Rare steroid receptor-negative basal-like tumorigeniccells in luminal subtype human breastcancer xenograftsKathryn B. Horwitz*†, Wendy W. Dye*, Joshua Chuck Harrell*, Peter Kabos‡, and Carol A. Sartorius*§

*Divisions of Endocrinology and ‡Medical Oncology, Department of Medicine, and †Department of Pathology, University of Colorado, Anschutz MedicalCampus, 12801 East 17th Avenue, Aurora, CO 80045

Edited by Irving L. Weissman, Stanford University School of Medicine, Stanford, CA, and approved February 20, 2008 (received for review July 2, 2007)

There are two major subtypes of human breast cancers: theluminal, estrogen, and progesterone receptor-positive, cytoker-atin 18-positive (ER�PR�CK18�) subtype, and the basalER�PR�CK18�CK5� subtype. Tumor-initiating cells (CD44�) havebeen described for human breast cancers; whether these arecommon to the two subtypes is unknown. We have identified arare population of cells that are both CD44� and ER�PR�CK5� inluminal-like ER�PR� T47D human breast tumor xenografts. Thetumor-isolated CD44� cell fraction was highly enriched for clono-genic (in vitro culture) and tumorigenic (in vivo reimplantation)cells compared with the CD44� cell fraction. Rare ER�PR�CK5� cellswere present within CD44�-derived colonies. Tumor-isolated cellsplaced in minimal media also contained rare ER�PR�CK5� cells atearly time points (<10 cells); however, this population did notexpand with increasing colony size. The number of ER�PR�CK5�

cells, conversely, increased linearly with colony growth. Similary,tumors originating in vivo from CD44� cells contained a rare staticER�PR�CK5� population, an intermediate ER�PR�CK5� popula-tion, and an expanding ER�PR�CK5� population. PutativeER�PR�CK5� transitional cells could be seen only in colonies ortumors treated with a progestin. We propose that luminal ER�PR�

breast tumors contain a minor ER�PR�CK5� population that hasthe capacity to generate the majority of ER�PR�CK18�CK5� cells.Luminal breast cancers are treated with endocrine therapies thattarget ER. The rare ER�PR�CK5� progenitor cells would escape suchtreatments and survive to repopulate the tumor.

cancer stem cell � cytokeratin 5 � estrogen receptor �progesterone receptor

The cancer stem cell theory proposes that tumors originatefrom stem cells that are capable of self-renewal while giving

off rapidly replicating progeny that comprise the tumor bulk (1).A small tumor-initiating subpopulation of cells, indicative ofstem cells, has been identified for several human epithelialcancers (2–6). Cells with stem-like properties also have beenfound in human tumor xenograft models and in many humancancer cell lines (7, 8). With regard to the normal human breast,stem cells are proposed to be contained in a cell fraction thatexpresses epithelial-specific antigen (ESA)� and �6 integrin(CD49f)�. Additional putative stem cell markers includeMusashi, the Polycomb group repressor Bmi-1, cytokeratins 5/6(CK5/6), and label-retention and side-population cells (reviewedin ref. 9). Currently, tumorigenic cells from primary humanbreast cancers have been identified solely by CD44�CD24�/low

ESA� expression (2).Approximately 80% of breast cancers fall into luminal sub-

types, which are estrogen receptor (ER) and progesteronereceptor (PR)-positive and express cytokeratin 18 (CK18)(ER�PR�CK18�). These tumors have a better prognosis andmore varied treatment options, including endocrine therapies,than their basal ER�PR�CK5� counterparts. Nonetheless, hor-mone resistance and consequent tumor recurrence remain aproblem for ER�PR� disease. The origin of luminal

ER�PR�CK18� and basal ER�PR�CK5� breast cancers re-mains under debate. The cancer stem cell theory proposes thatER�PR� and ER�PR� breast cancers arise from normal breastepithelial stem cells, leading to debate about the steroid receptorcontent of these nontransformed stem cells. Studies based onnormal mouse mammary gland models generally describe nor-mal stem cells as being restricted to basal cell layers of the gland,which are ER�PR� (10–12). However, ER�PR� label-retainingcells have also been described in the luminal epithelium of themouse mammary gland (13, 14). In the human breast, label-retaining and side-population cells are reportedly enriched forER expression (15). There is some speculation that ER�PR�

basal epithelial stem cells spawn ER�PR� progenitor cells foundin the luminal layer, which might also become tumorigenic (16).Along these lines, it has been proposed that ER�PR� breastcancers originate from ER�PR� luminal progenitor cells,whereas ER�PR� cancers arise from ER�PR� basal stem cells(17). This hypothesis is supported by gene expression profiling ofbreast cancer subtypes, which shows that luminal tumors areER�PR�CK18�, whereas basal tumors are ER�PR�CK5� (18).

Using an ER�PR� T47D human breast tumor xenograftmodel, we have identified rare CK5� tumor cells that areER�PR� and express the surface antigen CD44, a widelyproposed marker of breast cancer stem cells (2, 3, 5, 6). Thetumor-isolated CD44� fraction was highly enriched for clono-genic (Matrigel, agarose) and tumorigenic (in vivo) cells com-pared with the CD44� fraction. Analysis of marker expressiondemonstrated that initial colonies contained ER�PR�CK5�

cells that do not expand with colony growth over time. However,the tumor bulk is composed of ER�PR�CK5� cells. A putativetransient ER�PR�CK5� subpopulation could be observed onlywith progestin treatment. We propose that ER�PR�CK5� lu-minal breast cancer subtypes contain a rare tumorigenic basal-like ER�PR�CK5�CD44� subpopulation. This is important,because hormonal therapies targeting luminal ER�PR� tumorswould be ineffective against these rare ER�PR�CK5� tumor-initiating cells.

ResultsCK5� Cells in Luminal-Like ER�PR� Tumors Are CD44� and ER�PR�.Solid tumors generated from estrogen-treated ER�PR� T47Dhuman breast cancer xenografts predominantly express theluminal markers CK8 and CK18. However, a small (1%) sub-population of cells within these tumors express CK5, a marker

Author contributions: C.A.S. designed research; W.W.D., J.C.H., P.K., and C.A.S. performedresearch; W.W.D., P.K., and C.A.S. analyzed data; and K.B.H. and C.A.S. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

§To whom correspondence should be addressed. E-mail: carol.sartorius@uchsc.edu.

This article contains supporting information online at www.pnas.org/cgi/content/full/0706216105/DCSupplemental.

© 2008 by The National Academy of Sciences of the USA

5774–5779 � PNAS � April 15, 2008 � vol. 105 � no. 15 www.pnas.org�cgi�doi�10.1073�pnas.0706216105

Dow

nloa

ded

by g

uest

on

Janu

ary

17, 2

021

of the basal receptor-negative subtype of breast cancers (19).These CK5� cells lack the basal epithelial marker CK14 and themyoepithelial marker smooth muscle actin and increase sever-alfold in number when progestins are given during tumor growth(19). To characterize these cells, CK5� and CK5� cells wereisolated from tumor xenografts grown in mice treated withestradiol alone or with estradiol plus the progestin medroxy-progesterone acetate (6�-methyl-17�-hydroxyprogesterone ac-etate, MPA), using laser-capture microdissection of frozentumor sections immunost ained for CK5 [supporting infor mation(SI) Text]. Their gene ex pression profiles were then c ompared. Thisgenerated a list of genes that were significantly changed in the CK5�

compared with the CK5� cells, from both estrogen and estrogen-plus progestin-treated tumors (Dat aset S1 and Dat aset S2).

A volcano plot generated in Partek identified genes that werecoexpressed in CK5� cells, by their significance (P value) andtheir CK5�/CK5� ratio (Fig. 1A). The putative cancer stem cellmarker CD44 was among several genes whose expression wasup-regulated in CK5� compared with CK5� cells regardless ofhormone treatments (Fig. 1B). Dual CK5/CD44 immunohisto-chemistry on tumor sections confirmed that CD44 (red) iscoexpressed with CK5� (green) in both estrogen and estrogen-plus progestin-treated tumors (Fig. 1C). Additionally, tumorstreated with progestins had occasional CK5� cells that wereCD44�/low (arrow). This suggests that progestins increase aCK5�CD44� subpopulation independent of the CK5�CD44�

population.Dual immunohistochemistry for CK5/PR was performed on

sections of estrogen and estrogen-plus progestin-treated tumors(Fig. 2). Rare CK5� cells (green) in estrogen-treated tumorswere typically (95%) PR�, whereas the surrounding tumor cells

were PR� (red nuclei). Although the vast majority of CK5� cellsin estrogen- plus progestin-treated tumors were also PR�, a rarePR�CK5� subpopulation was increased to �20% in progestin-treated tumors (Fig. 2, �). Similarly, CK5� cells were mainlyER�, and an ER�CK5� subpopulation was increased by estro-gen-plus progestin treatment (Fig. S1).

The CD44� Cell Fraction from PR� Tumors Produces HeterogeneousColonies Containing PR�CK5� Cells in 3D Culture. We used thesurface marker CD44 to isolate cells from solid T47D xenografttumors, reasoning that the rare CK5� population would becontained within the CD44� fraction. T47D-ZsGreen cells wereused for these experiments to help distinguish the human tumorcells from host mouse cells during FACS sorting (Fig. 3A).Single-cell suspensions of tumor cells were depleted of Lin� hostcells, labeled with an Alexa Fluor 647 anti-CD44 antibody andCD44� cells (upper box), and CD44� cells (lower box) werecollected separately. The fraction of CD44� cells was consis-tently �1% in both estrogen and estrogen-plus progestin-treatedsolid tumors. A portion of sorted CD44� and CD44� cells werespun onto slides, fixed, and stained for CD44 and CK5. Approx-imately 75% of the sorted CD44� cells and 2% of CD44� cellswere CK5� by immunostain (Fig. S2). Flow cytometry alsodetermined that �67% of unsorted CD44� cells were CK5�

(Fig. S3).Sorted CD44� and CD44� cells were plated into separate

Matrigel-coated eight-well chambers at dilutions between 2 and4 � 103 in MEM supplemented with 5% FCS. After �3 weeks,large colonies reaching 200–500 �m in diameter were observedin the CD44� population (Fig. 3B). Rare surviving cells in theCD44� population formed colonies �50 �m in diameter. Whenthe number of CD44� cells was increased to �104 cells, coloniesformed more often, presumably due to contaminating CD44�

cells. Purity of flow sorts (postsort analysis) was determined tobe �99%. Clonal analysis on soft agar determined that 1.7/103

CD44� and 8.3/103 CD44� cells formed colonies. There was nodifference in colony formation between CD44� cells isolatedfrom estrogen or estrogen-plus progestin-treated tumors.

Serial sections of paraffin-embedded CD44� colonies werestained by immunohistochemistry for CD44, CD24, CK5, andPR (Fig. 3C). Cells in the 3D colonies ubiquitously expressedCD44 and were mostly CD24�/low, consistent with previouslydescribed tumorigenic breast cancer cells (2). CD44� coloniescontained rare peripherally located CK5� cells. The majority ofcells in the colonies were PR�. However, distinct PR� areascould be found in the vicinity of the rare CK5� cells. Because theCD44� cells failed to grow significant colonies, we were unableto identify and stain them in paraffin sections.

Established CD44� 3D colonies were either untreated (con-trol) or treated 24 h with MPA (�P). Dual immunohistochem-

Fig. 1. Coexpression of CK5 and CD44 in human breast tumor xenografts.ER�PR� T47D human breast tumor xenografts were grown in ovariectomizednude mice treated with estradiol alone (E2) or with E2 plus the progestin MPA(P). CK5� and CK5� cells, from E2 and E2�P sets were isolated by laser-capturemicrodissection of frozen tumor sections immunostained for CK5. Cells weresubjected to gene expression profiling. (A) Volcano plot of genes positivelycorrelated with CK5 expression by significance (P value) vs. fold change(CK5�/CK5�) from E2�P-treated tumors. Genes with the highest positivecorrelation are shown in red. Probes for CK5 (KRT5) and CD44 are indicated.(B) Gene expression profile of CD44 (relative expression, SEM) in CK5� andCK5� cells isolated from E2- and E2�P-treated tumors. (C) Immunohistochem-istry for CD44 (red) and CK5 (green) of paraffin sections from T47D tumorsgrown with E2 or E2�P. Sections were counterstained for DAPI (blue). Thearrow points to a rare CK5�CD44�/low cell. (Scale bar, 10 �m.)

Fig. 2. CK5� cells in T47D tumors are mainly PR-negative. Sections of T47Dtumors treated with estrogen (E2) or estrogen plus progestin (E2�P) werestained for nuclear PR (red) and cytoplasmic CK5 (green). The percentage ofCK5� cells that are also PR� was 5.1% (E2) and 19.8% (E2�P). A rare CK5�PR�

cell in the E2�P set is marked with an asterisk. (Scale bar, 10 �m.)

Horwitz et al. PNAS � April 15, 2008 � vol. 105 � no. 15 � 5775

CELL

BIO

LOG

Y

Dow

nloa

ded

by g

uest

on

Janu

ary

17, 2

021

istry for CD44/PR and CK5/PR was performed on sections ofCD44� colonies (Fig. 3 D and E). Control colonies (Fig. 3D)were uniformly CD44� (green) but were heterogeneous for PRexpression (red), containing both PR�CD44� and PR�CD44�

cells. The peripherally located CK5� cells (green) were PR�

(red, lower center, arrows; and merged). Colonies were alsoheterogeneous for ER expression, and the peripheral CK5� cellswere ER� (Fig. S4).

Like untreated colonies, progestin-treated colonies (Fig. 3E)uniformly expressed CD44 (green) and were heterogeneous forPR (red). However, progestin-treated colonies contained anincreased number of CK5� cells (green) compared with un-treated colonies, with the additional appearance of rarePR�CK5� cells (Fig. 3E, arrow). Rare ER�CK5� cells couldalso be seen in the progestin-treated colonies (Fig. S4).

Prevalence of Nonexpanding PR�CK5� Cells in Colony-Forming Units.To determine the relationship between colony formation andPR�CK5� cells, we prepared single-cell suspensions fromER�PR� tumor xenografts and plated 104 cells directly ontosix-well dishes with inserted glass coverslips in minimal media.Coverslips were removed, fixed, and stained for CK5/PR on days3, 7, and 14. Select coverslips were treated with MPA for 24 hbefore fixation. Colony-forming units (three or more cells percolony at day 3, 20 or more at day 7, and 50 or more at day 14)were photographed, scored for CK5 and PR expression, andgraphed as a function of positive cell number for each marker vs.colony size (Fig. 4A). In untreated control colonies, 96.9%(32/33) of colonies, �25 cells contained CK5� cells. The number

of CK5� cells did not increase as colony size grew (linearregression, m � �0.03), however, so CK5� cells were rapidlydiluted, and only 50% (8/16) of colonies with �25 cells containedCK5� cells. The vast majority (96.7%) of CK5� cells in controlcolonies were PR�. By contrast, the number of PR�CK5� cellsin controls grew linearly with colony size (m � 1.01). Progestin-treated colonies all (100%, 45/45) contained CK5� cells, whichincreased in number with colony size (m � 0.37). Like untreatedcolonies, PR� cells in progestin-treated colonies increased lin-early with colony size (m � 1.05). Thus, among CK5� cells inprogestin-treated colonies, 81% of cells were PR�CK5�, and19% were PR�CK5�. This suggests that a fraction of the PR�

cells retain the capacity to reactivate CK5 in response to aprogestin. However, as the colony expands, an increasing subsetof PR� cells lose the ability to activate CK5. Representativecolonies stained for PR (red) and CK5 (green) and used forquantitation in Fig. 4A are shown in Fig. 4B at early (10–15 cells,Fig. 4B Left) and later time points (60–80 cells, Fig. 4B Right)with vehicle (control, Fig. 4B Upper) or progestin treatment (Fig.4B Lower). Select coverslips were stained for CD44 and colony-forming units were CD44� (data not shown).

CD44� Cells Generate Tumors in Vivo That Acquire a Luminal Pheno-type Over Time. Fluorescent, tumor-isolated, FACS-sortedCD44� and CD44� cells were assessed for their tumorigenicityin vivo. Estrogen-supplemented ovariectomized nude mice wereinoculated in opposing abdominal mammary glands with 104

ZsGreen-tagged CD44� or CD44� cells in Matrigel. Mice weremonitored for tumor formation by fluorescence. Fig. 5A shows

Fig. 3. Tumor-isolated CD44� cells form heterogeneous colonies in 3D culture that contain CK5�PR� cells. (A) Lin�ZsGreen�CD44� cells were isolated fromestrogen-treated ER�PR� T47D tumors by FACS. For the depicted sort, 1.33% of ZsGreen� tumor cells were CD44� (upper box). Lin�ZsGreen�CD44� cells wereisolated from the lower boxed area. (B) Cells (2 � 103 CD44� and CD44�) were plated into eight-well chambers on Matrigel. Colonies were photographed after3 weeks. (Scale bars, 100 �m.) (C) Sections containing CD44� colonies were stained with antibodies against CD44, CK5, PR, and CD24. (Scale bar, 10 �M.) (D andE) Established CD44� colonies were treated with vehicle (D) or 10 nM MPA (P) for 24 h (E) before fixation and paraffin embedding. Sections were stained by dualimmunohistochemistry for CD44 (green)/PR (red), or CK5 (green)/PR (red). (Scale bars, 10 �m.) Arrows indicate CK5�PR� cells (D Lower Center) and a rare CK5�PR�

cell (E Lower Right).

5776 � www.pnas.org�cgi�doi�10.1073�pnas.0706216105 Horwitz et al.

Dow

nloa

ded

by g

uest

on

Janu

ary

17, 2

021

examples of bilateral dissected tumors at day 30 and tumors inintact mice at days 28 and 45. Tumor colonies were clearly visiblein the CD44� set by 28 days after injection; they increased in sizeand coalesced with time. Much smaller CD44� colonies tendedto form late presumably due to contaminating CD44� cells.Overall, CD44� tumors were visible earlier and grew larger in10/11 mice that received 103 to 105 cells (Table S1). In one mouseinjected with 104 cells, CD44� and CD44� cells yielded tumorsequal in size and appearance. We conclude that the CD44�

fraction is highly enriched for tumorigenic cells, consistent withour in vitro clonal analysis demonstrating that the sorted CD44�

population is 5-fold enriched in colony-forming cells comparedwith the CD44� population. In other published reports, limitingdilutions of CD44� cells have shown similar results, with CD44�

cells unable to form tumors at dilutions �103 (2, 3, 5, 6).Expression of markers was assessed in paraffin sections of

select 30- and 60-day control tumors or after injection of micewith 1 mg of MPA 24 h before necropsy. Sections were stainedfor CD44, CK5, PR, and CK18 (Fig. 5B). At 30 days, colony-likeclusters are ubiquitously CD44�, similar to the in vitro results.Rare CK5� cells (arrow) are seen in some, but not all, of thesein vivo colonies, and PR expression is heterogeneous. StrongCK18 expression confirms their luminal subtype designation. At

60 days, CD44� cells became more common, CK5� cells (arrow)remained rare, and the majority of cells were PR�CK18�.Tumors from mice treated with MPA had an increased numberof CK5� cells compared with controls at 45 days. MPA stillinduced CK5 expression in 70-day-old tumors, but to a lesserextent. ER expression was similarly heterogeneous in earliertumors (30 days) and more uniform in later tumors (60 days)(data not shown).

We conclude that the CD44� subpopulation of ER�PR�

breast cancers is enriched for tumor-initiating cells, as has beendescribed (2, 3, 5, 6). Within the CD44� cell fraction exists a rareER�PR�CK5� subpopulation. The presence of these cells earlyin colony formation, and their nonexpansion over time, suggeststhey are progenitors to luminal ER�PR�CK5� cells that developlater.

DiscussionMolecular genetic profiling has identified two main breast cancersubtypes: the luminal A and B subtype comprising the majorityof breast cancers (�85%) and the basal subtype, comprising3–15% of breast cancers (18). Other minor categories includeErb-B2-positive and ‘‘normal breast like’’ (18). Notably, luminaltumor subtypes are ER�PR�CK18�CK5� while the basal sub-types are ER�PR�CK18�CK5�. In accord with their steroidreceptor positivity, luminal breast cancers have a better prog-nosis than basal cancers (20–23).

The origin of these phenotypically diverse breast cancersubtypes has been explained by two competing theories. Theclonal evolution theory proposes that diverse populations intumors are the descendents of multiple transformed cells. Thegenetic diversity of subpopulations of breast cancer cells sup-ports the postulate that heterogeneous cancer cells may havedifferent origins (24). Alternatively, the cancer stem cell theoryproposes that only the long-lived stem cells of the normal breasthave the capacity to accumulate oncogenic mutations. Theidentification of a small subpopulation of cells in several humancancers that have the unique ability to reconstitute a heteroge-neous tumor in vivo supports this model (2–6). Theoretically, inthe cancer stem cell model, luminal breast cancers would arisefrom normal ER�PR� luminal stem/progenitor cells, whereasthe basal subtypes would originate from normal ER�PR� stemcells in the basal cell layer. Stingl et al. (27) describe two typesof progenitors in normal tissues of the human breast: luminal-specific progenitors and bipotent progenitors that generate bothluminal and myoepithelial cells. These cells could potentially beoncogenic targets.

We describe a rare population of cells having a basal-likephenotype (ER�PR�CK5�) that are contained within theCD44� fraction of luminal-like ER�PR� tumors. It has beenproposed that CK5� cells in the normal human breast representa stem/progenitor fraction that has the capacity to generatemyoepithelial (SMA�) or luminal epithelial (CK8/18�) lineages(25). Analogously, CK5� basal stem cells of the prostate areproposed to populate both the AR�PSA� basal and AR�PSA�

luminal epithelial cells (26). In our hands, rare CK5� cells canbe found in all ER�PR� human breast cancer cell lines, includ-ing T47D, MCF7, and ZR75 (data not shown). The stem-likeCD44�CD24� fraction of primary human breast cancers wasrecently described as ER�/lowPR�/low and enriched for CK5 (24).The CD44� fraction of our xenograft tumors cells was ESA� andCD49f� similar to previously described luminal-specific breastprogenitor cells (ref. 27; Fig. S5). Within the CD44� population,the ER�PR�CK5� are likely to be colony initiators, becausethey are present at early colony formation (3–10 cells percolony), but their population is static and does not grow withcolony expansion (Fig. 4A).

The role of estrogens and progestins in breast epithelial stemcells or tumorigenic breast cancer cells is undefined. Progester-

Fig. 4. Relationship between CK5 positivity, PR expression, and colony size.Cells isolated from ER�PR� T47D tumor xenografts were plated into minimalmedia in six-well dishes overlaid with glass coverslips and allowed to formcolonies for 3–14 days. Cells were treated with either vehicle (control) or 10 nMMPA for 24 h before fixation and staining for CK5 and PR. (A) Relationshipbetween colony size (log) and the number of cells positive for CK5 (greentriangles) and PR (red triangles) in vehicle (control) or progestin-treatedcolonies. Lines represent linear regression for CK5 (m � �0.03 � 0.01, r2 �0.18) and PR (m � 1.01 � 0.02, r2 � 0.98) in 49 control colonies and CK5 (m �0.37 � 0.04, r2 � 0.64) and PR (m � 1.05 � 0.03, r2 � 0.97) in 45 MPA-treatedcolonies. (B) Representative CK5 (green)/PR (red) immunostaining of vehicleand MPA-treated small colonies (3 days, 10–15 cells, Left) and large colonies(7–14 days, �50 cells, Right). Slides were counterstained with DAPI (blue).(Scale bars, 10 �m.)

Horwitz et al. PNAS � April 15, 2008 � vol. 105 � no. 15 � 5777

CELL

BIO

LOG

Y

Dow

nloa

ded

by g

uest

on

Janu

ary

17, 2

021

one is often considered to be a ‘‘differentiating’’ hormone in thebreast, although there are data to suggest that it has proliferativeactions as well (28). Clinical studies indicate that progestin usein hormone replacement therapy is associated with increasedbreast cancer incidence (29). The reasons for this are unclear. Inour hands, progestin treatment of CD44� colonies in vitro andearly tumors in vivo ‘‘reactivates’’ CK5 expression in a subpopu-lation of PR�CK5�, producing unusual PR�CK5� cells. Theextent of CK5 reactivation diminishes in larger colonies andtumors, suggesting that cells further removed from colonyinitiation have lost the capacity to express the CK5 marker. Wepostulate that the PR�CK5� state is transitory between thePR�CK5� progenitors and more differentiated PR�CK5� cells.This is supported by PR/CK5 immunohistochemistry on normalhuman breast and ductal carcinomas. In both cases, CK5 and PRexpression is mutually exclusive, although CK5� and PR� cellsare often found in close proximity (Fig. S6 and Fig. S7).

In summary, we postulate that the ER�PR� luminal breastcancer subtype contains a subpopulation of basal-likeER�PR�CK5� but luminal-specifying progenitor cells. If ourmodel is correct, then in the case of ER�PR� breast cancers,endocrine therapies that target ER would be ineffective againstthe ER�PR� progenitor cells, providing an avenue for tumorrecurrence.

Materials and MethodsReagents. T47D cells stably expressing ZsGreen were described in ref. 30.DMEM, MEM, and FBS were purchased from Life Technologies. Antibodies forimmunohistochemistry included CK5 mAb (XM26, Novacastra), CD44 mAb(156–3C11, Lab Vision), CD44 rabbit mAb (EPR1013Y, Epitomics), CD24 mAb(SN3b, Lab Vision), PR mAb (1294, Dako), PR rabbit mAb (SP2, Lab Vision), andCK18 rabbit mAB (AP1021, Calbiochem). Alexa Fluor 488- and 555- conjugatedsecondary antibodies were from Molecular Probes (In Vitrogen). An AlexaFluor 647 conjugated CD44 antibody and matched isotype control (IgG2a)antibody (F10–44-2, AbD Serotec) were used for FACS sorting.

Experimental Animals and Xenograft Tumor Growth. All animal procedures wereperformed under a protocol approved by the University of Colorado InstitutionalAnimal Care and Use Committee. Ovariectomized nu/nu mice were purchasedfrom Harlan Sprague–Dawley at 4–6 weeks of age. The growth of T47D human

breast cancercells intosolid tumors in thesemicehasbeendescribed inref.19.Forthe current experiments, tumors were grown in the no. 4 mammary glands for8–10 weeks. Animals were supplemented with silastic implants containing estra-diol only or estradiol plus the progestin MPA (19).

Isolation and FACS Sorting of Tumor Cells. Processing of xenograft tumors forcell sorting was performed by using a protocol adapted from Patrawala et al.(3). Briefly, solid T47D-ZsGreen tumors were removed from mice and mincedinto small pieces in 2 ml of DMEM � 5% FBS. Tumor pieces were collected bycentrifugation at 150 � g for 5 min, rinsed once in PBS, and digested inAccumax at 10 ml/g (Innovative Cell Technologies) for 30 min at 37°C, followedby 30 min at room temperature. Cells in the supernatant were collected andfiltered through a 40-�m nylon mesh (BD Biosciences). Lineage-positive (Lin�)host cells were depleted by using the mouse MACS Lineage Cell Depletion Kit(Miltenyi Biotech). The enriched human cancer cell suspension was blocked inphenol red free MEM plus 10% normal human serum (Innovative Research).Alexa Fluor 647-conjugated CD44 or the isotype control antibody were addeddirectly to the cells for 30 min at room temperature. Cells were washed with10� volume MEM and sorted by using a MoFlo sorter based on viability (DAPI),green fluorescence (human cells), and red fluorescence (CD44� cells).

3D Matrigel Cultures and Colony Formation in Minimal Media. Eight-wellchambers (Lab-Tek Chamberslide) were loaded with 100 �l of Growth FactorReduced Matrigel (Becton-Dickinson) and allowed to set for 30 min. SortedCD44� or CD44� cells (2–4 � 103 cells) were overlaid onto the gel in 200 �l ofMEM � 5% FBS. Cultures were allowed to grow for 3–4 weeks, preserved withHistoGel (Richard-Allen Scientific), fixed in 4% paraformaldehyde, and em-bedded in paraffin. Vehicle (ethanol) or MPA (10 nM) was added to chambers24 h before fixation. For assessment of early colony formation, unsorted singlecell suspensions (104 cells) were plated in six-well plates (35-mm wells) con-taining glass coverslips in minimal media (MEGM) containing B27 supplement(InVitrogen), 20 �g/ml each of EGF and FGF (Peprotech), and 4 �g/ml heparin(Sigma). Cells were fixed and stained for CK5 and PR and counterstained withDAPI at days 3, 7, and 14. At each time point, vehicle or 10 nM MPA was addedfor 24 h before fixation. Colonies were identified under the DAPI filter asclusters of three or more cells at day 3, 20 or more cells at day 7, and 50 or morecells at day 14. Colonies were counted for total number of cells and for thenumber of CK5�, PR�, and CK5�PR� cells. GraphPad Prism (version 4.0) wasused to plot data and perform linear regression analyses. Best-fit values forslope (m) and goodness of fit (r2) are noted.

In Vivo Tumorigenesis Studies. Cells were isolated from T47D-ZsGreen humanbreast tumor xenografts grown for 8–10 weeks in ovariectomized immature

Fig. 5. The CD44� fraction of T47D xenografts is enriched for tumorigenic cells, which form tumors that acquire luminal-like properties over time. An equalnumber of CD44� ESA�ZsGreen� and CD44� ESA �ZsGreen� cells (104) were implanted into opposite abdominal mammary glands of recipient estrogen-treatedovariectomized nude mice. Tumor formation was monitored by fluorescence imaging, and tumors were excised from select mice at the indicated dayspostimplantation. (A) Fluorescent image of CD44� and CD44� tumors in one mouse dissected after necropsy at day 30 (Left) and in vivo fluorescent images ofbilateral tumor pairs from representative mice bearing 28- and 45-day tumors (Left and Center). (B) Immunohistochemical analyses of CD44� tumors grown inmice for 30 or 60 days. Paraffin sections were stained for CD44, CK5 (positive cells indicated by arrows), PR, and CK18. Select mice were injected with 1 mg ofMPA (P) for 24 h before death. CK5 staining of these tumors is shown at days 45 and 70 (�P, far right). (Scale bars, 10 �m.)

5778 � www.pnas.org�cgi�doi�10.1073�pnas.0706216105 Horwitz et al.

Dow

nloa

ded

by g

uest

on

Janu

ary

17, 2

021

nude mice supplemented with estradiol, as described. For reimplantationstudies, tumors were digested with collagenase IV/hyaluronidase type I-S(Sigma) and human epithelial tumor cells positively selected by using EpCAMmicrobeads (Miltenyi), followed by labeling with CD44 antibody and FACS.Sorted cells were collected by centrifugation, resuspended in 200 �l of Ma-trigel, and injected into the no. 4 mammary glands. CD44� or CD44� cells (103,104, or 105 in different mice) were injected in opposite sides of the samemouse. Tumor growth was monitored weekly by fluorescence imaging withan Illumatool 9900 (Lightools Research). Tumors were harvested on days 30,45, 60, and 70 after injection and processed for immunohistochemistry asdescribed. Select mice were injected with 1 mg of MPA for 24 h before death.

Immunohistochemistry. Immunohistochemistry on paraffin sections of tumorswas described (19). For the current studies, samples were stained with asingle-mouse mAb (CK5, CD44, CD24, PR) and visualized by using the Envisionsystem (Dako) or dual stained by using the antibody combinations CK5/CD44,

CK5/PR, and CD44/PR, where the second antibodies were rabbit mAbs. Sec-ondary antibodies were anti-mouse and -rabbit Alexa Fluor 488 (green) and555 (red)-conjugated, respectively. A Nikon E600 microscope was used forphotography. Images were shot in black and white by using ImagePro soft-ware (Media Cybernetics) and merged in Adobe Photoshop 7.0.

ACKNOWLEDGMENTS. We also thank Dean Tang (M. D. Anderson CancerCenter, University of Texas, Houston) for providing a tumor cell isolationprotocol, Daniel Perez (University of Colorado Health Science Center) for helpwith computer graphics, and Paul Jedlicka for help with pathology. We thankthe flow cytometry, laser capture, and microarray core laboratories of theUniversity of Colorado Cancer Center for their services. This work was fundedby the Susan G. Komen Breast Cancer Foundation (Grant BCTR0402682, toC.A.S), the University of Colorado Cancer Center (C.A.S.), National Institutes ofHealth Grant CA26869 (to K.B.H.), the National Foundation for Cancer Re-search (K.B.H.), the Breast Cancer Research Foundation (K.B.H.), and the AvonFoundation (K.B.H.).

1. Reya T, Morrison SJ, Clarke MF, Weissman IL (2001) Stem cells, cancer, and cancer stemcells. Nature 414:105–111.

2. Al-Hajj M, Wicha MS, Benito-Hernandez A, Morrison SJ, Clarke MF (2003) Prospectiveidentification of tumorigenic breast cancer cells. Proc Natl Acad Sci USA 100:3983–3988.

3. Patrawala L, et al. (2006) Highly purified CD44� prostate cancer cells from xenografthuman tumors are enriched in tumorigenic and metastatic progenitor cells. Oncogene25:1696–1708.

4. O’Brien CA, Pollett A, Gallinger S, Dick JE (2007) A human colon cancer cell capable ofinitiating tumour growth in immunodeficient mice. Nature 445:106–110.

5. Prince ME, et al. (2007) Identification of a subpopulation of cells with cancer stem cellproperties in head and neck squamous cell carcinoma. Proc Natl Acad Sci USA104:973–978.

6. Dalerba P, et al. (2007) Phenotypic characterization of human colorectal cancer stemcells. Proc Natl Acad Sci USA 104:10158–10163.

7. Locke M, Heywood M, Fawell S, Mackenzie IC (2005) Retention of intrinsic stem cellhierarchies in carcinoma-derived cell lines. Cancer Res 65:8944–8950.

8. Hirschmann-Jax C, et al. (2004) A distinct ‘‘side population’’ of cells with high drugefflux capacity in human tumor cells. Proc Natl Acad Sci USA 101:14228–14233.

9. Clarke RB (2006) Ovarian steroids and the human breast: regulation of stem cells andcell proliferation. Maturitas 54:327–334.

10. Asselin-Labat ML, et al. (2006) Steroid hormone receptor status of mouse mammarystem cells. J Natl Cancer Inst 98:1011–1014.

11. Welm BE, et al. (2002) Sca-1(pos) cells in the mouse mammary gland represent anenriched progenitor cell population. Dev Biol 245:42–56.

12. Sleeman KE, et al. (2007) Dissociation of estrogen receptor expression and in vivo stemcell activity in the mammary gland. J Cell Biol 176:19–26.

13. Zeps N, Bentel JM, Papadimitriou JM, D’Antuono MF, Dawkins HJ (1998) Estrogenreceptor-negative epithelial cells in mouse mammary gland development and growth.Differentiation 62:221–226.

14. Booth BW, Smith GH (2006) ER alpha and PR are expressed in label-retaining mammaryepithelial cells that divide asymmetrically and retain their template DNA strands.Breast Cancer Res 8:R49.

15. Clarke RB, et al. (2005) A putative human breast stem cell population is enriched forsteroid receptor-positive cells. Dev Biol 277:443–456.

16. Dontu G, Al-Hajj M, Abdallah WM, Clarke MF, Wicha MS (2003) Stem cells in normalbreast development and breast cancer. Cell Prolif 36 Suppl 1:59–72.

17. Dontu G, El-Ashry D, Wicha MS (2004) Breast cancer, stem/progenitor cells and theestrogen receptor. Trends Endocrinol Metab 15:193–197.

18. Perou CM, et al. (2000) Molecular portraits of human breast tumours. Nature 406:747–752.

19. Sartorius CA, Harvell DM, Shen T, Horwitz KB (2005) Progestins initiate a luminal tomyoepithelial switch in estrogen-dependent human breast tumors without alteringgrowth. Cancer Res 65:9779–9788.

20. Abd El-Rehim DM, et al. (2004) Expression of luminal and basal cytokeratins in humanbreast carcinoma. J Pathol 203:661–671.

21. Malzahn K, Mitze M, Thoenes M, Moll R (1998) Biological and prognostic significanceof stratified epithelial cytokeratins in infiltrating ductal breast carcinomas. VirchowsArch 433:119–129.

22. Schaller G, et al. (1996) Elevated keratin 18 protein expression indicates a favorableprognosis in patients with breast cancer. Clin Cancer Res 2:1879–1885.

23. van de Rijn M, et al. (2002) Expression of cytokeratins 17 and 5 identifies a group ofbreast carcinomas with poor clinical outcome. Am J Pathol 161:1991–1996.

24. Shipitsin M, et al. (2007) Molecular definition of breast tumor heterogeneity. CancerCell 11:259–273.

25. Bocker W, et al. (2002) Common adult stem cells in the human breast give rise toglandular and myoepithelial cell lineages: a new cell biological concept. Lab Invest82:737–746.

26. Uzgare AR, Xu Y, Isaacs JT (2004) In vitro culturing and characteristics of transitamplifying epithelial cells from human prostate tissue. J Cell Biochem 91:196–205.

27. Stingl J, Raouf A, Emerman JT, Eaves CJ (2005) Epithelial progenitors in the normalhuman mammary gland. J Mammary Gland Biol Neoplasia 10:49–59.

28. Graham JD, Clarke CL (1997) Physiological action of progesterone in target tissues.Endocr Rev 18:502–519.

29. Rossouw JE (2002) Risks and benefits of estrogen plus progestin in healthy postmeno-pausal women: principal results from the Women’s Health Initiative randomizedcontrolled trial. J Am Med Assoc 288:321–333.

30. Harrell JC, et al. (2006) Estrogen receptor positive breast cancer metastasis: alteredhormonal sensitivity and tumor aggressiveness in lymphatic vessels and lymph nodes.Cancer Res 66:9308–9315.

Horwitz et al. PNAS � April 15, 2008 � vol. 105 � no. 15 � 5779

CELL

BIO

LOG

Y

Dow

nloa

ded

by g

uest

on

Janu

ary

17, 2

021