they are not curative. Although BCCcells are monomorphic, immunofluorescence microscopy revealsa complex hierarchical pattern of growthwith inward differentiationalong hair follicle lineages. Most BCC cells express the transcriptionfactor KLF4 and are committed to terminal differentiation. A smallCD200+ CD45− BCC subpopulation that represents 1.63 ± 1.11% ofall BCC cells resides in small clusters at the tumor periphery. By usingreproducible in vivo xenograft growth assays, we determined thattumor initiating cell frequencies approximate one per 1.5 million un-sorted BCC cells. The CD200+ CD45− BCC subpopulation recreatedBCC tumor growth in vivo with typical histological architecture andexpression of sonic hedgehog-regulated genes. Reproducible in vivoBCC growthwas achievedwith as few as 10,000 CD200+ CD45− cells,representing ∼1,500-fold enrichment. CD200− CD45− BCC cells wereunable to form tumors. These findings establish a platform to studythe effects of Smoothened antagonists on BCC tumor initiating celland also suggest that currently available anti-CD200 therapy be con-sidered, either as monotherapy or an adjunct to Smoothened antag-onists, in the treatment of inoperable BCC.

mouse model | skin cancer

Skin cancer is the most common of all malignancies, with >3.5million new cases diagnosed in the United States each year.

Seventy percent of skin cancers are basal cell carcinomas (BCCs)(1). An autosomal-dominant genetic disease resulting in BCC, basalcell nevus syndrome (i.e., Gorlin syndrome), is caused by germ-linemutations in the human homologue of patched 1 (PTCH1) re-ceptor, a component of the sonic hedgehog (SHH) growth factorsignaling pathway (2). Tissue-specific somatic mutation of the nor-mal PTCH1 allele in basal cell nevus syndrome leads to multipleBCCs, medulloblastomas, meningiomas, and rhabdomyosarcomas.PTCH1 encodes a transmembrane protein that, in the absence ofligand binding, inhibits the constitutively active G protein-coupledmembrane protein Smoothened (SMO). After binding SHH,PTCH1 fails to repress SMO, resulting in translocation of theKruppel-related zinc finger transcription factor Gli family mem-bers (GLI1 andGLI2) to the nucleus and subsequent expression ofhedgehog-regulated genes (e.g., GLI 1, K17, PDGFRα, PTCH1,BCL2). Downstream members of the SHH signaling pathway(SHH, SMO, GLI1, and GLI2) result in BCC when constitutivelyexpressed in murine skin or in human skin grafted onto mice,confirming the pivotal role of the SHH signaling pathway in BCCformation (3–9). Sporadic BCC tumors also harbor inactivatingmutations in PTCH1 (80–90%) or activating mutations in SMO(10–20%) (9). Together, these finding indicate that BCCs arisefrom constitutive activation of the SHH growth factor signalingpathway in keratinocytes.The hedgehog pathway is essential during embryogenesis but is

quiescent during adulthood, remaining active in only a few renewingadult tissues including hair follicles, bone marrow, and intestinalcrypts. Ligand-independent and ligand-dependent (autocrine and

paracrine) reactivation of the SHH pathway occurs in many tumors,including gastrointestinal, prostate, hematological, and neural can-cers (10–13). Because of the abundance and accessibility of tumortissues, BCC represents an attractive model to study therapeuticagents targeting the SHH pathway. A number of SMO inhibitorsare currently under development, and at least three are in clinicaltrials: GDC-0449/vismodegib (Genentech), LDE225/erismodegib(Novartis), and IPI-926/saridegib (Infinity). The Food and DrugAdministration has approved the use of vismodegib in the UnitedStates for the treatment of adults with metastatic BCC or locallyadvanced BCC that has recurred following surgery or who are notcandidates for surgery or radiation. Preclinical studies indicatethat these drugs are potent SMO antagonists, blocking ligand-dependent and ligand-independent activation (14–16). However,recent clinical studies demonstrated that, although patients withadvanced BCC can experience dramatic responses, BCC cellspersisted during treatment and retained the potential to regrow.One possible explanation is that tumor initiating cells (TICs) existwithin BCC, and that they are resistant to elimination by SMOantagonists (17–21).Our recent success in identifying primary human cutaneous

squamous cell carcinoma (SCC) TICs encouraged us to search forTICs in primary human BCC. In this report, we show that BCCcells differentiate along hair follicle lineages and that a smallsubpopulation of relatively undifferentiated BCC cells expressesthe human hair follicle bulge stem cell marker CD200. By using invitro and in vivo assays, we demonstrate that the CD200+ BCCpopulation is enriched for in vitro colony forming ability andcontains TICs that can recreate BCC growth in vivo.

ResultsHuman BCC Express Hair Follicle Differentiation-Specific Keratins.BCCstypically arise on hair-bearing skin, and BCC cells resemble basalcells of the hair follicle outer root sheath (ORS), explaining thename and presumed origin of this tumor (Fig. 1 A and B). Like-wise, BCC tumors from transgenic mouse models also demon-strate hair follicle differentiation, even though lineage tracingexperiments are divided as to the cell of origin between hair fol-licle bulge stem cells (22–26) and interfollicular epidermal cells(27). The process of hair growth is carefully choreographed andhair follicles consist of concentric cell layers characterized bydistinct patterns of hair follicle specific keratin heterodimer

Author contributions: S.H.R., M.O., M.C.U., J.C.V., and G.K.P. designed research; C.S.C.,A.B., C.L.Y., and G.K.P. performed research; S.H.R., N.V.H., and W.G.T. contributed newreagents/analytic tools; C.S.C., A.B., S.H.R., W.G.T., M.O., M.C.U., C.L.Y., J.C.V., and G.K.P.analyzed data; and C.S.C., S.H.R., M.C.U., and G.K.P. wrote the paper.

The authors declare no conflict of interest.

*This Direct Submission article had a prearranged editor.

Freely available online through the PNAS open access option.1Deceased October 30, 2010.2To whom correspondence should be addressed. E-mail: patelgk@cardiff.ac.uk.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1211655110/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1211655110 PNAS Early Edition | 1 of 6

MED

ICALSC

IENCE

S

expression during each step toward terminal differentiation (28).We first sought to determine if humanBCC expressed hair follicle-specific keratins by using RT-PCR and immunofluorescence toassess the extent of differentiation that would support the cancerstem cell model and the potential existence of TICs.All human BCCs studied (N = 20) contained cells that ex-

pressed human hair follicle ORS keratins from the basal (K5, K14,and K19) and suprabasal (K16 andK17) layers (Fig. 1C andD andSI Appendix, Fig. S1). K16, which is normally expressed in hairfollicle suprabasal terminally differentiated cells, showed re-stricted expression within the tumor cell mass (Fig. 1D and SIAppendix, Fig. S1 C and E), suggesting that BCCs undergo inwarddifferentiation. In normal hair follicles, SHH signaling from bulbmatrix cells induced K17 expression in the ORS (29) (Fig. 1D). Bycontrast, K17 was ubiquitously expressed throughout BCC tumors(Fig. 1D and SI Appendix, Fig. S1 D and E), consistent with on-cogenic SHH signaling in BCC (30). Whereas K16 and K17 ex-pression is mutually exclusive in the hair follicle, coincidentexpression was observed within BCC (SI Appendix, Fig. S1E).HumanBCC samples also expressed hair follicle keratins typical

of the companion layer, inner root sheath (IRS), cuticle, andmedulla. The companion layer keratin K75 was expressed by six of20 BCCs studied (Fig. 1 C and D and SI Appendix, Fig. S1D).Three of 20 BCCs also expressed IRS keratins, but only in a lim-ited number of cells (Fig. 1 C and D). Hair shaft keratins were notobserved in BCC. In summary, by conventional H&E staining,BCCs consisted of monomorphic cells disguising a complex pat-tern of hair follicle-specific keratin expression (SI Appendix, Fig.S1F) that implies hierarchical growth with differentiation andmultiple tumor cell subpopulations.

BCC Cells Express Human Hair Follicle Bulge Stem Cell Marker CD200.BCC cells express a diverse pattern of hair follicle differentiationdemonstrated by intracellular expression of hair follicle-specifickeratins. One approach to enrich TIC subpopulations in humanBCCs could involve characterization of human BCC heteroge-neity by using cell surface differentiation markers that identifykeratinocyte stem cells (KSCs). It has been assumed that BCCarises from KSCs because these cells are sufficiently long-lived tosustain the necessary mutations and they already possess thecapacity for self-renewal. The hair follicle bulge region and otherkeratinocyte subpopulations have been defined by clusters ofdifferentiation antigens: CD24, CD71, CD146, and CD200 (31).BCC tumor cell inward differentiation, as manifested by differ-

ential hair follicle keratin expression, was also demonstrable byusing cell surface markers of epidermal and hair follicle differen-tiation. CD24 localized to cells of the hair follicle IRS and also theinterfollicular stratified epidermis, and was similarly restricted tothe BCC inner tumor cell mass (SI Appendix, Fig. S2A). Thetransferrin receptor CD71 identified basal cells below the level ofthe bulge in hair follicles and was predominantly expressed in theoutermost cell layers of BCC tumor nodules (SI Appendix, Fig.S2A). CD146 localized to the lower portions of the hair folliclebasal layer and surrounding endothelial cells, whereas, in BCCsamples, CD146 expression was limited to blood vessels and wasabsent from tumor cells (SI Appendix, Fig. S2A). Human hairfollicle bulge KSCs reside within the ORS between the origin ofthe sebaceous gland and arrector pili muscle insertion, and expressthe cell surface protein CD200 (Fig. 2A). Small clusters of BCCtumor cells located in the basal and immediately adjacent supra-basal layers of occasional sections of tumor nodules also expressedCD200 (Fig. 2B and SI Appendix, Fig. S2A). Consistent with theinward pattern of differentiation, proliferation assessed by Ki67labeling occurred in the outer cell layer of BCC tumor nodules(SI Appendix, Fig. S2B) and in cells that also expressed the anti-apoptotic protein BCL2 (SI Appendix, Fig. S2C). We concludedthat, if present, TICs might also be located in the outer cell layerof BCC.BCC samples were efficiently dissociated into single cell sus-

pensions with as many as 88% of cells viable (SI Appendix, Fig.S3A) similar to that observed after dissociation of normal skinand SCC (32). We found that not all BCC tumors expressedEpCAM as assessed by immunohistochemistry. When using BCCin which the majority of tumors cells expressed EpCAM, we de-termined that dissociated BCC tumor samples contained highnumbers of EpCAM-positive tumor cells (49–62% of all cellsisolated, n = 6; SI Appendix, Fig. S3B), confirming adequate dis-sociation and survival of BCC tumor cells, including a subpopu-lation of EpCAM+CD200+ cells (SI Appendix, Fig. S3C). All BCCsamples contained a small CD200+ tumor cell population (1.63 ±1.11%; range, 3.96–0.05%; n = 21; Fig. 2C), irrespective of thehistological type. BCC also contained CD45+ tumor-associatedleukocytes that accounted for 13.81 ± 10.84% (n = 21) of all cellsand included a subpopulation of CD200+ CD45+ cells (0.66 ±0.7%; Fig. 2C). Thus, CD200+ BCC tumor cells could be distin-guished by flow cytometry with the pan-leukocyte marker CD45 toexclude tumor infiltrating leukocytes.BCCCD200+CD45− and CD200−CD45− subpopulations were

isolated by flow cytometry with greater than 86% and 98% purity,respectively (SI Appendix, Fig. S3D). To assess SHH signaling,flow-sorted BCC tumor cell cDNA was compared with cDNAfrom intact BCC tumor tissue and the GLI1-overexpressing sar-coma cell line SJSA-1. Sustained SHH signaling leads to expressionof hedgehog-regulated genes, including the transcription factorGLI1 that augments the pathway (33). Both CD200+ CD45− andCD200− CD45− tumor cell populations expressed the humanhedgehog-regulated genes K17, PDGFRα, and GLI1 as expected(Fig. 2D andE and SI Appendix, Fig. S3E). Loss of GLI2 expressionwas apparent in the CD200+CD45− subpopulation. In contrast, theCD200− CD45− population maintained GLI2 expression similar tothat observed in SJSA-1 cells and hair follicles, highlighting a po-tential functional difference between these two populations. TheCD200+ CD45− subpopulation also exhibited almost twofold

A

B

C

D

Ha

ir F

ollic

le

BC

C

K14 & K16 K14 & K17 K14 & K75 K14 & K28

Fig. 1. Human BCC expresses hair follicle differentiation-specific keratins. (A)Human BCC. (B) Histology section showing interconnected islands of relativelymonomorphic darkly stained BCC tumor cells. (C) RT-PCRwith equal amounts ofcDNA from hair follicle-rich scalp tissue and two different BCC samples (BCC1and BCC2) showing expression of GAPDH and hair-specific keratins representingdistinct layers of hair follicle differentiation, including ORS, companion layer,IRS, cuticle, and matrix. (D) Double-label immunofluorescence characterizeskeratinocyte populations in hair follicle (Upper) and BCC (Lower), using K14labeling to contrast expression of suprabasal ORS (K16 and K7), companion(K75), and IRS (K28) layer keratins. (Scale bars: 100 μm.)

2 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1211655110 Colmont et al.

more proliferating cells than the CD200− CD45− cells, 7.26% vs.4.60%, respectively (SI Appendix, Fig. S3F). In summary, theCD200+ CD45− and CD200− CD45− BCC tumor cell populationsdemonstrated activated hedgehog signaling consistent with thegenetic basis for BCC.The development of an in vitro colony forming efficiency assay, as

was used to identify CD133+ primary human SCC TICs (32), couldtestBCCsubpopulationsbefore in vivoassessment.BCCcells formedcellular aggregates atop irradiated 3T3 feeder layers in tissue culture,similar to what was observed with primary human SCC and remi-niscent of transformed cells, but BCC cells did not exhibit anchorage-

independent growth (Fig. 3A). BCC spheroidal colonies could bequantified and were proportional to the number of cells plated, al-though the absolute number of colonies varied among the nine tumorsamples and experiments (SI Appendix, Fig. S4A). In vitro, BCCcolonies maintained active hedgehog signaling (Fig. 3B). Coloniesfrom unsorted BCC cells could be passaged in vitro and when im-planted into nudemice gave rise to tumors (Fig. 3C).When 105 flow-sorted cells were plated from five different BCC samples, CD200+CD45− sorted cells gave rise to threefold more colonies than theCD200− CD45− subpopulation (P < 0.005), which also gave rise tofewer colonies than unsorted cells (P < 0.01; Fig. 3D). CD200+CD45− sorted cells also gave rise to larger colonies in vitro (SI Ap-pendix, Fig. S4B) that resulted in tumor growth in vivo, in contrast tothe smaller colonies from theCD200−CD45− cells that did not formtumors in vivo. Thus, human BCC contained a relatively smallCD200+ CD45− tumor cell subpopulation that demonstratedincreased colony forming efficiency relative to unfractionatedtumor cells.

Human BCC Growth in Vivo Is Dependent on a “Humanized” Stromaand Etoposide Pretreatment. BCC growth in vivo has been difficultto achieve, even when tumor fragments containing stromal com-ponents were grafted into a variety of immunodeficient mice (SIAppendix, Table S1). We grafted 17 different human BCC tumor

C

CD200 & K17A B Dapi K14

CD200 Merge

H & E

GLI1 GLI2 K17

GLI1/2 negative control K17 negative control D

E

Fig. 2. The hair follicle KSC marker CD200 identifies a subpopulation of BCCtumor cells. (A) Double-label immunofluorescence of CD200 (red) togetherwith suprabasal ORS K17 (green) expression. (Lower: Higher-magnificationview.) Nonspecific CD200 labeling is seen within the nonviable inner aspect ofthe hair follicle adjacent to the hair shaft. (B) CD200 (green) expression bya small subset of basal and immediately adjacent suprabasal BCC tumor cellswithin K14-labeled (red) tumor nodules. The panel for each fluorescence labelis shown, along with a merged image (Lower). (C) Representative flow cyto-metric analysis of human BCC sample cell suspensions labeled with isotypecontrols (Left) and with CD45 and CD200 (Right) showing the CD200+ CD45−

subpopulation (3.94%) of interest as well as CD200+ CD45+ (0.19%) cells. (D)Human BCC tumor sections labeled by immunohistochemistry demonstrateexpression of GLI1, GLI2, and K17. (E) Three different BCC samples flow-sortedfor CD200+ CD45− and CD200− CD45− tumor subpopulations were comparedwith the GLI1-overexpressing cell line SJSA-1 (ATCC), three different BCC tissuesamples (BCC1, BCC2, BCC3) and hair follicle-rich scalp tissue (HF). Expression ofdownstream hedgehog signaling targets GLI1, PDGFRα, and K17 was assessedby RT-PCR. GAPDH was used as internal cDNA control. (Scale bars: 100 μm.)

A

C H&E

Gli1

Gli2

K17

B

D

Fig. 3. CD200+ BCC cells demonstrated increased colony-forming efficiency.(A) Unsorted BCC cells formed large and small tightly packed adherentspheroidal colonies when plated onto irradiated NIH 3T3 murine embryonicfibroblast feeder layers. (B) RT-PCR of SJSA-1, fresh human BCC tumor tissue(BCCt), cultured BCC cells (BCCc), and NIH 3T3 fibroblast cDNA demonstrat-ing expression of hedgehog-regulated genes using human specific primers.Human- and murine-specific GAPDH primers were used to determine therelative contributions of human vs. murine cell cDNA in the BCC cell sample.(C) Tissue sections of xenografted BCC colonies reveal tumor nodules withH&E staining and immunohistochemistry reveal tumor nodules after 12 wkin vivo. (D) Colony forming efficiency was used to estimate the relative TICfrequency within 105 cells from CD200+ CD45− vs. CD200− CD45− vs. unsortedpopulations from five different BCC tumor samples. (Scale bars: 100 μm.)

Colmont et al. PNAS Early Edition | 3 of 6

MED

ICALSC

IENCE

S

samples as 0.5-cm3 fragments into dorsal s.c. spaces of athymicnude (n = 17), SCID-beige (n = 5), and nonobese diabetic//SCID(n = 5) mice. After 12 wk, histological analyses of graft sites failedto demonstrate BCC in any recipient mice (SI Appendix, Fig. S5A).In vivo propagation of primary human SCC required the genera-tion of a stromal bed before tumor grafting, achieved by implantinga glass disk or Gelfoam dressing into the s.c. space 2 wk beforeimplantation of tumor tissue (34). This approach failed to allowpropagation of BCC in vivo in athymic nude mice or SCID-beigemice (n = 4 each; SI Appendix, Fig. S5A). We hypothesized thatresidual inflammatory cells present in athymic nude mice and, toa lesser extent, SCID-beige mice might hinder tumor growth afterinitial creation of the stromal bed. Similar to the preparation ofmurine mammary fat pads with etoposide, which induced myelo-suppression before human breast cancer engraftment, we createdstromal beds and administered i.p. etoposide 1 d before graftingBCC tissue. By using this approach, we achieved xenograft tumorgrowth after 12 wk from six of seven different primary human BCCsamples (SI Appendix, Fig. S5A). Consistently, BCC growth oc-curred in only athymic nude but not SCID-beige mice, suggestingthat a residual inflammatory milieu is essential for this in vivo BCCmodel. Tumor xenografts measured 3 to 8 mm in diameter, con-sistent with the slow growth of BCC in humans.Histology confirmedthat the xenografts recreated the original BCC tumor architectureandmaintained active SHH signaling (SI Appendix, Fig. S5B). Thus,BCC tumor growth in athymic nude mice was dependent on thecreation of a stromal bed and etoposide pretreatment.To begin to test the cancer stem cell hypothesis, it was necessary

to successfully graft fractionated cell suspensions from primary hu-man BCC. Analogous to our findings with grafting of primary hu-man SCC cell suspensions (34), it was necessary to “humanize”xenograft stromal beds. One million (106) normal primary humanfibroblasts were first suspended inMatrigel and implanted with glassdiscs or Gelfoam dressings. After 13 d, mice were treated with i.p.etoposide, and, on day 14, BCC xenograft cell suspensions werecoinjected with an additional 106 primary human normal fibroblastssuspended in Matrigel into the prepared graft sites (Fig. 4 A and B).This approach yielded successful xenograft tumor growth of 12 of 13xenografts from 10 different primary human BCC when 3 million ormore unsorted BCC cells were implanted (SI Appendix, Table S2).Tumor growth was not reproducible when 1 million unsorted pri-mary human BCC cells or fewer were implanted, irrespective of thehistological grade of the original tumor (Fig. 4C). The histologicalpatterns of xenograft tumors matched the original primary humanBCC histologies and tumors also maintained active SHH pathwaysignaling (Fig. 4D). The dose-dependence of engraftment supportsthe existence of a small number of TICs in human BCC. Based on

a limiting dilution analysis, we calculated the TIC frequency in hu-manBCC to be less than one per 1.5million (SI Appendix, Table S3).

CD200+ CD45− BCC Subpopulation Is Enriched for TICs. To determineif CD200+ CD45− primary human BCC cells were enriched forTICs, we grafted 52 athymic nude mice with varying numbers ofcells from 14 different BCC tumor samples (SI Appendix, Table S4)after isolation of CD200+ CD45− and CD200− CD45− subpopula-tions. After 12 wk, xenograft sites were harvested and analyzed byhistology. CD200− CD45− cells did not give rise to tumors in xe-nografts (0 of 14) involving eight different BCC samples, even when3× 106 tumor cells were implanted. In contrast, CD200+CD45− cellsreproducibly formed tumors, initiated with as few as 10,000 cells inour in vivo assay (Fig. 5A). CD200+CD45−humanBCCcells formedtumors resembling the original BCC and maintained active SHHsignaling and differentiation (Fig. 5B). Based on limiting dilutionanalysis, the TIC frequency in the CD200+ CD45− subpopulationapproximated one in 822 (SI Appendix, Table S5). Thus, the CD200+

CD45− subpopulation was enriched for TICs more than 1,500-fold.Of equal importance, we determined that CD200−CD45−BCCcellsdid not exhibit TIC activity. Because BCC xenografts were small andgrew slowly, serial in vivo transplantation of the CD200+ CD45−

population was not attempted. Taken together, these findings sup-port the existence of CD200+ TICs in human BCC.The expression of CD200 on BCC TICs and human hair follicle

bulge stem cells raised the possibility that BCCTICs arose from hairfollicle bulge KSCs. Analogous to human hair follicle bulge stemcells, CD200+ CD45− BCC cells expressed K15 (SI Appendix, Fig.S6) (35). The ability of adult tissue stem cells and TICs to self-renewled us to study the expression of transcription factors involved inembryonic stem cell maintenance and self-renewal. Kruppel-likefactor 4 (KLF4) has activator and repressor transcriptional activitiesand is a key regulator during embryogenesis, in which it preventsdifferentiation by regulating NANOG expression. However, inmature skin, KLF4 is normally expressed in the differentiated celllayers (36). Consistent with their differentiated state in BCC, KLF4expression was restricted to the CD200− CD45− subpopulation (SIAppendix, Fig. S6). The proto-oncogene C-MYC is associated withstem and transient amplifying cell proliferation, but continued ex-pression leads to epidermal stem cell depletion and terminal dif-ferentiation (37). In BCC, C-MYC was expressed by CD200+

CD45− and CD200− CD45− subpopulations (SI Appendix, Fig. S6).In contrast, the POUdomain transcription factorOct3/4, homeoboxtranscription factorNanog, and the telomerase reverse transcriptaseTert were expressed by tumor tissue but not the sorted BCC sortedpopulations (SI Appendix, Fig. S6). Hence, CD200+ BCC cellsclustered at the tumor periphery collectively do not express the

A

B

D

C

10mm

H&E Gli1

Gli2 K17

Fig. 4. Development of an in vivo BCC growth assay. (A) Schematic of the in vivo TIC assay, including the creation of a humanized stromal bed and etoposidepretreatment. When ≥3 × 106 unsorted BCC cells were implanted, reproducible in vivo BCC growth was achieved (B). In vivo BCC growth was dependent on thedose of unsorted BCC cells implanted (C). Active in situ hedgehog signaling was confirmed by histology and immunohistochemistry (D). (Scale bars: 100 μm.)

4 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1211655110 Colmont et al.

regulator of keratinocyte differentiation KLF4 and are exclusivelyenriched with cells that can initiate tumor growth.

DiscussionBCC arise from keratinocytes with mutations leading to constitu-tively active SHH growth factor signaling. Unlike the multiple ge-netic lesions required during stepwise carcinogenesis inmany othercancers, fewer “hits” are required for the development of BCC,perhaps explaining the absence of precursor lesions and why BCCis the most common malignancy in subjects of white race. SHH-expressing keratinocytes demonstrate continued proliferationand are resistant to p21CIP1/WAF1-induced replicative senescence(38). As BCCs grow, they continue to exhibit hair follicle differ-entiation with inward expression of differentiation-associated hair-specific keratins and the differentiation-associated protein CD24in central regions, whereas cell proliferation mostly occurs at thetumor periphery. A small number ofBCC cells identified by the cellsurface marker CD200 reside as clusters at the tumor periphery,and are not transcriptionally programed toward terminal differ-entiation, as they do not express the transcription factor KLF4.These CD200+ BCC cells are also unique in that they lack ex-pression of GLI2 in response to SHH signaling and instead rely onGLI1, contrary to the currently held view of SHHsignaling inBCC.Collectively, these findings suggest that BCC cells are not uniformand undergo hierarchical differentiation as proposed by the cancerstem cell model, with TICs residing in a clustered and relativelyundifferentiated CD200+ BCC cell precursor population.CD200 is a highly conserved type-1 membrane glycoprotein that

is expressed primarily by normal myeloid cells. However, CD200expression is also observed in a number of malignancies, includingrenal carcinoma, ovarian carcinoma, colon carcinoma, melanoma,acute myeloid leukemia, multiple myeloma, and chronic lympho-cytic leukemia (39, 40). Expression of the cognate receptorCD200R is restricted to myeloid cells and T lymphocytes (41).Ligand receptor interaction confers an immunosuppressive signalto immune cells. T lymphocytes down-regulate Th1 cytokines andinstead express IL-10 and exhibit regulatory T-cell activity (42).CD200 KO mice exhibit expansion and activation of tissue spe-cific macrophages, with rapid onset of experimental autoimmunediseases (41). The immune modulator protein CD200 is also ex-pressed by human hair follicle bulge KSCs, presumably to protectthese cells from immunological attack (31). Intriguingly, inter-spersed interfollicular keratinocytes that also express CD200 donot exhibit stem cell activity (43). As CD200+ CD45− BCC cells

express K15, these cells may arise from mutated CD200+ humanhair follicle bulge KSCs that also express the hair-specific keratinK15. Although expression of CD200 and K15 is not regulated bySHH, this does not exclude the possibility that BCC arises fromtransformed interfollicular or hair follicle differentiated kerati-nocytes. Putative TICs in multiple cancer cell lines have also beenfound to express CD200 (44). In human acute myeloid leukemiaand multiple myelomas, CD200 expression is associated with poorprognosis (45–47). In summary, CD200 is expressed by BCC TICsand hair follicle bulge KSCs from which they may be derived, andmay help protect both cell populations from immunological attack.To confirm the presence of BCC TICs, we developed a unique

in vivo assay. Similar to many other cancers, BCC growth is de-pendent on the presence of stromal cells. We implanted glass discsor Gelfoam dressings together with 1 million primary humanfibroblasts, a strategy we developed to propagate primary humanSCC xenografts (33), to create a receptive stromal bed. i.p. ad-ministration of etoposide before tumor implantation was also re-quired, in analogy to the method described for primary humanbreast cancer xenografts (48). With this approach, tumor growthwas successful in athymic nude mice, which lack T lymphocytes,but not SCID-beige mice, which lack both T and B lymphocytesand have reduced natural killer cell numbers. We hypothesize thatsome inflammatory cells are important during the initial phase ofstromal bed formation but then hinder tumor engraftment, asetoposide-induced myelosuppression was also found to be neces-sary. BCC grafts in this model grew slowly, consistent with the rateof growth of BCC observed in humans and mouse models. Thus,despite its complexity, the model we describe faithfully recreatedhuman BCC growth from dissociated tumor cells and allowedcharacterization of TICs in BCC.Recently, drugs that simultaneously inhibitmultiple growth factor

pathways (e.g., tyrosine kinase receptor inhibitors), single pathways(VEGF receptor, TGF-β receptor, EGF receptor, and SMOantagonists), mutated targets (B-Raf inhibitors), and downstreamsignaling targets (MEK inhibitors) have been developed. Althoughmalignancies in patients often show initial responses to these drugs,cancer recurrence is frequently observed. This study demonstratesthe existence of TICs that may drive BCC growth in patients as wellas in mice, and these cells may be resistant to killing by SMOantagonists (SI Appendix, Fig. S7). Although not tested, our datawould also suggest that currently available anti-CD200 neutralizingantibody alone or in combination with SMO antagonists might bebeneficial in the treatment of inoperable and metastatic BCC.

Materials and MethodsImmunofluorescence and Immunohistochemistry. Immunofluorescence andimmunohistochemistry were performed by using standard techniques, as pre-viously described (25, 27),with the followingprimary antibodies: pancytokeratin(clone AE1/3; Dako), CD200 (clone MRC OX104; Serotec), GLI1 (GTX27523;GeneTex), GLI2 (GTX27195; GeneTex), BCL2 (clone 124; Dako), CD24 (cloneML5;BD Pharmingen), CD71 (clone M-A712; BD Pharmingen), CD146 (clone P1H12;BD Pharmingen), CD200 (clone MRC OX104; Serotec), Ki67 (clone Mib1; Dako),and human cytokeratins K14 (binding site, clone PH503), K16 (gift from RebeccaPorter, Department of Dermatology and Wound Healing, School of Medicine,Cardiff University, Cardiff, UK), K17 (Thermo Scientific Pierce, cloneE3), K28 (giftfrom Rebecca Porter), and K75 (gift from Rebecca Porter).

For cytospin analysis, cells were washed and suspended in PBS solution,100-μL aliquots of cells were added to each slide in a Cytofunnel (Shandon;Thermo Scientific) and spun at 100 × g for 5 min in a cytocentrifuge(Shandon Cytospin 2 cytocentrifuge; Thermo Scientific), fixed in acetone,and labeled with a pancytokeratin and anti-GLI1 antibody.

Flow Cytometry. Tumor samples were subjected to mechanical and enzymaticdissociation as previously described (27). Samples were analyzed and flow-sorted by using FACSCalibur and a FACSAria flow cytometer (BD Biosciences)with mouse fluorochrome-conjugated IgG subtype isotype controls (BD Phar-mingen). Live cell gates were created by using 7-amino-actinomycin D (BDPharmingen) to label dead cells. Cells were labeled according tomanufacturer’sinstructions with fluorochrome-conjugated antibodies: CD200-AF647 (Serotec),EpCAM-APC (BD Pharmingen), and CD45-FITC (BD Pharmingen). Cell cycleanalysis was performed using propidium iodide/RNase staining buffer (BDPharmingen), and data were analyzed by using FlowJo software (Tree Star).

A B H&E Gli1

Gli2 K17

K19 K16

Fig. 5. The CD200+ CD45− BCC cell subpopulation contains TICs. (A) Flow-sorted CD200+ CD45− and CD200− CD45− subpopulations from 14 differentfresh BCC samples were grafted in varying numbers. The CD200+ CD45−

subpopulation grafts (n = 38) gave rise to reproducible BCC growth with asfew as 104 cells implanted, whereas no growth was observed in grafts fromthe CD200− CD45− subpopulation (n = 14) when as many as 3 × 106 cells wereimplanted. (B) In all cases, BCC growth was confirmed by histology andverified by immunohistochemistry to demonstrate active hedgehog signal-ing and differentiation. (Scale bars: 100 μm.)

Colmont et al. PNAS Early Edition | 5 of 6

MED

ICALSC

IENCE

S

In Vitro Assay and Tissue Culture. Unsorted or sorted cells were plated onto 50Gy irradiated 3T3 murine fibroblast feeder layers in 10-cm tissue culture Petridishes in keratinocyte serum-free media (Keratinocyte-SFM; Gibco) supple-mented with 20 ng/mL EGF, 10 ng/mL FGF-2 and 0.15 ng/mL bovine pituitaryextract, 25 U/mL penicillin, 25 μg/mL streptomycin, and 10 μg/mL amphotericin.Media were changed every 3 d, and the number of spheroidal BCC colonieswas counted on day 14 by using an inverted light microscope.

Transplantation of Dissociated Human BCC Cells. Athymic nude homozygousfoxn1nu (Jackson Labs), SCID-beige, and nonobese diabetic/SCID (Taconic) micewere housed and used under conditions approved by the animal care and usecommittee at the National Cancer Institute or carried out under the terms ofa UKGovernment HomeOffice project license (ethics approval is included in theUK Home Office project license). Mice were anesthetized, and Gelfoam dress-ings (Johnson and Johnson) or sterilized glass discs were implanted into thedorsal s.c. space, togetherwith 106 primary human fibroblasts suspended in 100μL of Matrigel (BD Biosciences), for single cell suspension experiments, andwoundswere closedwith surgical staples (Mikron). After 13d, etoposide (30mg/kg) was administered (diluted in serum-free HBSS for a final injection volume of200 μL). On day 14, mice were anesthetized, glass discs were removed (as ap-propriate), and BCC cells together with 106 primary human fibroblasts that hadbeen suspended in 100 μL of Matrigel were injected into s.c. spaces or, alter-natively, into residual Gelfoamdressings. After 12wk,micewere euthanized viaCO2 inhalation, and tumors were removed for analysis.

RT-PCR. Tissue specimens were microdissected to remove the overlying epi-dermis. Tissue or cultured cells were homogenized in TRIzol (Invitrogen)followed by RNA isolationwith an RNeasy kit (Qiagen) per themanufacturer’sinstructions. Superscript III (Invitrogen) or iScript cDNA synthesis (Bio-Rad)were used to create cDNA. All PCR reactions were carried out using PlatinumTaq (Invitrogen) and specific primers (SI Appendix, Tables S6 and S7). Totaland human- and mouse-specific GAPDH were used as the housekeepinggenes for the amplifications.

Statistical Analysis. Paired t tests were used to compare the colony formingefficiencies of unsorted vs. CD200+ CD45− vs. CD200− CD45− subpopulations.For in vivo limiting dilution assays, the frequencies of cancer-initiating cellswere calculated by using L-Calc software (Stem Cell Technologies), with χ2

analysis to determine internal consistency.

ACKNOWLEDGMENTS. We thank Ms. Rachelle Graham, a summer student,for characterization of BCC by immunohistochemistry; and Drs. AndrewMontemarano (Rockledge Skin Cancer Clinic), Kurt Maggio (Walter ReedArmy Medical Center Dermatology Service), Martin Braun (Braun and BraunMDs), and Andrew Morris and Richard Motley (Welsh Institute of Derma-tology) for providing tumor tissue samples. This research was supportedby the Intramural Research Program of the National Institutes of Health,National Cancer Institute, Center for Cancer Research.

1. American Cancer Society (2012) Cancer Facts and Figures 2012. Available at: www.cancer.org/Research/CancerFactsFigures/CancerFactsFigures/cancer-facts-figures-2012.Accessed March 3, 2012.

2. Johnson RL, et al. (1996) Human homolog of patched, a candidate gene for the basalcell nevus syndrome. Science 272(5268):1668–1671.

3. Grachtchouk M, et al. (2000) Basal cell carcinomas in mice overexpressing Gli2 in skin.Nat Genet 24(3):216–217.

4. Hahn H, et al. (1996) Mutations of the human homolog of Drosophila patched in thenevoid basal cell carcinoma syndrome. Cell 85(6):841–851.

5. Oro AE, et al. (1997) Basal cell carcinomas in mice overexpressing sonic hedgehog.Science 276(5313):817–821.

6. Xie J, et al. (1998) Activating Smoothened mutations in sporadic basal-cell carcinoma.Nature 391(6662):90–92.

7. Dahmane N, Lee J, Robins P, Heller P, Ruiz i Altaba A (1997) Activation of the tran-scription factor Gli1 and the Sonic hedgehog signalling pathway in skin tumours.Nature 389(6653):876–881.

8. Fan H, Oro AE, Scott MP, Khavari PA (1997) Induction of basal cell carcinoma featuresin transgenic human skin expressing Sonic Hedgehog. Nat Med 3(7):788–792.

9. Epstein EH (2008) Basal cell carcinomas: Attack of the hedgehog. Nat Rev Cancer8(10):743–754.

10. McMahon AP, Ingham PW, Tabin CJ (2003) Developmental roles and clinical signifi-cance of hedgehog signaling. Curr Top Dev Biol 53:1–114.

11. Bhardwaj G, et al. (2001) Sonic hedgehog induces the proliferation of primitive hu-man hematopoietic cells via BMP regulation. Nat Immunol 2(2):172–180.

12. Scales SJ, de Sauvage FJ (2009) Mechanisms of Hedgehog pathway activation incancer and implications for therapy. Trends Pharmacol Sci 30(6):303–312.

13. Jiang J, Hui CC (2008) Hedgehog signaling in development and cancer. Dev Cell 15(6):801–812.

14. Tremblay MR, et al. (2009) Discovery of a potent and orally active hedgehog pathwayantagonist (IPI-926). J Med Chem 52(14):4400–4418.

15. De Smaele E, Ferretti E, Gulino A (2010) Vismodegib, a small-molecule inhibitor of thehedgehog pathway for the treatment of advanced cancers. Curr Opin Investig Drugs11(6):707–718.

16. Tremblay MR, McGovern K, Read MA, Castro AC (2010) New developments in thediscovery of small molecule Hedgehog pathway antagonists. Curr Opin Chem Biol 14(3):428–435.

17. Von Hoff DD, et al. (2009) Inhibition of the hedgehog pathway in advanced basal-cellcarcinoma. N Engl J Med 361(12):1164–1172.

18. Metcalfe C, de Sauvage FJ (2011) Hedgehog fights back: Mechanisms of acquiredresistance against Smoothened antagonists. Cancer Res 71(15):5057–5061.

19. Skvara H, et al. (2011) Topical treatment of Basal cell carcinomas in nevoid Basal cellcarcinoma syndrome with a smoothened inhibitor. J Invest Dermatol 131(8):1735–1744.

20. Sekulic A, et al. (2012) Efficacy and safety of vismodegib in advanced basal-cell car-cinoma. N Engl J Med 366(23):2171–2179.

21. Tang JY, et al. (2012) Inhibiting the hedgehog pathway in patients with the basal-cellnevus syndrome. N Engl J Med 366(23):2180–2188.

22. Grachtchouk V, et al. (2003) The magnitude of hedgehog signaling activity definesskin tumor phenotype. EMBO J 22(11):2741–2751.

23. Hutchin ME, et al. (2005) Sustained Hedgehog signaling is required for basal cellcarcinoma proliferation and survival: Conditional skin tumorigenesis recapitulates thehair growth cycle. Genes Dev 19(2):214–223.

24. Yang SH, et al. (2008) Pathological responses to oncogenic Hedgehog signaling in skinare dependent on canonical Wnt/beta3-catenin signaling. Nat Genet 40(9):1130–1135.

25. Grachtchouk M, et al. (2011) Basal cell carcinomas in mice arise from hair follicle stemcells and multiple epithelial progenitor populations. J Clin Invest 121(5):1768–1781.

26. Wang GY, Wang J, Mancianti ML, Epstein EH, Jr. (2011) Basal cell carcinomas arisefrom hair follicle stem cells in Ptch1(+/-) mice. Cancer Cell 19(1):114–124.

27. Youssef KK, et al. (2010) Identification of the cell lineage at the origin of basal cellcarcinoma. Nat Cell Biol 12(3):299–305.

28. Schweizer J, Langbein L, Rogers MA, Winter H (2007) Hair follicle-specific keratins andtheir diseases. Exp Cell Res 313(10):2010–2020.

29. Bianchi N, Depianto D, McGowan K, Gu C, Coulombe PA (2005) Exploiting the keratin17 gene promoter to visualize live cells in epithelial appendages of mice.Mol Cell Biol25(16):7249–7259.

30. Depianto D, Kerns ML, Dlugosz AA, Coulombe PA (2010) Keratin 17 promotes epi-thelial proliferation and tumor growth by polarizing the immune response in skin.Nat Genet 42(10):910–914.

31. Ohyama M, et al. (2006) Characterization and isolation of stem cell-enriched humanhair follicle bulge cells. J Clin Invest 116(1):249–260.

32. Patel GK, et al. (2012) Identification and characterization of tumor-initiating cells inhuman primary cutaneous squamous cell carcinoma. J Invest Dermatol 132(2):401–409.

33. Regl G, et al. (2002) Human GLI2 and GLI1 are part of a positive feedback mechanismin basal cell carcinoma. Oncogene 21(36):5529–5539.

34. Patel GK, Yee CL, Yuspa SH, Vogel JC (2012) A humanized stromal bed is required forengraftment of isolated human primary squamous cell carcinoma cells in immuno-compromised mice. J Invest Dermatol 132(2):284–290.

35. Lyle S, et al. (1998) The C8/144B monoclonal antibody recognizes cytokeratin 15 anddefines the location of human hair follicle stem cells. J Cell Sci 111(Pt 21):3179–3188.

36. Segre JA, Bauer C, Fuchs E (1999) Klf4 is a transcription factor required for estab-lishing the barrier function of the skin. Nat Genet 22(4):356–360.

37. Gandarillas A, Watt FM (1997) c-Myc promotes differentiation of human epidermalstem cells. Genes Dev 11(21):2869–2882.

38. Fan H, Khavari PA (1999) Sonic hedgehog opposes epithelial cell cycle arrest. J CellBiol 147(1):71–76.

39. Minas K, Liversidge J (2006) Is the CD200/CD200 receptor interaction more than justa myeloid cell inhibitory signal? Crit Rev Immunol 26(3):213–230.

40. Moreaux J, Veyrune JL, Reme T, De Vos J, Klein B (2008) CD200: A putative thera-peutic target in cancer. Biochem Biophys Res Commun 366(1):117–122.

41. Wright GJ, et al. (2003) Characterization of the CD200 receptor family in mice andhumans and their interactions with CD200. J Immunol 171(6):3034–3046.

42. Hoek RM, et al. (2000) Down-regulation of the macrophage lineage through in-teraction with OX2 (CD200). Science 290(5497):1768–1771.

43. Terunuma A, et al. (2008) Behavior of human foreskin keratinocytes expressing a hairfollicle stem cell marker CD200. J Invest Dermatol 128(5):1332–1334.

44. Kawasaki BT, Mistree T, Hurt EM, Kalathur M, Farrar WL (2007) Co-expression of thetoleragenic glycoprotein, CD200, with markers for cancer stem cells. Biochem BiophysRes Commun 364(4):778–782.

45. Kawasaki BT, Farrar WL (2008) Cancer stem cells, CD200 and immunoevasion. TrendsImmunol 29(10):464–468.

46. Moreaux J, et al. (2006) CD200 is a new prognostic factor in multiple myeloma. Blood108(13):4194–4197.

47. Tonks A, et al. (2007) CD200 as a prognostic factor in acute myeloid leukaemia.Leukemia 21(3):566–568.

48. Al-Hajj M, Wicha MS, Benito-Hernandez A, Morrison SJ, Clarke MF (2003) Pro-spective identification of tumorigenic breast cancer cells. Proc Natl Acad Sci USA100(7):3983–3988.

6 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1211655110 Colmont et al.