IDENTIFYING AND TARGETING TUMOR-INITIATING CELLS IN THE TREATMENT OF BREAST CANCER

Abstract

Breast cancer is the most common cancer in women (exclusive of skin cancer), and is the second leading cause of cancer-related deaths. Although conventional and targeted therapies have improved survival rates, there are still considerable challenges in treating breast cancer, including treatment resistance, disease recurrence, and metastasis. Treatment resistance can be either de novo - due to traits that tumor cells possess prior to treatment, or acquired, - due to traits that tumor cells gain in response to treatment. A recently proposed mechanism of de novo resistance invokes existence of a specialized subset of cancer cells defined as tumor-initiating cells (TICs), or cancer stem cells (CSC). TICs have the capacity to self-renew and regenerate new tumors that consist of all clonally-derived cell types present in the parental tumor. There are data to suggest that TICs are resistant to many conventional cancer therapies, and survive treatment in spite of dramatic shrinkage of the tumor. Residual TICs can then eventually regrow resulting in disease relapse. It is also hypothesized that TIC may be responsible for metastatic disease. If these hypotheses are correct, targeting TICs may be imperative to achieve cure. In this review, we discuss evidence for breast TICs and their apparent resistance to conventional chemotherapy and radiotherapy, as well as to various targeted therapies. We also address the potential impact of breast TIC plasticity and metastatic potential on therapeutic strategies. Finally, we describe several genes and signaling pathways that appear important for TIC function that may represent promising therapeutic targets.

Introduction

In recent years, breast cancer patients, clinicians, and basic scientists have rightfully celebrated small, but significant, improvements in breast cancer outcome. These improvements have been attributed to better methods for early detection, enhanced screening efforts, and the availability of more effective targeted therapeutics for the treatment of the two largest clinically-defined subtypes of breast cancers - those that express the estrogen receptor (ER⁺) (with, or without co-expression of the progesterone receptor (PR)), and those with overexpression or amplification of the human epidermal growth factor 2 (ErbB2) gene (a.k.a. HER2⁺). These two subtypes of breast cancer account for approximately 70% and 15–20% of all cases, respectively. The remaining subtype, “Triple Negative” breast cancer (TNBC), lacks expression of ER, PR, and HER2. To date, there are no targeted agents to combat TNBC and its prognosis remains poor.

Despite recent progress, several major clinical problems remain. Chief among these are the issues of treatment resistance, disease recurrence, and metastasis. For example, whereas many ER⁺ tumors respond to ER-targeted therapies (antiestrogens (e.g. tamoxifen and aromatase inhibitors like anastrozole, letrozole, and exemestane)), de novo and acquired resistance is common (Burstein, et al. 2014). Similarly, recent clinical trials showed that up to 64% of HER2⁺ patients can show pathological complete response to combination treatment with dual anti-HER2 targeted therapy (Cortazar, et al. 2014; de Azambuja, et al. 2014; Gianni, et al. 2012; Schneeweiss, et al. 2013). However, a significant percentage of patients are resistant to these agents. In TNBC, treatment generally involves use multiagent chemotherapy along with surgery. Unfortunately, not all patients receiving chemotherapy show clinical benefit, and side effects can be significant. In the case of disease recurrence, the recurrent cancer can be refractory to the original treatment.

Breast cancer has long been recognized as a heterogeneous disease, and this heterogeneity has been invoked to explain, at least in part, differences in treatment response, recurrence potential, and metastatic behavior. Tumor heterogeneity exists at the histological and molecular levels within a single tumor (intratumoral), and between different tumors (intertumoral). Recent gene expression profiling is beginning to reveal the full extent of intertumoral heterogeneity. For example, independent of the three clinically-defined subtypes of breast tumors, at least six molecular subtypes of breast cancer have been identified: luminal A, luminal B, HER2-enriched, normal-like, basal-like, and claudin-low (Herschkowitz, et al. 2007; Perou, et al. 2000). The luminal subtypes are generally ER⁺. The HER2-enriched subtype is typically ErbB2⁺ and are also generally ER⁻. Tumors in the basal-like subtype are generally triple-negative. To date, approximately 60–70% claudin-low tumors identified have been triple-negative (Prat and Perou 2011). More recently, TNBC have been evaluated in large numbers and show at least 6 subclasses (Lehmann, et al. 2011).

Although less well-studied than intertumoral heterogeneity, breast tumors also show intratumoral heterogeneity. As in the normal mammary gland, where cellular heterogeneity has been recognized and studied for decades, phenotypic heterogeneity at the cellular level is also common within breast tumors. For example, as is the case in the normal mammary gland in which only 30–40% of cells express ER and PR, in ER⁺ breast tumors, ER⁺ cells express variable levels of ER protein, and up to 99% of all tumor cells may not express any detectable ER at all (Hammond, et al. 2011; Harvey, et al. 1999). In similar fashion, PR is not generally expressed in every cell in PR⁺ tumors. While not currently useful in clinical decision making, expression of many other protein markers (e.g. cytokeratin 5, CD44, CD24, PTCH1, SMO) is also known to vary from cell to cell in some breast cancers (Abd El-Rehim, et al. 2004; Marotta, et al. 2011; Moraes, et al. 2007). In addition to simple variability at the level of protein expression, there is now known to be genetic heterogeneity within tumors. For instance, single cell sequencing data demonstrate that there are extensive clonal diversities within a single tumor, resulting from low-frequency point mutations that evolved during tumor development. Noticeably, the mutation frequency is 13 times higher in TNBCs versus in luminal type tumors, suggesting an increase in heterogeneity in TNBC (Wang, et al. 2014b).

It stands to reason that the observed phenotypic and genetic heterogeneity within a given tumor likely results in functional heterogeneity. Most importantly from a clinical perspective, these heterogeneous tumor cell subpopulations may show different responses to therapy, different division potentials, and varied metastatic properties. Therefore, a better understanding of the mechanisms that contribute to intratumoral heterogeneity in breast tumors, particularly regarding the tumor-initiating subpopulations, is critical to improve current treatment options.

Heterogeneity and the Cancer Stem Cell Hypothesis

The cancer stem cell (CSC) hypothesis has been developed in part to explain the intratumoral heterogeneity. According to this hypothesis, many cancers have a unique subset of cells, referred to as CSCs, that have the capacity to self-renew and give rise to other cancer cell types, creating an hierarchically organized tumor (Visvader and Lindeman 2012). Furthermore, these CSCs are thought to be the main drivers of tumor growth. Evidence also indicates that CSCs are more resistant to conventional therapies, suggesting that they also play an important role in mediating tumor relapse (Creighton, et al. 2009). Therefore, this hypothesis provides a plausible explanation for different types of treatment failure, although the mechanisms of resistance and the proportion of CSCs may vary between different tumors. We and many other groups prefer the term “tumor-initiating cell (TIC)” instead of CSC, to distinguish them from normal stem cells, and to emphasize their tumor initiating capacity.

The identification of TICs in breast tumors

Putative breast TIC subpopulations can be isolated by FACS using several combinations of cell surface and non-cell surface markers (Table 1), followed by functional assays demonstrating their enriched tumorigenic potential. For example, subpopulations of human breast tumor cells have been shown to have enriched tumorigenic potential using limiting dilution transplantation (LDT) assays (Al-Hajj, et al. 2003; Ginestier, et al. 2007). In addition, serial transplantation assays have been performed to demonstrate the self-renewal capacity of breast TICs, and the heterogeneity of the regenerated daughter tumors has provided evidence of their differentiation capacity (Al-Hajj et al. 2003). Mammosphere formation efficiency (MSFE) assays have also been developed as in vitro surrogates for LDT assays to demonstrate the self-renewal capacity of breast TICs (Dontu, et al. 2003). However, MSFE and regenerative cell frequency may not necessarily correlate in all cases, which may limit the interpretation of in vitro assays (Moraes et al. 2007).

Table 1.

Functionally defined TIC markers for human breast tumors and mouse mammary tumors. nr=not reported.

Human Breast TIC markers	Tumor Subtypes	TIC frequency before enrichment	TIC frequency after enrichment	References
CD44+/CD24−/low	nr	nr	nr	Al-Hajj et al. 2003
ALDH+	Basal-like	nr	nr	Ginestier et al. 2007, Liu et al. 2014
STAT3-GFP	Claudin-low (cell line xenografts)	nr	nr	Wei et al. 2014
EpCAM+/CD49f+	Triple Negative	nr	1/71	Lee et al. 2014
CD44+/CD49fhi/CD133/2hi	ER−	nr	nr	Meyer et al. 2010

Mouse Mammary TIC markers	Tumor Subtypes	TIC frequency before enrichment	TIC frequency after enrichment	References
Sca1+/CD24+	Luminal (MMTV-Neu)	nr	1/303	Liu et al. 2007
CD49fhi/CD61hi	Luminal (MMTV-Neu)	nr	1/70	Lo et al.2012
CD24+/CD49f+	Luminal (MMTV-PyMT)	nr	nr	Ma et al. 2012
CD29lo/CD24+/CD61+	Basal-like (MMTV-Wnt1)	nr	1/303	Vaillant et al. 2008
Thy1+/CD24+	Basal-like (MMTV-Wnt1)	nr	nr	Cho et al. 2008
EpCAMlow/CD49fhi	Basal-like (MMTV-Wnt1)	nr	1/79	Feng et al. 2014
CD24hi/CD49fhi	Basal-like (MMTV-Wnt1)	1/685	1/71	Lee et al. 2014
CD29hi/CD24+	Basal-like (p53-null)	1/21,583	1/302	Zhang et al. 2008
TOP-GFP	Basal-like (p53-null)	nr	1/48 ~ 1/225	Zhang et al. 2010
CD24+/CD29+	Basal-like (Brca1-deficient)	nr	nr	Vassilopoulos et al. 2008

To isolate TICs from breast tumors, researchers frequently use fluorescent-conjugated antibodies to cell surface markers, which are specifically expressed or enriched in TIC populations, combined with fluorescence-activated cell sorting (FACS). AI-Hajj et al. published the first markers identified to enrich for human breast TICs in 2003 (Al-Hajj et al. 2003). The authors used two cell surface markers, the glycoproteins CD44 and CD24, to isolate CD44⁺/CD24^−/low/Lin⁻ cells from primary human/xenograft breast tumors and showed that this cell population is more tumorigenic than the other cell populations, suggesting it is enriched for TICs. Importantly, this tumor-initiating advantage is maintained during serial passages and the new generations of tumors contain phenotypically diverse populations (Al-Hajj et al. 2003). Interestingly, CD44 and CD24 markers fail to enrich for TICs in several murine xenograft models that are ER⁻ and triple-negative (Meyer, et al. 2010), suggesting that TIC populations may vary between different breast cancer subtypes.

Surrogate markers for CD44⁺/CD24⁻ cells have also been suggested, including ganglioside GD2 (a glycosphingolipid) and protein C receptor (PROCR) (Battula, et al. 2012; Shipitsin, et al. 2007b). Almost all GD2⁺ cells are CD44⁺/CD24⁻ and knockdown of GD3 synthase, an enzyme involved in the synthesis of GD2, in GD2⁺ cells reduces their TIC properties, suggesting that GD2 and CD44 may be phenotypic indications of underlying mechanisms that drive TIC function. PROCR is a cell surface receptor specifically expressed in CD44⁺ cells. Unlike CD44, which is also expressed in leukocytes and myofibroblasts, PROCR expression is restricted to CD44⁺ epithelial cells. PROCR was initially identified as an embryonic stem cell (ESC) marker (Ramalho-Santos, et al. 2002), suggesting it may play a general role in stem cell function.

More recently, integrins have been identified as human breast TIC markers; as they are differentially expressed between breast TICs and non-TICs. For example, the α6 integrin (CD49f) marks TICs in ER⁻ and triple negative breast tumor xenografts (Lee, et al. 2014; Meyer et al. 2010). Using a combination of CD49f, CD44, and CD24 may further enrich for breast TICs from patient breast tumor samples, compared to CD44 and CD24 alone, as MSFE assays show that the self-renewal capacity is predominantly in the CD44^high/CD24^low/CD49f⁺ population, but not in the CD44^high/CD24^low/CD49f⁻ population (Ghebeh, et al. 2013).

A different set of cell surface proteins have been used to enrich for TICs from mouse mammary tumors. In MMTV-Neu tumors, the stem cell antigen 1 (Sca1)⁺/CD24⁺ combination and the β3 integrin CD61 combined with CD49f (CD49f^hi/CD61^hi) enrich for TICs (Liu, et al. 2007; Lo, et al. 2012). In MMTV-PyMT tumors, the CD24⁺/CD49f⁺ combination enriches for TICs (Ma, et al. 2012). In MMTV-Wnt1 tumors, the CD61⁺/CD29^lo/CD24⁺ combination, the thymocyte antigen 1 (Thy1)⁺/CD24⁺ combination, the EpCAM^low/CD49f^hi combination, and the CD24^hi/CD49f^hi combination all enrich for TICs (Cho, et al. 2008; Feng, et al. 2014; Lee et al. 2014; Vaillant, et al. 2008). In both p53-null and Brca1-deficient tumors, TICs are enriched in the β1 integrin (CD29)^hi/CD24⁺ cells (Vassilopoulos, et al. 2008; Zhang, et al. 2008).

Non-cell surface markers have also been used successfully to enrich for TICs. For example, aldehyde dehydrogenase (ALDH) activity has been used to identify stem cells in mouse mammary glands and human breast tumors (Ginestier et al. 2007). Results showed that ALDH⁺ cells perform better than negative cells in LDT assays and generate tumors that recapitulate the heterogeneity of the parental tumors, in terms of enzyme activity. However, another study found that although many basal/mesenchymal breast cancer cell lines are positive for ALDH activity, the majority of luminal type breast cancer cell lines are negative (Charafe-Jauffret, et al. 2009), suggesting that distinct breast TIC populations may exist in different types of breast tumors. This study also revealed the limited overlap between the CD44⁺/CD24⁻ and ALDH1⁺ populations, suggesting that there may be distinct breast TIC populations even within individual tumors. Therefore, similar to the restricted application of CD44⁺/CD24^−/low/Lin⁻ markers, ALDH activity may only be useful for certain breast tumor subtypes.

As several cellular signaling pathways are key regulators of breast TIC function, fluorescent reporters of their downstream activity can also serve as functional breast TIC markers. For example, a recent study showed that a Wnt signaling reporter is preferentially activated in TICs in basal-like, p53-null mouse mammary tumors, and that cell populations isolated based on expression of the reporter are highly enriched for TICs (Zhang, et al. 2010). The utility of this reporter has not been explored fully in models of human breast cancer.

More recently, a lentiviral STAT3-EGFP reporter was shown to be a potent and functionally relevant breast TIC marker in claudin-low cell line xenograft models of human breast cancer (Wei, et al. 2014) using both MSFE and limiting-dilution transplantation. These fluorescent pathway reporters facilitate the visualization of specific signaling pathway activity in live cells and, more broadly, facilitate stem cell research by enabling direct FACS sorting.

It is worth noticing that there are huge variations between TIC frequencies estimated by different TIC markers. For example, breast cancer cell line MDA231 contains 85% CD44⁺/CD24⁻ cells vs. 0.88% ALDEFLUOR⁺ cells, while breast cancer cell line SKBR3 contains 0% CD44⁺/CD24⁻ cells vs. 95.3% ALDEFLUOR⁺ cells (Charafe-Jauffret et al. 2009; Sheridan, et al. 2006). We summarized in Table 2 the percentage of marker expressing cells in frequently used breast cancer cell lines.

Table 2.

Frequency of TIC markers in frequently used breast cancer cell lines.

Cell Line	% CD44+/CD24−			% CD44+/CD24−/ESA+	% ALDH+
BT474	0	0
HCC1937		57			2.26
HS578T	86	64			0.6
MCF10A	17	3	16	0.25	0.32
MCF12A		5
MCF7	0	0	0	0	0.2
MDA-MB-231	85	76	98	1.8	0.88
MDA-MB-436	72				2.65
MDA-MB-453		0			3.54
MDA-MB-468	3	0
SKBR3	0	0			95.3
SUM1315	97		92	2.4
SUM149			5	2.3	5.96
SUM159		55	95	1.7	5.49
SUM225			1	0.6	2.12
T47D	0	0			0
ZR75.1	0	0			1.02
References	Sheridan et al. 2006	Marotta et al. 2011*	Fillmore et al. 2008*	Fillmore et al. 2008*	Charafe-Jauffret et al. 2009

^*Percentage estimated from bar graph

Researchers tried to consolidate the discrepancy by applying different TIC markers to different molecular subtypes of breast cancers (Ricardo, et al. 2011), as well as arguing that different set of markers represent TICs in different states of MET/EMT (Liu, et al. 2014). Unfortunately, even within the same molecular subtypes, using the same TIC marker set, marker expression does not correlated with TIC frequency estimated by functional assays. For example, claudin-low type human breast cancer cell lines SUM159 and MDA231 contain comparable percentages of CD44⁺/CD24⁻ cells (95% vs. 98%) (Fillmore and Kuperwasser 2008), yet SUM159 exhibits a MSFE much higher than MDA231, while MDA231 display a much higher tumorigenic capacity than SUM159 judged by LDT assays (Wei et al. 2014). Similarly, although the percentage of CD44⁺/CD24⁻/ESA⁺ cells in SUM149 and SUM159 correlate well with their MSFEs (Fillmore and Kuperwasser 2008), it fails to predict the relative tumorigenic capacity of SUM159 and MDA231 cells (Wei et al. 2014).

Furthermore, the lack of correlation between MSFEs and tumorigenic capacities is widely observed. For example, breast cancer cell line SUM149 has a higher MSFE after normalization to proliferation, but lower tumor-formation frequency in LDT assay as compared to SUM159 (Fillmore and Kuperwasser 2008). A similar but more dramatic result was observed between SUM159 and MDA231 xenograft cells (Wei et al. 2014). Recent study also showed that marker bearing cells in different cell lines response to therapies inconsistently, which highlights a lack of correlation between TIC markers and treatment responses (Liu et al. 2014).

To date, none of the identified breast TIC markers are universal. Thus, continuing to search for novel markers, or using novel combinations of current markers, will presumably improve TIC identification and isolation, and lead to a better understanding of breast TICs. Because of the inconsistent interpretation of TIC markers within and across breast cancer samples, it is important to evaluating TIC frequency rigorously by LDT assays whenever possible, rather than simply rely on TIC markers.

Breast TICs may be resistant to conventional systemic therapies

The content of TICs in breast tumors has been shown to be closely related to clinical outcome. For example, most of the triple-negative tumors belong to either basal-like or claudin-low subtypes (Herschkowitz, et al. 2012), with claudin-low tumors having worse prognosis compared to luminal A tumors and a clear enrichment for a TIC-associated signature (Herschkowitz et al. 2012; Prat, et al. 2010). The aggressiveness of primary breast cancers has also been associated with their TIC frequency, with poorly differentiated tumors displaying a higher content of TICs by xenotransplantation experiments (Pece, et al. 2010; Usary, et al. 2013). Consistent with these findings, accumulating evidence has revealed the resistance of breast TICs to diverse types of breast cancer therapies.

Resistance to chemotherapy

Eventual relapse following chemotherapy has been a major challenge in treating breast cancer (1998). The existence of de novo chemoresistant breast TICs could be a major cause of disease relapse. There are two essential pieces of evidence that support this idea: i) Chemotherapy treatment enriches for cells expressing markers of breast cancer TICs. ii) Tumors enriched in markers of breast TICs are comparatively resistant to chemotherapy treatment. For the former, it has been shown that, following chemotherapy, residual breast cancers and cancer cell lines have increased CD44^high/CD24^low subpopulations, increased MSFE and tumor-initiating efficiency, and enriched expression of TIC signature genes (Creighton et al. 2009; Fillmore and Kuperwasser 2008; Li, et al. 2008; Yu, et al. 2007). For the later, it has been shown that breast tumors that have been classified as claudin-low, a molecular subtype that is enriched for TIC features, are relatively resistant to chemotherapy (Usary et al. 2013).

The hypothesized quiescent nature of breast TICs may confer their chemoresistance, because efficient induction of apoptosis by chemo-drugs usually requires cell division (Moore and Lyle 2011; Naumov, et al. 2003). There are several lines of evidence showing that breast TICs are slow cycling: i) Markers from the gene expression signature derived from quiescent human normal mammary stem cells can be used to isolate breast TICs, suggesting the corresponding quiescent nature of breast TICs (Pece et al. 2010). ii) The dye retention side-population (SP) is enriched for TICs in breast cancer cell lines, and shows increased expression of negative cell cycle regulators (Goodell, et al. 1996; Hirschmann-Jax, et al. 2004; Patrawala, et al. 2005; Zhou, et al. 2007). iii) The CD44⁺/CD24^−/low cells isolated from a panel of breast cancer cell lines are slow cycling and resistant to chemotherapies (Fillmore and Kuperwasser 2008). Hence, although it may not be true in all tumors, the slow-cycling nature of breast TICs may serve as a potential mechanism for chemoresistance (Moore and Lyle 2011).

Another potential mechanism that may confer chemoresistance to breast TICs is the increased presence of ATP binding cassette (ABC) transporters, which actively pump drugs out of cells, relative to the bulk of the tumor. The breast cancer resistance protein (BCRP) was first identified in a multi-drug resistant subline of MCF7 and is involved in in vitro multidrug resistance (Doyle, et al. 1998). Likewise, the increased expression of the P-glycoprotein (Pgp), another well-characterized ABC transporter, is associated with doxorubicin resistance in multiple breast cancer cell lines (Turton, et al. 2001). In patient breast tumor samples, high BCRP expression is correlated with high HER2 expression, lymph node metastasis, and advanced stage of breast cancer (Xiang, et al. 2011). Pgp shows higher expression after treatment with chemotherapy, suggesting that tumor cells expressing Pgp may resist to chemotherapy (Rudas, et al. 2003). Furthermore, there is evidence suggesting that those proteins are preferentially expressed in breast TICs. For example, the SP cells of MCF7 cells, which was enriched for TICs, expresses higher levels of ABC transporters than non-SP cells (Zhou et al. 2007). On the other hand, in the MDA-MB-435 cell line, cells expressing BCRP are fast cycling and are not enriched in breast TICs, as compared to BCRP⁻ cells (Patrawala et al. 2005). These data suggest that different mechanisms may play roles mediating TIC chemoresistance in different models, and that there may be a trade-off between slow cycling and drug efflux properties in conferring drug resistance.

Resistance to radiotherapy

Malignant cells are usually rapidly dividing, and their DNA damage repair systems frequently fail to perform dependably (Nie 2012). Therefore, malignant cell growth can be effectively controlled by radiotherapy. More than 50% of breast cancer patients receive radiotherapy during the treatment of their disease (Langlands, et al. 2013). Unfortunately, distant metastasis and local recurrence still occur due to treatment resistance (Langlands et al. 2013), especially in patients with basal-like, triple-negative breast cancer (Nguyen, et al. 2008).

Accumulating evidence suggests that breast TICs are radioresistant (Morrison, et al. 2011; Phillips, et al. 2006; Prat et al. 2010; Woodward, et al. 2007). For example, isolated from p53-null mouse mammary tumors carried by irradiated mice, the CD24⁺/CD29⁺ subpopulations, the cell type enriched for TICs, resolve their γH2Ax DNA damage foci more rapidly than non-TIC subpopulations, suggesting that mouse mammary TICs have a more effective DNA repair system (Zhang et al. 2010). Furthermore, in the human breast cancer cell line MCF7, radiation enriches for breast TIC subpopulations (Phillips et al. 2006; Woodward et al. 2007). Similarly, in two patient-derived TNBC xenograft models (MC1 and BCM-2665A), radiation enriches for the ALDH⁺ subpopulation (Atkinson, et al. 2010). In addition, clinical data show that triple-negative breast tumors, which include the breast TIC-enriched claudin-low subtype, are generally more resistant to radiotherapy as compared to other types of tumors (Prat et al. 2010). Together, these data suggest that breast TICs are more resistant to radiotherapy. However, the resistance of breast TICs to radiation is not without controversy. One study showed the depletion of TICs in one patient-derived xenograft after radiation, as measured by the percentage CD44⁺/CD24^−/low/Lin⁻ cells, ALDH1 levels, and MSFE, but the opposite was observed in a second independent PDX model (Zielske, et al. 2011). This controversy may be due to the use of inappropriate breast TIC markers, or it may indicate that some breast tumors have a radiosensitive TIC population.

Resistance to endocrine therapy

Endocrine therapy drugs, such as tamoxifen and aromatase inhibitors, have been the cornerstone in treating ER⁺ breast cancer and have significantly decreased mortality (2005; Jensen and Jordan 2003; Johnston and Dowsett 2003). However, disease recurrence occurs in up to 25–30% of the patients within 5 years following tamoxifen treatment (2005; Howell and Wardley 2005). For patients with recurrence, de novo and acquired resistance become major challenges for successful treatment (2005). A range of mechanisms have been postulated to account for tamoxifen resistance, from lack of ERα expression (Clarke, et al. 2003) to increased expression of growth factor tyrosine kinases (Dowsett, et al. 2006; Giltnane, et al. 2007; Knowlden, et al. 2003; Massarweh, et al. 2006).

The role of breast TICs in endocrine therapy resistance is controversial. On one hand, even in ER⁺ tumors, breast TICs are predicted to be ER⁻, because CD44⁺ cells are ER⁻ and the expression of ALDH1 is inversely correlated with ER and PR expression (Clarke, et al. 1997; Ginestier et al. 2007; Harrison, et al. 2013; Shipitsin et al. 2007b), suggesting that breast TICs themselves would not respond to endocrine therapy directly. On the other hand, estrogen has been shown to expand the ER⁻ breast TIC population via paracrine FGF/Tbx3 or EGFR/Notch signaling (Fillmore, et al. 2010; Harrison et al. 2013), suggesting that they could be indirectly sensitive to endocrine therapy. However, a recent study provided evidence to support the predicted resistance of ER⁻ cells to endocrine therapy by showing the expansion of the ER⁻PR⁻ subpopulation in ER⁺PR⁺ breast cancers in response to anti-estrogen treatment (Haughian, et al. 2012). This result suggests that the ER⁻ breast TICs are likely to be resistant to endocrine therapy.

Resistance to anti-HER2 therapy

About 15–20% of breast cancer patients have breast tumors of the HER2⁺ subtype, meaning the tumor cells have over expression typically accompanied by HER2 gene amplification (Allred 2010; Slamon, et al. 1987). HER2-targeting drugs, such as trastuzumab, greatly improve the prognosis of patients with HER2⁺ breast cancer. However, 50% of recurrences are due to de novo resistance and, of the patients with HER2⁺ metastatic tumors that do respond to trastuzumab initially, the majority acquire resistance within 1–2 years of treatment (Chung, et al. 2013; Lan, et al. 2005).

Although anti-HER2 agents have been shown to target breast TICs with, or sometimes without, HER2 amplification (Ithimakin, et al. 2013; Korkaya, et al. 2009; Magnifico, et al. 2009), several lines of evidence support a role for breast TICs in trastuzumab resistance: i) HER2 overexpression expands the breast TIC population in vitro and in vivo (Cicalese, et al. 2009; Korkaya, et al. 2008). ii) Signaling pathways that are known to generate trastuzumab resistance also expand the breast TIC population (Chakrabarty, et al. 2013; Ding, et al. 2014; Hanker, et al. 2013; Korkaya, et al. 2012). iii) Long-term trastuzumab treatment of resistant cells enriches the breast TIC population as measured by TIC markers (Korkaya et al. 2012; Reim, et al. 2009). Taken together, these results suggest that combining breast TIC-targeting agents with HER2-targeting agents will benefit patients that do not respond to anti-HER2 therapies alone.

Plasticity of breast TICs and potential therapeutic implications

There is evidence suggesting that breast TICs and non-TICs are inter-convertible, either spontaneously or through induction (Gupta, et al. 2011; Iliopoulos, et al. 2011; Kim, et al.; Meyer, et al. 2009). The apparent plasticity of breast TICs has also been hypothesized to be an obstacle in treating breast cancer. Although facilitating the transformation of breast TICs to non-breast TICs can be an effective therapeutic strategy, the opposing transformation would obviously be problematic (Visvader and Lindeman 2012). Indeed, it has been reported in breast cancer cell lines, that a subset of cells in certain phenotypic states will return toward proportions of phenotypic equilibrium eventually by stochastically transitioning between states following the Markov model, which also predicts that breast TICs and non-TICs are capable of converting to each other (Gupta et al. 2011). In support of this model, it was shown that in multiple breast cancer cell lines, a single CD44⁺/CD24⁺ cell gives rise to the CD44⁺/CD24⁻ progeny in vitro, and that xenograft tumors generated by CD44⁺/CD24⁺ cells are similar to tumors initiated by CD44⁺/CD24⁻ cells (Meyer et al. 2009). Likewise, another study showed that in transformed MCF10A cells, the CD44⁺/CD24^−/low stem-like subpopulation is able to rapidly convert back to other cell types until it reaches an equilibrium proportion of 10% that is similar to those of the parental cell line, and that the non-TICs can give rise to breast TICs in vivo, demonstrating that interconversion between these cell states is possible (Iliopoulos et al. 2011). Furthermore, in transformed human mammary epithelial cells (HMLER cells), CD44^low cells convert spontaneously into CD44⁺ cells in vitro and in vivo (Chaffer, et al. 2011), demonstrating another case of breast TIC plasticity.

The underlying mechanism of such conversions also emerged recently, and was found to be associated with the EMT program. For instance, knocking down of FOXC2 leads to inhibition of mesenchymal phenotype and reduction in tumor-initiating capacity in breast cancer cell lines (Hollier, et al. 2013). Furthermore, it was shown that the ZEB1 promoter undergoes conformational changes in response to the TGFβ signal to drive breast cancer cell plasticity (Chaffer, et al. 2013). Similarly, the conversion of the luminal-like CD44⁺/CD24⁺ cells into basal/mesenchymal-like CD44⁺/CD24⁻ cells depends on the Activin/Nodal initiated TGFβ signaling (Meyer et al. 2009). These studies have been performed predominantly in cell line models, and the degree to which plasticity occurs in vivo in PDX models has not been established.

This “plastic CSC” phenomenon may add additional layers of complexity when treating breast cancer. In particular, if TICs can be generated de novo from non-TICs within a tumor, therapies designed to target TICs may ultimately fail. In that case, targeting both TICs and non-TICs is imperative. Targeting the de-differentiation mechanism, which could be due to gene mutations, epigenetic modifications, or stochastic events, is also promising (Marjanovic, et al. 2013).

Metastatic potential of breast TICs and therapeutic implications

The proposed enhanced metastatic potential of breast TICs may be another treatment hurdle. Because TICs have the ability to generate tumors at orthotropic sites, it has been hypothesized that they could also generate tumors at the metastasis sites (Brabletz, et al. 2005). For example, it has been shown that spontaneous lung metastases and primary breast tumors share similar CD44⁺ profiles that are enriched for breast TICs (Liu, et al. 2010), suggesting that TICs are involved in spontaneous metastasis. Furthermore, ER⁻ metastatic tumors are found in ER⁺ patients, demonstrating the metastasis potential of ER⁻ TICs (Fehm, et al. 2008; Lower, et al. 2005). The epithelial-mesenchymal transition (EMT) program is thought to enable the invasion of tumor cells into the stroma, and therefore may initiate the early steps of the metastatic process (Scheel and Weinberg 2012).

Importantly, the EMT program has been coupled with breast TIC formation. For instance, human mammary epithelial cells (MECs) undergoing EMT also show enhanced TIC properties (Mani, et al. 2008), and the expression of EMT regulators forces non-TICs into the TIC state (Chaffer et al. 2013; Hollier et al. 2013; Li, et al. 2009). Moreover, residual breast cancer cells following chemotherapy display EMT features (Creighton et al. 2009), and knockdown of TWIST reverses chemotherapy-induced multidrug resistance (MDR) and EMT concurrently (Li et al. 2009), indicating the formation of drug-resistant breast TICs during EMT. Additionally, mesenchymal-epithelial transition (MET), a reverse process of EMT, is thought to be essential during the last step of metastasis (Scheel and Weinberg 2012). Contrary to the established pro-EMT signaling involving TGFβ and TWIST, TGFβ and Id1 signaling was reported to induce the mesenchymal-epithelial transition (MET) and stem-like phenotypes in breast cancer cells by targeting TWIST (Stankic, et al. 2013). Together, these data link breast TIC formation to both EMT and MET programs, suggesting the metastatic potential of TICs

Breast TICs were also thought to possess enhanced metastasize capacity. For example, CD44⁺/CD24⁻ human breast cancer stem-like cells not only express EMT markers (Mani et al. 2008), but also display an increased incidence and burden of metastasis when administrated through tail veil injection (Croker, et al. 2009). In addition, ALDH expression is associated with the MET state, and ALDH activity is largely due to the ALDH1A3 isoform that is associated with metastasis (Liu et al. 2014; Marcato, et al. 2011). The metastatic potential of TICs provides insight into the mechanism of cancer progression. Targeting TICs may therefore help eradicate primary tumors and prevent metastasis simultaneously. However, due to the complicated involvement of some common programs in both the MET and the EMT processes, considerable caution is required to pinpoint the different progression stages in each patient.

Pathways that regulate breast TICs and confer resistance

TICs may survive conventional therapy and contribute to recurrence and metastasis later, even in a rapidly shrinking tumor. Therefore, combination of drugs targeting TICs and non-TICs, and combination of parameters measuring tumor volume and TIC functions, is likely required for future clinical studies In order to develop TIC targeting drugs, signaling pathways utilized by breast TICs, as well as the therapeutic agents targeting those pathways, are being intensively studied.

Hedgehog signaling

Hedgehog (Hh) was first identified in a genetic screen for genes required for Drosophila embryonic patterning (Driever and Nusslein-Volhard 1988). In mammals hedgehog signaling functions in multiple tissue/cell types in the developing embryo to direct organogenesis, including the ventral-dorsal pattern formation in the neural tube and anterior-posterior pattern formation in the limb (Ingham and McMahon 2001). Despite the debate on its functional significance in postnatal mammary gland development (Lewis and Visbal 2006), paracrine Hh signaling has been shown to stimulate proliferation and expand the progenitor population in the mouse mammary gland in transgenic mice (Garcia-Zaragoza, et al. 2012; Visbal, et al. 2011). Furthermore, activation of Hh signaling promotes mammosphere formation of normal mammary stem cells, whereas inhibition of Hh signaling by cyclopamine exerts the opposite effects, suggesting a role for Hh signaling in normal mammary stem cells (Liu, et al. 2006; Moraes et al. 2007). Importantly, HH signaling is activated in the Lin⁻/CD44⁺/CD24^−/low human breast TIC subpopulation, and promotes mammosphere formation in a p53-null mouse mammary tumor model through Bmi-1, a polycomb protein overexpressed in the Lin⁻/CD29^H/CD24^H subpopulation. (Liu et al. 2006; Zhang et al. 2008). More recently, it was reported that GLI1, a downstream mediator of HH signaling, stimulates tumor initiation in triple-negative breast tumors (Goel, et al. 2013), and GLI1 is required for breast TIC self-renewal in ER⁺ breast cancer cells (Sun, et al. 2014). These findings imply that HH signaling components could be targets for treating breast TICs.

WNT signaling

Wnt1 was originally identified as a proto-oncogene as it was retrieved from a oncogenic integration site of mouse mammary tumor virus (MMTV) (Nusse and Varmus 1982), and Wnt1 transgenic overexpression generates mammary tumors in mice within 6 months (Tsukamoto, et al. 1988). Wnt signaling is required for normal mammary stem cell function and Wnt-responsive cells show enriched stem cell activity in the mammary gland (Andl, et al. 2002; van Amerongen, et al. 2012). Therefore, Wnt1-induced tumor formation may be due to the ability of Wnt signaling to transform mammary stem cells (Li, et al. 2003; Zeng and Nusse 2010). Wnt-responsive cells are also enriched in TICs of mouse mammary tumors (Zhang et al. 2010), which are located close to distorted blood vessels (Vadakkan, et al. 2014; Zhu, et al. 2013), supporting a role for WNT signaling in breast TICs. Accordingly, a WNT inhibitor decreases the TIC frequency of a RAS-transformed mesenchymal subpopulation of human MECs (Scheel, et al. 2011). Importantly, WNT signaling plays a pivotal role in the radioresistance of breast TICs. For example, TICs isolated from p53-null mouse mammary tumors have enriched Wnt signaling activity and more effective DNA damage repair systems (Zhang et al. 2010). In addition, clinically relevant doses of radiation specifically enrich the SP and the Sca1⁺ stem-like subpopulations in MECs with activated Wnt signaling, but not in wild type MECs (Woodward et al. 2007). Finally, radiation enriches for Sca1⁺ cells, and these cells show elevated Wnt signaling activity and fewer γH2Ax DNA damage foci as compared to Sca1⁻ cells (Chen, et al. 2007). Together, these suggest that WNT signaling inhibitors have the potential to sensitize the resistant breast TICs to radiotherapy. Recent studies by Gunther and colleagues reported that both the luminal and basal populations were required for efficient tumor formation in the MMTV-driven Wnt1 genetically engineered mouse model, and was dependent on luminal Wnt 1 expression (Cleary, et al. 2014)

WNT/β-catenin signaling is activated by the binding of a WNT ligand to its receptor Frizzled and co-receptor LRP5/6 (lipoprotein receptor-related protein5/6), making Frizzled and LRP5/6 candidate targets for inhibiting WNT signaling (MacDonald, et al. 2009). Salinomycin inhibits WNT/β-catenin signaling by inducing LRP6 degradation, and has also been identified as a selective inhibitor of breast TICs in a high throughput screen (Gupta, et al. 2009; Lu, et al. 2011; Lu and Li 2014). A number of natural dietary components were also found to inhibit breast TICs and downregulate the WNT pathway, including curcumin, piperine, and sulforaphane (Kakarala, et al. 2010; Park, et al. 2010). Despite this progress, the development of potent WNT inhibitors remains challenging (Anastas and Moon 2013). Interestingly, pSTAT3 binds to the promoter region of β-catenin directly and activates β-catenin transcription in MCF7 and BT474 breast cancer cell lines, and pSTAT3 and β-Catenin staining is significantly correlated in primary breast tumors (Armanious, et al. 2010). Investigating the crosstalk between STAT3 and WNT signaling may potentially provide novel intervention targets to inhibit WNT signaling.

NOTCH signaling

In vertebrates, there are four transmembrane Notch receptor proteins Notch1-4. Notch4 is a proto-oncogene that is capable of inducing mouse mammary tumors when abnormally expressed by MMTV insertion (Gallahan and Callahan 1987, 1997). Notch pathway activators promote, whereas Notch pathway inhibitors suppress, the formation of secondary mammospheres from normal mammary stem cells, suggesting that Notch signaling plays an important role in normal stem/progenitor cell function (Dontu, et al. 2004). Moreover, both NOTCH1 and NOTCH4 antibodies decrease the MSFE of breast tumor cells derived from PDX or patient samples, suggesting their function in breast TICs (Farnie, et al. 2007; Qiu, et al. 2013). Importantly, NOTCH4 signaling activity is 8-fold higher in the ESA⁺/CD44⁺/CD24^low TIC subpopulation, and NOTCH4 inhibition reduces the TIC frequency in human breast cancer cell lines in vivo (Harrison, et al. 2010). More recently, it was found that in basal-like breast cancer, NF-κB induces the expression of JAG1, a Notch ligand, in non-TICs, which stimulates NOTCH signaling in TICs in trans, leading to breast TIC expansion (Yamamoto, et al. 2013). Another study showed that NOTCH signaling is activated in ER⁻ cells, whereas it is inhibited in ER⁺ cells by estradiol, making NOTCH signaling a potential target to overcome endocrine therapy resistance in ER⁻ cells, the putative breast TIC population (Rizzo, et al. 2008).

NOTCH signaling is activated by γ-secretase cleavage of its intracellular domain. Therefore, γ-secretase inhibitors (GSIs) have been used as pan-NOTCH signaling inhibitors (Bray 2006). For example, the GSI inhibitor MRK-003 was shown to successfully eliminate TICs in a mouse model of ErbB2 breast cancer, as measured by LDT assays (Kondratyev, et al. 2012). More recent and promising data suggesting that the MRK-003 reduces breast TICs in patient-derived xenograft tumors, and that MK-0752, a MRK-003 analog, may reduce breast TICs in patient biopsies in a Phase I clinical trial, as measured by MSFE and the expression levels of stem cell markers (Schott, et al. 2013). However, it has also been shown that NOTCH4 knockdown is more efficient than GSIs in targeting TICs, and this is likely because NOTCH4 activity is increased and NOTCH1 activity is decreased in breast TICs (Harrison et al. 2010). Interestingly, a recent study showed that in some triple-negative breast cancer lines, the activated NOTCH1-ICD is only increased in the CD44⁺/CD24^low breast TICs, but not in the CD44⁺/CD24^neg breast TICs. Therefore, the CD44⁺/CD24^low stem-like population is sensitive to GSIs, but CD44⁺/CD24^neg subpopulation, which is also stem-like, is resistant (Azzam, et al. 2013). These data suggest the heterogeneous nature of NOTCH signaling within breast TICs, and indicate that, similar to other proposed breast cancer therapies, GSIs may need to be combined with other treatments to maximize their efficiency.

IL-6/STAT3 signaling

STAT3 belongs to a family of latent transcription factors (Chatterjee-Kishore, et al. 2000), and it plays a pivotal role in early embryogenesis, acting to keep ESCs in an undifferentiated state (Matsuda, et al. 1999; Raz, et al. 1999; Torres and Watt 2008). Recent evidence suggests it also plays a role in regulating stem cells in solid tumors, including breast tumors. For example, in breast cancer models, treatment with the STAT3 pathway agonist IL-6 expands the CD44^high/CD24^low stem-like subpopulation (Iliopoulos et al. 2011), whereas shRNA-mediated STAT3 knockdown decreases the TIC frequency (Zhou et al. 2007). Additionally, in patient breast tumor samples, the majority of CD44^high/CD24^low cells are positive for pSTAT3 staining (Marotta et al. 2011). A more recent report linked high STAT3 activity to high autophagy function that contributes to chemoresistance in breast cancers (Maycotte, et al. 2014). Taken together, these data suggest that STAT3 is associated with breast TIC function.

Recently, our laboratory showed that STAT3 signaling is activated preferentially in TICs of claudin-low type human breast cancer, further suggesting that STAT3 signaling regulates breast TIC function (Wei et al. 2014). Consistent with these data, a recent study showed that the anti-malarial drug chloroquine inhibits Jak2/STAT3 signaling, decreases the MSFE in a number of TN breast cancer cell lines, and decreases the TIC frequency in MDA231 xenograft tumors (Choi, et al. 2014). Importantly, it has also been shown that breast TIC-derived IL-6 recruits mesenchymal stem cells, which communicate with breast TICs by secreting additional cytokines and chemokines, creating a breast TIC-promoting microenvironment (Liu, et al. 2011). Furthermore, an inflammatory feedback loop has been shown to enhance breast TIC function, in which breast cancer cells under long term trastuzumab treatment are highly enriched in the CD44⁺/CD24⁻ population and secrete high levels of IL-6, which in turn acts through the IL-6 receptor to expand the CD44⁺/CD24⁻ stem-like population (Korkaya et al. 2012).

STAT3 signaling is typically induced by the binding of IL-6-type cytokines to gp130 receptors, which activates JAK kinases (Yu and Jove 2004). Therefore, STAT3 signaling can be inhibited by JAK inhibitors and small molecules that directly target STAT3. The former is exemplified by ruxolitinib, which is currently being tested in phase II trials to treat breast cancer (Quintas-Cardama and Verstovsek 2013). The latter is exemplified by piperlongumine, which was shown to block mammosphere formation of patient-derived xenograft tumor cells and inhibit breast tumor growth in vivo (Bharadwaj, et al. 2014). Other in vivo used STAT3 inhibitors includes Cmp188, the, competitive small molecule inhibitor compete with STAT3 to bind pY-peptide (Xu, et al. 2009). In addition, chloroquine, an anti-malarial drug, was reported to inhibit JAK/STAT3 signaling and reduce the CD44⁺/CD24^low/− subpopulation in triple-negative breast tumors (Choi et al. 2014).

Other methods to inhibit STAT3 signaling include blocking the IL-6 receptor. For example, combined treatment using an anti-IL-6 receptor antibody and trastuzumab reduces the TIC frequency in trastuzumab-resistant HER2⁺ breast cancer cells (Korkaya et al. 2012). Together, these data suggest that STAT3 inhibitors would be promising therapeutic drug candidates that target both breast TICs and their niche.

IL-8/CXCR1/2 signaling

CXCR1 is a member of the G-protein-coupled receptor family, and it is highly expressed in ALDH⁺ cells in a number of breast cancer cell lines (Charafe-Jauffret et al. 2009). Treatment with the CXCR ligand IL-8 increases the ALDEFLUOR⁺ populations and the MSFE in breast cancer cell lines (Charafe-Jauffret et al. 2009; Singh, et al. 2013), whereas treatment with a CXCR1 inhibitor depletes the breast TIC subpopulation (Ginestier, et al. 2010). IL-8 is also secreted by mesenchymal stem cells in response to signals from breast TICs and reinforces breast TIC function (Liu et al. 2011). More recently, it was reported that IL-8/CXCR1/2-induced mammosphere formation is mediated partially by EGFR/HER2 signaling (Singh et al. 2013), which provides a rationale for using a combined therapeutic strategy. Indeed, the small molecule CXCR1 inhibitor repertaxin was shown not only to effectively target breast TICs, but also to enhance the efficacy of lapatinib in treating HER⁺ breast cancers (Singh et al. 2013).

IL-8 is the most well-studied CXCR1/2 ligand, and antibodies against IL-8 are being tested in clinical trials to treat inflammatory diseases (Skov, et al. 2008). However, the benefits of inhibiting IL-8 may be limited, because a broad range of other ligands can also stimulate CXCR1/2 (Bieche, et al. 2007). Therefore, blocking CXCR1/2 function provides much more specificity. For instance, repertaxin shows potent inhibition of breast TICs in vitro, and reduces tumor growth and metastasis in vivo (Ginestier et al. 2010). It also has an additive effect when used in combination with lapatinib to treat HER⁺ breast cancers (Singh et al. 2013), and has undergone clinical testing in combination with chemotherapy to treat patients with HER2⁻ metastatic breast cancer (AF Schott 2012).

TGFβ signaling

TGFβ family members are multifunctional peptides that regulate cell growth and differentiation (Heldin, et al. 1997). TGFβ signaling has also been shown to maintain the undifferentiated state of human ESCs (James, et al. 2005). The role of TGFβ signaling in mammary carcinogenesis is controversial, because it is known to impair tumorigenesis, but it also triggers the EMT/MET program to facilitate metastasis (Muraoka-Cook, et al. 2005; Siegel, et al. 2003; Stankic et al. 2013; Tang, et al. 2007). Importantly, TGFβ and its receptor are specifically expressed in CD44⁺ cancer cells, and their expression results in a more mesenchymal appearance of those cells (Shipitsin, et al. 2007a). Together, these data link TGFβ signaling to a mesenchymal/stem-like state of breast cancer cells. Surprisingly, a recent finding suggests that TGFβ/Id1 signaling also induces MET and stem-like phenotypes in breast cancer cells, and this only occurs in cells that have already undergone an EMT. Hence, TGFβ is predicted to facilitate metastasis during both the EMT and MET steps, in a sequential manner (Stankic et al. 2013). More recently, an oscillating TGFβR3-JUND signaling circuit was uncovered in basal-like breast cancer cells, suggesting an additional underlying mechanism for intratumoral heterogeneity (Wang, et al. 2014a). All of these findings highlight a context-dependent role of TGFβ in breast TIC plasticity and metastasis.

Several examples illustrate the importance of TGFβ signaling in chemotherapy-induced breast TIC enrichment and metastasis. In the 4T1 mouse mammary tumor model, the chemotherapy drug doxorubicin induces EMT and promotes a stem-like phenotype by activating TGFβ signaling, which is effectively blocked by a TGFβ type I receptor kinase inhibitor (TβRI-KI). Combining doxorubicin with TβRI-KI significantly reduces tumor growth and lung metastasis in vivo (Bandyopadhyay, et al. 2010). Similarly, the chemotherapy drug paclitaxel enriched for breast TICs in triple-negative breast cancers as indicated by MSFE assay and TIC markers (Bhola, et al. 2013), and this effect is prevented by treating with LY2157299, a small molecule TβRI-KI (Bhola et al. 2013). However, considering the multiple functions of TGFβ in plasticity and metastasis of breast TICs (Stankic et al. 2013; Wang et al. 2014a), additional caution is required when targeting this signaling pathway.

Integrin signaling

Integrins are environmental sensors, and importantly, they are differentially expressed between breast TICs and non-TICs. For example, the β3 subunit integrin CD61 was used to enrich a breast TIC subpopulation in MMTV-Wnt1 mouse mammary tumors since the CD29^lo/CD24⁺/CD61⁺ subset shows increased TIC frequency than the CD29^lo/CD24⁺/CD61⁻ subset (Vaillant et al. 2008). Furthermore the α6 integrin subunit CD49f is a biomarker for TICs in ER⁻ and triple negative breast cancer xenografts (Lee et al. 2014; Meyer et al. 2010). Recently, an autocrine loop between VEGF and integrin signaling has been uncovered in breast TICs. In this loop, VEGF signaling activates α6β1(CD49fCD29) integrin, which enhances the expression of Gli, leading to the transcription of the VEGF receptor neuropilin-2 (NRP2) (Goel et al. 2013). Another study demonstrated that VEGF signaling transcriptionally regulates an RNA-splicing factor that regulates the splicing of α6Bβ1 (CD49fBCD29) integrin. This splice form of α6β1 (CD49fCD29) integrin defines a mesenchymal-like subpopulation within the CD44^high/CD24^low cells, and it is critical for breast TIC function (Goel, et al. 2014). Also recently, the integrin α(v)β3 (α(v)CD61) and the KRAS/NFKB signaling pathway were shown to drive resistance to EGFR inhibition in breast TICs (Seguin, et al. 2014), suggesting a function of integrin proteins in breast TICs

In addition, focal adhesion kinase (FAK) is a major mediator of Integrin signaling. Targeted deletion of Fak in mouse mammary epithelium decreases the content of ALDH⁺ stem-like populations in primary MMTV-PyMT mammary tumors, and decreases the tumorigenicity of ALDH⁺ cells in vivo (Luo, et al. 2009). Functional studies show that integrin signaling actively maintains the mammary cancer stem/progenitor population (Luo et al. 2009). Further studies suggest distinct roles of FAK in mammary stem cells and progenitors, and that it plays correspondingly different roles in claudin-low and luminal-like type breast cancer cells, suggesting a potential relationship between normal stem cells/progenitors and different subtypes of breast cancers (Luo, et al. 2013). Together, these recent discoveries of detailed mechanisms involving integrin signaling in breast TICs provide novel intervention opportunities.

Treatment with anti-β1 integrin antibodies enhances the response of human xenograft breast tumors to radiotherapies (Park, et al. 2008), and normalizes structures of breast cancer cell lines in 3D cultures (Park, et al. 2006) In addition, the recent finding that integrin signaling in breast TICs drives resistance to EGFR inhibitors, suggests potential novel strategies to sensitize breast tumors to receptor tyrosine kinase inhibition (Seguin et al. 2014). Cilengitide, an integrin antagonist currently being tested in clinical trials (Goodman and Picard 2012), reduces bone metastasis in a mouse mammary tumor model (Bauerle, et al. 2011). A more recent study also reported that the increased stiffness of the extracellular matrix in tumors can activate integrin signaling to drive tumor progression, suggesting another new intervention method (Mouw, et al. 2014).

microRNAs

microRNAs (miRNA) are short noncoding RNAs that bind to complementary sequences in mRNAs, resulting in suppression of translation and/or degradation of targeted mRNAs (Bartel 2009). Importantly, over 30 miRNAs have been reported as differentially expressed between breast TICs and non-TICs (Shimono, et al. 2009), suggesting miRNAs may function to regulate breast TICs. Indeed, let-7 family members are downregulated in mammospheres formed from primary breast tumor cells as compared to bulk tumor cells, and forced expression of let-7a reduces MSFE (Yu et al. 2007). miR-200 family members are also downregulated in CD44⁺/CD24⁻ cells, and forced expression of miR-200c reduces tumor initiating capacity (Shimono et al. 2009). Interestingly, miR-93 is highly expressed in non-TICs in a claudin-low type breast cancer, but it is not differentially expressed between TICs and non-TICs in luminal type breast cancers. Forced expression of miR-93 decreases the TIC frequency in claudin-low type tumors, whereas it increases the TIC frequency in luminal type tumors, suggesting that the same miRNA can have distinct roles in regulating TICs in different types of breast cancer (Liu, et al. 2012). Moreover, Lin28, a negative regulator of let-7, is a direct target of the WNT-β-catenin pathway. Loss of function of Lin28 reverses WNT-mediated let-7 inhibition and breast TIC expansion (Cai, et al. 2013). These data suggest that miRNAs can be mediators of WNT-regulated breast TIC function, and as such, are promising targets for inhibiting breast TIC expansion. Taken together, these data suggest that miRNAs could be useful agents to target in treating breast cancer.

Inhibition of miRNAs can be achieved by intravenous injection of antagomirs, which are modified antisense oligos that are complementary to the miRNA sequence (Krutzfeldt, et al. 2005). In one example, anti-let-7 oligos were introduced into breast cancer cells, resulting in the enhanced self-renewal of non-TICs (Yu et al. 2007). Alternatively, synthesized artificial miRNAs can be introduced that mimic the effect of a miRNA. For instance, let-7 mimics have been administered to mouse models of lung cancer, leading to reduced tumor growth (Esquela-Kerscher, et al. 2008). However, targeting breast TICs by manipulation of miRNAs remains challenging. One problem is that each miRNA potentially regulates hundreds of genes simultaneously, which leads to poor specificity. Also, cancer cells tend to produce mRNAs with shorter 3′ UTRs, which often contain the miRNA target sites, making them resistant to miRNA mimics (Mayr and Bartel 2009).

Panels of inhibitors/activators have been developed for each of the signaling pathways mentioned above. To evaluate their proposed effects on TICs, MSFE and LDT assays should be applied to measure their efficiency. We provided a summary of activatory and inhibitory agents targeting those signaling pathways in Table 3, emphasizing their effectiveness on TIC function.

Table 3.

Summary of pathway specific inhibitors/activators which show effectiveness on TIC function.

Pathways	Samples (subtypes and/or manipulations)	Inhibitors	Activators	Effective Dose	MSFE	TIC frequency	References
HH	SUM1315 (NPR2high/shNPR2)		BMI-1		increase
	SUM1315, patient samples		SHH		increase		Goel et al. 2013
	SUM1315 (NPR2low/RAS)	shGLI			decrease
	MCF7, HCC1428	shGLI			decrease		Sun et al. 2014
WNT	HMLE (shEcad), MCF7(RAS); 4T1	salinomycin		1 nM, 8 μM, 8 μM	decrease		Gupta et al. 2009
	HMLE (shEcad), MCF7(RAS); 4T1	salinomycin		5 mg/kg		decrease
	MCF7	curcumin		5 μM	decrease		Kakarala et al. 2010
	MCF7	piperine		10 μM	decrease
	HMLE (MSP) or HMLE (MSP/Twist)	DKK1		0.5 μg/ml	decrease
	HMLE (MSP) or HMLE (MSP/Twist)	SFRP1		1 μg/ml	decrease		Scheel et al. 2011
	HMLE (MSP/RAS)	SFRP1		1 μg/ml	decrease	decrease
	MDA231	shβ-catenin			decrease		Liu et al. 2012
NOTCH	patient samples (DCIS)	Notch4 antibody			decrease		Farnie et al. 2007
	MCF7, MDA231, BT474, patient samples	DAPT		10 μM	decrease
	MCF7	shNotch1			decrease	decrease	Harrison et al. 2010
	MCF7	shNotch4			decrease	decrease
	MMTV-ErbB2 tumor	MRK-003		0.125 μM		decrease	Kondratyev et al. 2012
	MMTV-ErbB2 tumor cells	MRK-003		0.01 μM	decrease
	HCC1973		JAG1		increase		Yamamoto et al. 2013
	PDX	MRK-003		100 mg/kg	decrease
	PDX	MRK-003		100 mg/kg		decrease	Schott et al. 2013
	Phase I patients	MK-0752		300 mg – 800 mg	decrease
	PDX, SUM149 xenograft	Notch1 antibody			decrease		Qiu et al. 2013
IL-6/STAT3	MCF7	shSTAT3				decrease	Zhou et al. 2007
	MCF7	IS3 295		10 μM		decrease
	Hs578t, MDA231, SUM159, HCC1937	chloroquine		1 μM	decrease		Choi et al. 2014
	MDA231 xenograft	chloroquine		10 mg/kg		decrease
	PDX cells	piperlongumine		5.8 μM, 0.1 μM	decrease		Bharadwj et al. 2014
	MDA231/SUM159 xenograft cells (STAT3-GFP+)	Stattic		5 μM	decrease		Wei et al. 2014
IL-8/CXCR1/2	HCC1954, SUM159, MDA453		IL-8	50–100 ng/ml	increase		charafe-Jauffret et al. 2009
	SUM159, HCC1954	CXCR1 antibody		10 μg/mL	decrease
	SUM159, HCC1954	repertaxin		100 nM	decrease		Ginestier et al. 2010
	PDX	repertaxin		15 mg/kg		decrease
	patient samples		IL-8	100 ng/mL	increase		Singh et al. 2013
TGFβ	SUM159		TGFβ1	2.5 ng/ml	increase
	BT549	SMAD4 siRNA			decrease		Bhola et al. 2013
	SUM159 xenograft	LY2157299		100 mg/kg		decrease
Integrin	MMTV-PyVT (ALDH+)	FAK KO			decrease		Luo et al. 2009
	SUM1315 (shα6), MDA231 (shα6), MDA435 (shα6)		α6B		increase	increase	Goel et al. 2014
	MDA231	α6B TALENs			decrease
microRNA	SKBR3 xenograft		let-7		decrease	decrease	Yu et al. 2007
	PDX cells (CD44+CD24−/low)		miR-200c			decrease	Shimono et al. 2009
	SUM159		miR-93			decrease	Liu et al. 2012
	MCF7		miR-93			increase

Abbreviations: Ecad, E-cadherin; MSP, Mesenchymal Subpopulation; MDA231, MDA-MB-231; DCIS, Ductal Carcinoma In Situ; PDX, Patient-Derived Xenograft.

Summary

The hypothesized existence of resistant breast TICs provides a potential mechanism to explain treatment failure, recurrence, and metastasis in treatment of breast cancer using current therapeutic methods. Therefore, inhibiting the pathways that are essential for breast TIC function is a promising strategy to overcome drug resistance. Furthermore, combining conventional therapies with breast TIC-targeting drugs should improve long-term patient outcome by eliminating TICs that potentiates future recurrence and metastasis. Moreover, effective therapeutic strategies should address concerns regarding the dynamic properties of breast TICs that could become problematic during breast cancer treatment, such as their plasticity and enhanced metastatic potential. Finally, while simply targeting one major signaling pathway is unlikely to eliminate breast TICs completely, sequential or simultaneous administration of inhibitors targeting multiple signaling pathways may render long-lasting and synergistic effects on TICs. Alternatively, given our success in eliminating non-TICs, inducing the differentiation of TICs to other cell types by manipulating essential signaling pathways, followed by conventional therapies, could be a safe and effective strategy. Nonetheless, genome instability in stem-like cells predicts that the TICs are an evolving entity that quickly adapts to alternative survival mechanisms. We will have to continue to build on our knowledge about breast TICs by revealing novel mechanisms and alternative signaling pathways that are essential for their maintenance and function.