FGFR2 Promotes Breast Tumorigenicity through Maintenance of Breast Tumor-Initiating Cells

Emerging evidence suggests that some cancers contain a population of stem-like TICs (tumor-initiating cells) and eliminating TICs may offer a new strategy to develop successful anti-cancer therapies. As molecular mechanisms underlying the maintenance of the TIC pool are poorly understood, the development of TIC-specific therapeutics remains a major challenge. We first identified and characterized TICs and non-TICs isolated from a mouse breast cancer model. TICs displayed increased tumorigenic potential, self-renewal, heterogeneous differentiation, and bipotency. Gene expression analysis and immunostaining of TICs and non-TICs revealed that FGFR2 was preferentially expressed in TICs. Loss of FGFR2 impaired self-renewal of TICs, thus resulting in marked decreases in the TIC population and tumorigenic potential. Restoration of FGFR2 rescued the defects in TIC pool maintenance, bipotency, and breast tumor growth driven by FGFR2 knockdown. In addition, pharmacological inhibition of FGFR2 kinase activity led to a decrease in the TIC population which resulted in suppression of breast tumor growth. Moreover, human breast TICs isolated from patient tumor samples were found enriched in a FGFR2+ population that was sufficient to initiate tumor growth. Our data suggest that FGFR2 is essential in sustaining the breast TIC pool through promotion of self-renewal and maintenance of bipotent TICs, and raise the possibility of FGFR2 inhibition as a strategy for anti-cancer therapy by eradicating breast TICs.


Introduction
Stem cells may play an essential role not only in regenerative capacity, but also in the development of cancer [1]. The unique property of stem cells to self-renew must be tightly regulated to obey genetic restriction and meet their environmental needs. Dysregulation of self-renewal may be a key event underlying cancer. When breast undergoes changes of tissue renewal and massive tissue expansion during pregnancy, stem cells in the invading tips of mammary glands, known as terminal end buds (TEBs) are believed to be responsible for these cellular dynamics [2]. The abilities of stem cells in TEBs to extensively proliferate and invade different regions of the organ resemble those of solid tumors.
The notion that tumors are heterogeneous and that tumor cells share certain properties with normal stem cells led to the hypothesis that tumors may contain a subset of self-renewing, stem-like tumorigenic cells, TICs, which drive tumor initiation and growth. In this hypothesis, only the tumor initiating cells are capable of unlimited self-renewal, extensive proliferation, and give rise to heterogeneous progenies while differentiated progenies have a limited proliferative potential [3,4]. The demonstration of TICs in many types of cancer including leukemia, breast, brain, colon, skin, head and neck, liver, and pancreatic cancer supports the concept of tumor hierarchy [5][6][7][8][9][10][11][12][13]. Furthermore, TICs have been proposed to be responsible for tumor recurrence. Based on this view, therapeutic strategies for selectively eradicating tumorinitiating cells should lead to successful curative therapies for cancer. However, there is little evidence to support this concept, mainly due to the poor understanding of the molecular mechanisms underlying tumor initiation and the stem-like function of TICs.
Mouse models have been attractive tools to study tumorinitiating cells with their unlimited transplantation assays in many types of cancer [14][15][16]. The recent evidence suggests that the different environments between mice and human can influence the xenotransplantation assay in which human tumor cells that are able to survive in a foreign environment may be selected [17,18]. Therefore, using mice models we can avoid some of the intrinsic problems of measuring stemness through human into mouse xenotransplantation.
To understand the molecular mechanisms underlying tumor initiation and stem-like properties of TICs, we investigated to identify the regulators that are critical for maintenance of a TIC pool. To do so, we used a mouse mammary tumor virus-polyoma middle T (MMTV-PyMT) transgenic mouse, which is a model of human breast cancer with distinct stages of tumor progression from premalignant stage to malignant carcinomas [19,20]. The cellular origins of this breast tumor that can self-renew and drive tumor initiation are uncertain. The ability to isolate and characterize TICs allowed us to compare gene expression and biological functions between stem-like TICs and differentiated non-TICs. This comparison and functional studies with genetic manipulations demonstrated that FGFR2 is essential in sustaining the breast TIC pool through promotion of self-renewal and inhibition of differentiation of TICs. Moreover, we found human breast TICs enriched in a FGFR2+ population which was sufficient to initiate tumor growth.

Breast Tumor Initiating Cells Are Highly Enriched in the CD29 high CD24 + Population
To understand the molecular mechanisms underlying tumor initiation and self-renewal, we first identified potential TIC and non-TIC populations that were isolated from breast tumors of MMTV-PyMT transgenic mice. Given that many functional properties of normal stem cells are shared by TICs, we isolated and examined subpopulations from the breast tumors, based on expression of cell-surface markers CD29 and CD24, which were used to enrich for mouse normal mammary stem cells [21]. Hematopoietic and endothelial cell lineages from the tumor cells were depleted by staining with antibodies against CD45, TER119 and CD31. Five subpopulations (CD29 high CD24 + , CD29 med CD24 + , CD29 low CD24 + , CD29 high CD24 2 , and CD29 low CD24 2 ) were gated and isolated from the breast tumors by fluorescence-activated cell sorting (FACS), and the purity of the sorted populations was assessed by FACS analysis (Figure 1A and S1). To determine the frequency of TICs in these subpopulations, we performed limiting dilution analyses of tumors by injecting the purified tumor cells in decreasing numbers into NOD/SCID mice ( Figure 1B). The calculated frequency of TICs in the CD29 high CD24 + subpopulation was ,1 in 122, whereas the frequencies of TICs in the other four subpopulations (CD29 med CD24 + , CD29 low CD24 + , CD29 high CD24 2 , and CD29 low CD24 2 ) were as low as 1 in 3,133 to 1 in 9,935 ( Figure 1B), indicating that breast tumor initiating cells are highly enriched in the CD29 high CD24 + subpopulation.
Limited tumor growth from the CD29 med CD24 + , CD29 low CD24 + , CD29 high CD24 2 , and CD29 low CD24 2 subpopulations was observed when a higher number of cells was injected ( Figure 1C). Tumors derived from CD29 high CD24 + population grew faster than other four subpopulations even with low number of cells implanted. For example, tumor growth from 1,000 to 5,000 cells of CD29 high CD24 + population was faster than that from 5,000 cells of most of the other subpopulations ( Figure 1C). In particular, tumor growth from even 500 CD29 high CD24 + cells was faster than that from 1,000 to 5,000 cells from the other four subpopulations ( Figure S2). These results suggest that the CD29 high CD24 + subpopulation is enriched for TICs and has significantly increased tumorigenic potential compared with the other four subpopulations. CD29 high CD24 + Cells Have Self-Renewal and Heterogeneous Differentiation Potentials To determine whether tumor cells in the CD29 high CD24 + population can self-renew and generate heterogeneous progenies, tumors generated from the CD29 high CD24 + population were examined by flow cytometry and serial transplantation. The secondary tumors arising from the primary CD29 high CD24 + population were FACS-sorted into CD29 high CD24 + and non-CD29 high CD24 + (all other four) populations, and then tertiary tumors were generated from the transplanted secondary CD29 high CD24 + population ( Figure 1D). Secondary and tertiary tumors arising from the primary and secondary CD29 high CD24 + population, respectively, contained heterogeneous populations that were originally seen in the primary tumors ( Figure 1D). Limiting dilution analyses of tumors derived from as few as 5 cells revealed that the calculated frequency of TICs in the secondary CD29 high CD24 + population was increased to ,1 in 22, whereas the frequency of TICs in the secondary non-CD29 high CD24 + population was as low as ,1 in 1,208 ( Figure 1E). Importantly, the significant increase in TIC frequency (,1 in 22) in the secondary CD29 high CD24 + population compared to that (,1 in 122) in the primary CD29 high CD24 + population ( Figure 1B) suggests that more tumor initiating cells were generated in the CD29 high CD24 + population through serial transplantations, indicating self-renewing capability of the TICs in the CD29 high CD24 + cells. Very limited tertiary tumor growth from secondary non-CD29 high CD24 + population was evident compared with the CD29 high CD24 + population ( Figure 1F). The tertiary tumor growth from 1,000 secondary non-CD29 high CD24 2 cells was slower than that from 100 secondary CD29 high CD24 + cells ( Figure 1F). These results suggest that the TICs in the CD29 high CD24 + population are capable of self-renewal and heterogeneous differentiation, indicative of stem cell-like properties.
To reconfirm the self-renewal capability of the CD29 high CD24 + tumor cells, we cultured the primary tumor cells isolated from different subpopulations and subjected them to a mammosphere assay in which self-renewing cells can proliferate as threedimensional floating cell clusters (mammospheres) under nonadherent conditions [22]. The primary tumor cells were cultured at a clonal density to ensure that each mammosphere was generated from a single cell. The CD29 high CD24 + cells displayed markedly (P,0.02) higher ability to form mammospheres (.50 mm in diameter) than tumor cells from the other subpopulations ( Figure 2A). The few spheres that formed from the CD29 high CD24 2 cells were mostly necrotic ( Figure 2A). Approximately 16 mammospheres were formed per 2000 seeded cells in the CD29 high CD24 + subpopulation, indicating that 1 in 125 cells can form mammospheres. This frequency (1:125) of mammosphere-forming cells in the CD29 high CD24 + population was similar to the frequency (1:122) of TICs in that primary population, calculated by limiting dilution analyses of tumors ( Figure 1B). These results suggest that the CD29 high CD24 + population is enriched for cells that show increased self-renewal activity compared to the other populations.

FGFR2 Is Preferentially Expressed in Breast TICs
To identify genes that regulate stem-like properties in TICs, we analyzed gene expression profiles of stem-like TICs in comparison with lineage-restricted non-TICs. Microarray analysis was performed to examine the expression levels of mRNAs in a TIC (CD29 high CD24 + ) and three non-TIC populations (CD29 low CD24 + , CD29 high CD24 2 and CD29 low CD24 2 ) that were FACS-sorted from primary MMTV-PyMT tumors. The fold changes of expression values in the TIC population compared to the three non-TIC populations were analyzed (for microarray data filtering, see materials and methods). This analysis led to the identification of seven genes of which the mRNA levels are markedly upregulated in TICs compared to non-TICs: syndecan 4 (SDC4), gamma-aminobutyric acid A receptor subunit alpha 4 (GABRA4), fibroblast growth factor receptor 2 (FGFR2), colony stimulating factor 3 (CSF3), Forkhead-box A1 (FOXA1), ethanolamine kinase 1 (ETNK1), and CDC42 binding protein kinase gamma (CDC42BPG) ( Figure 3A). While these genes have not been directly examined for their relationship to stem-like properties of tumor-initiating cells, the majority of these genes have been suggested to be implicated in development, stem cell activities and mammary carcinogenesis [30][31][32][33][34][35][36]. We further validated the mRNA levels of the seven genes in TICs and non-TICs using quantitative real-time PCR ( Figure 3B). TICs exhibited significantly higher (17 to 39-fold) expression of mRNAs of FOXA1, FGFR2, and GABRA4 compared to non-TICs.
Recent studies have shown that FGFR2, which encodes a receptor tyrosine kinase, is critical for the maintenance and proliferation of terminal end buds (TEBs), where stem cells are found to be active during mammary gland development [37,38]. FGFR2 also may play an important role in human breast cancer [32,[39][40][41]. However, the mechanism by which FGFR2 functions as a risk factor in breast cancer remains unknown. Given our expression profiles displaying the high expression of FGFR2 in breast TICs, we investigated the role of FGFR2 in tumor initiation through the self-renewal and maintenance of breast TICs.
To determine whether the upregulated expression of FGFR2 mRNA in TICs (microarray data and Figure 3B) corresponds to an increased level of FGFR2 protein, we analyzed the expression levels of FGFR2 protein in TICs and non-TICs. Flow cytometry analysis revealed that expression of FGFR2 protein is higher (5.4 to 10.2-fold) in TICs compared to CD29 low CD24 + and CD29 high CD24 2 cells ( Figure 3C).

Loss of FGFR2 Suppresses Breast Tumor Growth and Inhibits Oncogenic Signaling
To validate FGFR2 oncogenic function in MMTV-PyMT breast tumors, we transduced primary MMTV-PyMT breast tumor cells with lentiviral short hairpin RNAs (shRNAs) targeting FGFR2 (shFGFR2). Two different shRNAs targeting different regions of FGFR2 displayed greater than 70% knockdown of the FGFR2 mRNA level ( Figure 4A) and a strong reduction in the level of FGFR2 protein ( Figure 4D) in tumor cells stably transduced with the lentiviral shFGFR2s. The shFGFR2s significantly inhibited cell proliferation in vitro ( Figure 4B) and anchorage independent growth in soft agar ( Figure 4C) as compared to non-targeting shRNA (shNT). This suppressed growth was accompanied by markedly reduced phosphorylation of Erk1/2 MAPK and decreased phosphorylation of FRS2, downstream targets of FGFR2 signaling ( Figure 4D and S6), indicating the dependence of downstream oncogenic signaling on FGFR2 expression.
To evaluate the effect of shFGFR2 on breast tumor growth in vivo, breast tumor cells transduced with the lentiviral shFGFR2 were injected in the mammary fat pad of NOD/SCID mice. Loss of FGFR2 resulted in marked suppression of breast tumor growth ( Figure 4F). To exclude the possibility that the observed growth inhibition was due to off-target effects of the shRNAs, we performed rescue experiments by ectopically expressing shRNAresistant FGFR2 ( Figure 4E) in tumor cells suppressed by shFGFR2. Restoration of FGFR2 rescued the shFGFR2-driven growth defect and was sufficient to re-initiate tumor growth in vivo , whereas the empty vector control failed to do so ( Figure 4F), suggesting that the antiproliferative effects of shFGFR2 are not offtarget effects.

Loss of FGFR2 Impaired Self-Renewal of Breast TICs and led to a reduced TIC Pool
To investigate whether the impaired growth of breast tumor cells transduced with shFGFR2 was due to reduced numbers of TICs, we analyzed frequencies of TIC and non-TIC subpopulations in shRNA-transduced breast tumors using flow cytometry. Indeed, shFGFR2-transduced tumor cells exhibited a strong reduction (64 to 70%) in the TIC subpopulation (CD29 high CD24 + ) compared to shNT-transduced tumor cells ( Figure 5A and 5B). In contrast, the shFGFR2-transduced tumor cells displayed a significant increase (65 to 67%) in the non-TIC (CD29 high CD24 2 ) subpopulation compared to shNT-transduced tumor cells ( Figure 5A and 5B). These results suggest that loss of FGFR2 leads to decreased numbers of TICs and increased number of lineage-restricted non-TICs.
To determine whether decreased number of TICs upon FGFR2 loss was due to an impaired self-renewal capacity of TICs, we transduced the purified TICs with lentiviral shFGFR2 and then formation. The CD29 high CD24 + population isolated from the secondary tumors displayed significantly (P = 3.6610 212 ) higher frequency of TICs compared to non-CD29 high CD24 + populations. (F) Comparison of the growth of tertiary breast tumors in NOD/SCID mice which were generated from purified tumor cells of the secondary CD29 high CD24 + or non-CD29 high CD24 + populations. doi:10.1371/journal.pone.0051671.g001 FGFR2 Maintains a Tumor-Initiating Cell Pool PLOS ONE | www.plosone.org examined their ability to form TIC spheres. The sphere-forming capability of TICs was significantly reduced (8 to 19-fold) by loss of FGFR2 as compared with that of shNT-transduced TICs ( Figure 5C).
We next investigated whether loss of FGFR2 affects the differentiation potential of the breast tumor cells. The shRNAtransduced tumor cells were differentiated on collagen-coated plates and then costained for lineage-specific markers K18, K14, and SMA ( Figure 5D). The quantification of immunofluorescence images revealed that shFGFR2-transduced tumor cells had ,3fold lower frequency of bipotent precursor-like cells (K18 + K14 + ) compared to the control cells ( Figure 5D and 5E). In contrast, the majority of FGFR2-knockdown tumor cells gave rise to the terminally differentiated, myoepithelial lineage (SMA + and K18 2 K14 + ) ( Figure 5D and 5E). As the non-TIC populations were shown to enrich for SMA+ cells ( Figure 2C and 2D), this data suggest that the increase in SMA+ cells in the FGFR2 shRNAtransfected cells results from the decrease in the TIC population and the increase in the non-TIC population upon FGFR2 knockdown, which is consistent with the flow cytometry data ( Figure 5A and 5B).
Furthermore, we examined the differentiation potential of the transplanted tumors in vivo upon FGFR2 knockdown. The immunofluorescence staining for K18, K14 and SMA in those tumors revealed that shFGFR2-transduced cells containing enriched non-TIC populations generated the breast tumors with markedly reduced numbers of bipotent precursor-like cells (K18 + K14 + ) and increased numbers of myoepithelial cells (K18 2 K14 + and SMA + ) ( Figure S3A and S3B), compared to the shNT-transduced tumors, indicating terminal differentiation of a majority of tumor cells into myoepithelial lineages in tumors in vivo, in accordance with the observation in vitro ( Figure 5D and 5E). The shNT-transduced tumors contained the bipotent precursor-like cells (K18 + K14 + ), luminal epithelial cells (K18 + K14 2 ) and myoepithelial cells (K18 2 K14 + and SMA + ) ( Figure S3A and S3B). These immunohistology patterns of shNT-transduced tumors were very similar to those of the tumors derived from primary MMTV-PyMT tumor cells. Ectopic expression of FGFR2 in shFGFR2-transduced tumors led to a considerable increase in bipotent precursor-like cells (K18 + K14 + ) ( Figure S3A), suggesting that restoration of FGFR2 rescued the defect in bipotent capacity driven by knockdown of FGFR2. We also examined the frequencies of TIC and non-TIC subpopulations in shRNAtransduced breast tumors by flow cytometry. Tumors transduced with lentiviral shFGFR2 displayed a strong reduction (,63%) in the TIC subpopulation (CD29 high CD24 + ) and a significant increase (,71%) in the non-TIC (CD29 med CD24 + ) subpopulation compared to shNT-transduced tumors ( Figure S3C). Conversely, restoration of FGFR2 in shFGFR2-transduced tumors led to a marked increase (,60%) in the TIC subpopulation compared to shFGFR2-transduced tumors ( Figure S3C). Taken together, these results suggest that FGFR2 plays an important role in sustaining the breast TIC pool through promotion of self-renewal and maintenance of bipotency of TICs. To investigate whether inhibition of FGFR2 kinase activity would suppress the growth of breast tumors and TICs, we treated breast tumors with TKI258 (formerly CHIR258), an inhibitor of receptor tyrosine kinases including FGFRs [42][43][44][45][46]. Treatment of PyMT breast tumors with TKI258 resulted in growth suppression of tumors ( Figure 6A). The growth inhibition by TKI258 in vivo was accompanied by significantly reduced phosphorylation of FGFR2 and Erk1/2 ( Figure 6B), indicating the dependence of breast tumor growth on activation of FGFR2 and its downstream targets. To determine whether the inhibited growth of tumors treated with FGFR inhibitor was due to reduced numbers of TICs, we analyzed TIC and non-TIC subpopulations in TKI258-treated breast tumors using flow cytometry. Treatment of tumors with the FGFR inhibitor in vivo resulted in a ,44% decrease in the TIC subpopulation (CD29 high CD24 + ) compared to the control group ( Figure 6C and 6D). In contrast, the treatment with the FGFR inhibitor led to an increase in the non-TIC subpopulations (,34% increase in CD29 med CD24 + , ,66% in CD29 low CD24 + , and  ,64% in CD29 high CD24 2 subpopulations) ( Figure 6C and 6D). These results suggest that inhibition of FGFR2 kinase activity can suppress the growth of breast tumors probably by blocking the selfrenewing capacity of TICs.

Human Breast TICs Were Found Enriched in a FGFR2+ Population that Was Sufficient to Initiate Tumor Growth
Thus far, we demonstrated the role of FGFR2 in the maintenance of mouse breast TICs. We next explored whether FGFR2 might play a role in maintaining a stem-like TIC pool in human breast cancer. First, we examined expression levels of FGFR2 in patient-derived breast tumors. Quantitative real-time PCR revealed a strong increase (32-to 293-fold) in FGFR2 mRNA levels in 2 of 26 breast tumors (,8%) compared to a normal breast sample ( Figure 7A). Furthermore, flow cytometry analysis confirmed that high (BT5 and BT12) or low level (BT8 and BT25) of FGFR2 protein ( Figure 7B Figure 7C). We also investigated whether the FGFR2+ population is enriched for human breast tumor-initiating cells by monitoring aldehyde dehydrogenase (ALDH) activity since elevated ALDH activity has been shown in normal and malignant human mammary stem cells [47,48]. Flow cytometry analysis revealed ,6-fold higher ALDH activity in the FGFR2+ population compared to the FGFR2-population, indicating that human breast TICs were enriched in the FGFR2+ population that was sufficient to initiate tumor growth ( Figure 7D). This result suggests that FGFR2 may play important roles in the proliferation and survival of human breast tumor-initiating cells.

Discussion
Many different types of tumors contain a subset of stem-like tumor-initiating cells, which drive tumor initiation and recurrence. Thus, therapeutic strategies that selectively target TICs are proposed to lead to the development of successful anti-cancer therapies. The molecular mechanisms underlying tumor initiation and stem-like function of TICs, however, are poorly understood.
To understand these mechanisms, we investigated to identify essential regulators for the maintenance of breast TICs. We first identified and characterized TICs and non-TICs isolated from a mouse breast cancer model. We demonstrated that the breast TICs, which are highly enriched in the CD29 high CD24 + subpopulation, drive tumor initiation and possess stem-like capabilities including self-renewal, heterogeneous differentiation, and bipotency, compared with other subpopulations. Comparison of gene expression between breast TICs and non-TICs revealed that FGFR2 is upregulated in the breast TICs. Our studies suggested that FGFR2 is essential in sustaining the breast TIC pool through promotion of self-renewal and maintenance of bipotency of TICs. Our studies contribute to the characterization of stem-like function of breast tumor-initiating cells and to the understanding of the associated molecular mechanisms. In addition, our results targeting FGFR2 suggest that there are specific targets that may be used to successfully eliminate TICs in human cancers. This is the first report demonstrating convergence of two recent findings: FGFR2 function in breast cancer and in normal mammary stem cell maintenance. FGFR2 was recently identified as a risk factor in breast cancer from several genome-wide association studies for breast cancer [32,[39][40][41]. Indeed, FGFR2 is amplified and overexpressed in 4-12% of human breast cancers [49][50][51]. A mouse mammary tumor virus (MMTV) insertional mutagenesis screen for genes associated with mammary cancer also identified FGFR2 and FGF10 [52]. Independent of these roles of FGFR2 in breast cancer, recent studies in mammary gland development revealed that FGFR2 is essential in the maintenance of terminal end buds (TEBs) where mammary stem cells are active during mammary gland development [37,38]. Although independent studies demonstrated that FGFR2 plays a significant role in normal or malignant breast tissue, none of these findings made FGFR2 signaling as a functional link between the normal mammary stem cells and malignant mammary stem cells.
It has been speculated that the self-renewal regulators in normal stem cells may be shared by malignant stem cells or TICs, since malignant stem cells have been proposed to derive from the transformation of normal stem or progenitor cells through dysregulated self-renewal [4]. The potential role of FGFR2 in mammary stem cell maintenance in TEBs [37,38] is further supported by the requirement of FGF10-FGFR2IIIb signaling for embryonic and postnatal mammary gland development [53,54]. In addition, a role for FGFR2 in stem cell maintenance is reinforced by the evidence of FGFR2 functions in a variety of stem cells. These functions include self-renewal and proliferation of the undifferentiated state of multipotent trophoblast stem cells during embryogenesis [55][56][57], proliferation of osteogenic stem cells [58], and maintenance of self-renewal and undifferentiated growth of human embryonic stem cells [59].
Activating somatic mutations of FGFR2 have been reported in some cancers including lung squamous cell carcinoma, gastric cancer, cervical carcinoma, and endometrial carcinoma, implicating its role in cancer development [60][61][62]. We sequenced the regions of exons (exon 3, 6, 7, 8, 9, 12, and 14) in which somatic mutations of FGFR2 have been reported in the other carcinoma cases [60,63]. Although we found one SNP, rs1047100 (696ARG) in exon 6 of FGFR2 in 60% of the 30 primary human breast tumors analyzed, we failed to detect any of the previously described FGFR2 somatic mutations in either 30 primary human breast tumors or in MMTV-PyMT tumors (unpublished data), suggesting that FGFR2 somatic mutations may be an infrequent event in breast cancer. However, several recent genome-wide association studies demonstrated that germ-line polymorphisms in intron 2 of FGFR2 are associated with breast cancer susceptibility [32,[39][40][41] emphasizing the importance of FGFR2 in breast cancer. The mechanism by which FGFR2 functions as a risk factor in breast cancer, however, remains unknown. The present findings that FGFR2 is an essential regulator for the maintenance of breast TICs may answer questions, at least in part, in breast cancer.
Alternative splicing in the third immunoglobulin-like domain of FGFR2 mRNA results in the formation of FGFR2IIIb and FGFR2IIIc isoforms ( Figure S4A). This differential splicing determines the ligand specificity of FGFR2 in a tissue specific manner [64,65]. Epithelial cells express FGFRIIIb, whereas mesenchymal cells express FGFR2IIIc ( Figure S4A). To investigate which isoform of FGFR2 is present in MMTV-PyMT tumors, we sequenced the cDNAs generated from mRNA regions common to both FGFR2IIIb and FGFR2IIIc isoforms ( Figure S4A and  S4B). This analysis revealed that MMTV-PyMT tumors express FGFR2IIIb isoform which is exclusively expressed by epithelial cells. Further studies will be required to understand a functional link between the FGF ligands (to the particular FGFR2 isoform) and the FGFR2 roles in TIC maintenance.
TKI258 is an orally bioavailable kinase inhibitor of the FGFRs. Its activity against VEGFR and PDGFR, along with the FGFR inhibitory activity, is responsible for the potent anti-angiogenic component of the compound. TKI258 is now being investigated in phase III and II clinical trials in renal cell carcinoma, breast cancer, myeloma and urothelial cancer. In our studies we have shown that use of TKI298 leads to an inhibition of tumor growth in this mouse model of breast cancer and this data could serve as a predictor for its use in breast cancer patients. Given the multitargeted nature of TKI258, our results emphasize the need to develop FGFR2 specific inhibitors.
Our gene expression comparison between breast TICs and differentiated non-TICs led to the identification of upregulated genes in breast TICs, including FGFR2, FOXA1, GABRA4, SDC4, CSF3, ETNK1, and CDC42BPG. These genes have been shown to be involved in mammary carcinogenesis, development and stem cell activities [30,31,[33][34][35][36]. However, the expression of these genes has not been previously investigated regarding their involvement in maintenance of tumor-initiating cells. Further studies will also be required to understand any functional link between FGFR2 and other genes that were upregulated in TICs. It is possible that the inhibition of other TIC-specific genes along with FGFR2 inhibition produces synergistic anti-tumor activity, more effective at eradicating tumor initiating cells.

Mouse Strain, Clinical Samples, and Tumor Cell Preparation
MMTV-PyMT mice (FVB/n strain) [19], NOD.CB17-Prkdc scid and NOD.Cg-Prkdc scid Il2rg tm1WJ1 /SzJ mice were bred and maintained according to institutional guidelines [the Genomics Institute of the Novartis Research Foundation (GNF) Institutional Animal Care and Use Committee]. The GNF Biomedical Institutional Review Board approved this specific study, and the breast cancer specimens were handled and maintained according to protocols approved by GNF Biomedical Institutional Review Board, ensuring that samples were delivered to GNF with unique identifiers and GNF had no access to any personal patient information, other than pre-existing pathology provided by Asterand (Detroit, MI). Establishment of primary human tumor models for in vivo validation studies was performed in accordance with the animal protocols (P06-147 and P09-243) approved by the GNF Institutional Animal Care and Use Committee for this specific study. Primary tumors were isolated from 3-to 6-monthold MMTV-PyMT mice, and were mechanically dissociated and digested in serum-free mammary epithelial growth media (MEGM, Lonza, Walkersville) containing 200-300 U/ml collagenase (Worthington). The resulting tumor suspension was treated with 0.25% trypsin for 1-2 min and 16 RBC (red blood cell) lysis buffer for 3 min, and filtered to obtain single tumor cells.

Cell Staining and Flow Cytometry
Hematopoietic and endothelial cell lineages from the tumor cells were depleted by staining with antibodies against CD45 and/ or TER119 and CD31. In most cases, FACS-sorted CD45 2 tumor cells contained less than ,0.4% of hematopoietic (TER119 + ) and endothelial (CD31 + ) cell lineages. Otherwise, tumor cells were stained with antibodies against TER119-biotin (BD Pharmingen) and CD31-biotin (BD Pharmingen) to deplete hematopoietic and endothelial cell lineages by magnetic-activated cell sorting

Drug Administration
TKI258 solution was formulated in water freshly every day. Daily oral administration of TKI258 at 50 mg/kg was initiated in tumor-bearing NOD/SCID mice when mammary tumors reached ,150 mm 3 in volume. The tumors were measured every 2-4 days.

Transplantation of Breast Tumor Cells
For limiting dilution transplantation of tumor cells, sorted primary or secondary MMTV-PYMT tumor cells were counted in decreasing numbers (5000 to 5 cells) and resuspended in MEGM media with 50% matrigel, and injected into the mammary fat pads of 4-to 6-week-old NOD.CB17-Prkdc scid or NOD.Cg-Prkdc scid Il2rg tm1WJ1 /SzJ mice. Ten thousand transduced MMTV-PYMT tumor cells with lentiviral shRNAs and retroviral cDNAs were injected. Eighty thousand tumor cells of two sorted populations (FGFR+ and FGFR22) from primary human breast tumor BT5 were injected into the mammary fat pad of the NOD/SCID mice. The tumors were measured once a week. Secondary and tertiary tumors were harvested when the size of those tumors reached the size similar to primary tumors that were isolated from 3-to 6month-old MMTV-PyMT mice.

Proliferation and Soft Agar Assay
Proliferation of tumor cells was determined by cell viability using CellTiter-Glo reagent (Promega). Three thousand transduced cells were plated in 96-well Primaria plates for cell viability assays. To determine the anchorage-independent growth, five thousand or ten thousand tumor cells were seeded in a top layer of 0.3% agar in MEGM with supplements that was placed on the top of a bottom layer of 0.6% agar in DMEM without any supplement in 12-well plate. Colonies (.100 mm in diameter) were counted 3 weeks after seeding.

Immunofluorescence and Quantification
PTA-fixed cells and formalin-fixed paraffin-embedded tumor sections were stained with antibodies against Cytokeratin 18 (Abcam), Cytokeratin 14 (Covance), smooth muscle actin (SMA) (Sigma), and FGFR2 (Santa Cruz). While one representative image per group was presented, at least three images from the different positions in the plate per group were collected using the ImageXpress Ultra laser scanning confocal microscope (Molecular Devices). Image quantification was achieved using the multiwavelength cell scoring analysis module, and an average of three values from three images (different positions) per group was calculated.

Statistics and Limiting Dilution Analyses of Tumors
Limiting dilution analyses of tumors were performed using the limdil function of the 'statmod' package (http://bioinf.wehi.edu. au/software/elda/) for the R statistical programming environment (http://www.r-project.org). 95% confidence intervals were computed for tumor-initiating cell frequencies. The single hit Poisson assumption was validated by likelihood ratio tests and accepted for all dilution series (P.0.05) [66]. To evaluate the difference between groups, the Student's t test was used. Differences of P,0.05 were considered significantly different.

RNA Isolation and Quantitative Real-Time PCR (qRT-PCR)
Total RNA from tumor cells was extracted with the RNeasy kits (QIAGEN) or TRIzol (Invitrogen), according to the manufacturer's instructions. cDNA was prepared using the SuperScript II or III reverse transcriptase (Invitrogen) and subjected to qRT-PCR on 7700HT Fast real-time PCR system (Applied Biosystems) using gene-specific primers. qRT-PCR was performed with either Taqman (Applied Biosystems) or SYBR Green (Applied Biosystems). Gene expression was normalized by b-actin (endogenous control) and calibrator sample using the DCt method to calculate RQ (relative quantity). The difference between the threshold cycle of a target and the threshold cycle of the endogenous control was calculated as DCt = Ct(target)2Ct(endogenous control). The difference between the average DCt value of a target sample and the average DCt for the calibrator sample was calculated as DDCt(test sample) = AvgDCt(test sample)2AvgDCt(calibrator sample). Expression fold value (RQ, Relative Quantity) is calculated as RQ = 2 -DDCt . The following Taqman Gene Expression Assays (Applied Biosystems) were used: mouse FGFR2

Microarray Analysis
RNeasy kits (QIAGEN) or TRIzol (Invitrogen) was used to extract total RNA from tumor cells, according to the manufacturer's instructions. The integrity and concentration of RNA were determined via microfluidic analysis on an Experion instrument (BioRad). GeneChip Two-Cycle Target Labeling Assays (Affymetrix, Santa Clara, CA) were used to amplify cRNA from 40 ng total RNA. Standard Affymetrix protocols were used to process Affymetrix Mouse430_2 microarrays (Affymetrix, Santa Clara, CA). All CEL file images were processed as a single group using gcRMA. The microarray data were serially sorted by fold changes of expression values in the TIC population compared to the other three non-TIC populations in MMTV-PyMT mammary tumors, The data sets were then filtered by three categories following; (1) as the maximum intensity of expression values among four populations is greater than 100, (2) as the fold change of expression values in the CD29 high CD24 + population compared to CD29 low CD24 2 population is greater than 4, and (3) as the minimum intensity of expression values among two samples of CD29 high CD24 + population is greater than 17. This analysis led to the identification of genes that are significantly upregulated in TICs compared to non-TICs. The microarray data described will be deposited in NCBIs Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo/) and will be accessible through a GEO Series accession number.

Immunoblotting and Immunoprecipitation
Tumors were homogenized, and tumor cells were lysed on ice, immunoprecipitated for FGFR2 and immunoblotted with antiphosphotyrosine to evaluate phosphorylation of FGFR2. The membranes were reprobed to evaluate total level of FGFR2 protein. The following antibodies were used for immunoblotting: FGFR2 (Sigma or Santa Cruz Biotech), p44/42 MAP Kinase (Erk1/2) (Cell Signaling Technology), phospho-p44/42 MAP Kinase (p-Erk1/2) (Cell Signaling Technology), b-actin (Cell Signaling Technology), phosphor-tyrosine (R&D Systems), and phospho-FRS2-a (Cell Signaling Technology). under the differentiation condition. The sorted cells from various populations were plated at the same cell density on collagencoated plates. The differentiated cells were stained for the luminal epithelial marker (K18, green), the myoepithelial markers (K14 and SMA, red), and DAPI (nuclei, blue). A significant portion of CD29 high CD24 + cells contains K18 + K14 + (bipotent precursorlike), whereas the majority of cells from the other subpopulations contain lineage-restricted cells. (TIF) Figure S6 Inhibition of Oncogenic Signaling by Loss of FGFR2. Inhibition of downstream target activation upon FGFR2 knockdown. Primary MMTV-PyMT breast tumor cells were transduced with lentiviral short hairpin RNAs (shRNAs) targeting FGFR2 (shFGFR2). The shFGFR2-4 efficiently knocked down the expression of FGFR2 protein and inhibited phosphorylation of Erk1/2 and FRS2, as evidenced by immunoblotting with antiphospho-ERK1/2 (p-ERK1/2) and anti-phospho-FRS2 (p-FRS2). The shFGFR2-1 that partially knocked down the FGFR2 expression had little effect on ERK or FRS2 phosphorylation.

Supporting Information
The membranes were reprobed for actin and Erk1/2 as loading controls. (TIF)