The seminal study by Al-Hajj et al. [4] showed that CD44⁺/CD24^−/low/lin⁻ (lack of expression of CD2, CD3, CD10, CD16, CD18, CD31, CD64 and CD140b) breast cancer cells obtained mainly from pleural effusions could form tumors when as few as 200 were injected into the mammary fat pad of nonobese diabetic (NOD)/severe combined immunodeficient (SCID) mice. In patients with breast cancer, the presence of CD44⁺/CD24^−/low cells did not provide clinically relevant prognostic information notwithstanding the finding that the fraction of CD44⁺/CD24^−/low cells was higher in tumors developing distant metastasis than in those that did not [10] and was detected in 50/50 bone marrow specimens with cytokeratin-positive cells, that is, in the so-called disseminated tumor cells [8]. Moreover, compared to normal breast epithelium, they showed a signature accounting for 186 differentially expressed genes (invasiveness gene signature, IGS) mainly involved in the IκB/NFκB (nuclear factor of kappa light polypeptide gene enhancer in B-cells 1) and RAS/MAPK (mitogen-activated protein kinase) pathways and in the epigenetic control of gene expression. This signature was predictive of clinical outcome not only in breast cancer, but also in lung and prostate cancers and in medulloblastoma [7]. Among cell surface markers, also the expression of CD133, a putative stem cell marker for tumor types of different origin, has been reported to identify breast cancer cell subpopulations endowed with tumor-initiating capability [11].

A recently proposed and currently very popular approach used to identify BCIC is based on enzymatic activity of aldehyde dehydrogenase 1 (ALDH1), a detoxifying enzyme responsible for metabolization of aldehydes and oxidization of retinol to retinoic acid and related to the stemness-related markers Oct-4 and BMI-1 and proven to label stem/progenitor cells in neural and hematopoietic systems and in the mammary gland [12]. Use of the ALDEFLUOR approach (by flow cytometry) and/or the detection of ALDH1-positive cells at the cytoplasmic level by immunohistochemistry identified highly tumorigenic cells not only within breast cancer specimens but also in brain tumors, leukemia and multiple myeloma. In breast cancer, the fraction of ALDH1-positive cells correlates with the molecular subtypes, being highly expressed in the basal-like and HER2 subtypes, but not with other patho-biologic features [13] and is predictive of clinical outcome in breast cancer series in different clinical settings [12,14], even if such a finding is not univocal [13].

The capability to exclude vital dyes, such as Hoechst 33342 or rhodamine-123, is characteristic of cells overexpressing transmembrane transporters, like the ATP-binding cassette (ABC) molecule ABCG2/BCRP (ATP-binding cassette, sub-family G/breast cancer resistance protein-1), and might identify cells accounting for the so-called side population. In breast epithelium, side population cells are limited in number (0.5–3%) and are endowed with stem cell properties [15]. Moreover, side population cells are also reported to exist in breast cancer and mainly in human breast cancer cell lines [16]. Notwithstanding the proper identification of the side population still represents a controversial issue, the “side population” cancer cells proved to be resistant to toxins and drugs used in anticancer treatment and to exhibit a higher tumor-initiating capability than non-side population cells, and have been suggested to be enriched in CSC [17].

BCIC have also been isolated and in vitro propagated for some passages as non-adherent spheres of undifferentiated cells (mammospheres) in medium without serum and supplemented with growth factors. The culture of breast cancer cells as mammospheres is a relatively new technique based on methods used for growing neurospheres and recently applied by Dontu et al. [5,18] also to mammary epithelial cells to enrich mammary stem cells. We adopted such an approach to propagate in vitro as non-adherent mammospheres breast tumorigenic cells with stem/progenitor cell-like properties isolated from clinical tumors as well as from established cancer cell lines (MCF-7 and more recently, 734B and estrogen receptor (ER)-positive and ER-negative human breast xenografts transplanted in nude mice). Mammosphere-derived breast cancer cells display self-renewal properties, overexpress stemness (Oct4 and BMI1) and cytoprotective markers (the IAP survivin protein) and telomerase activity, and are highly tumorigenic when injected at low concentrations in immunocompromized mice [19].

The possibility to propagate in vitro BCIC using a functional approach, such as the sphere-forming assay, might represent a potentially valid and convenient tool to enrich cell cultures for stem cell-like cancer cells, in order to molecularly characterize and challenge them against conventional, investigational and novel therapies targeting pathways biologically relevant for CSC. However, the yield of BCIC from primary tumors is highly variable and often contaminated by more differentiated cells, mainly when the number of BCIC is modest, and culture conditions may be suboptimal and not adequate to support their propagation, since after a few in vitro passages, induction of differentiation and/or senescence results in loss of stemness.

Side population, expression of surface markers, ALDH1 activity and capability to form non-adherent spheres have also been recognized in cancer cell lines that have been established from different solid tumor types [17]. Stem cell-like cancer cells isolated by distinct approaches from established breast cancer cell lines have been considered as surrogate models of cultures of tumor-initiating cells derived from clinical tumors (which are difficult to obtain and in vitro successfully propagate for high-throughput screening of lead compounds or drug development, which in turn requires a large number of stem-like cells) or of direct xenografts of surgical tumor specimens in immunocompromized animals (which are successful in less than 20% of breast cancers). However, much effort should be made to assess the reliability of the system after several in vitro passages, as well as its independence of experimental conditions and its genetic stability. In fact, in human breast cancer cell lines, the propagation of non-adherent spheres, which are endowed with high tumorigenicity, rather than adherent cells, may result in a more rearranged karyotype, and display additional and more complex chromosomal rearrangements [20], which puts into question their use for high-throughput screening.

Our experience with breast cancer cells obtained from established cell lines, clinical tumors and pleural effusions, and xenografts of human tumors, indicates that a common, distinctive feature of stem cell-like breast tumor-initiating cells is currently not available. Indeed, the CD44⁺/CD24^−/low status commonly used to identify tumor-initiating breast cancer cells is present in a substantial fraction of cells derived from pleural effusions or from ER-negative breast cancer cell lines and xenografts, whereas it is detectable in a lower fraction of cells from ER-positive tumors, even growing as non-adherent mammospheres. Conversely, other putative stemness markers, such as the presence of a side population expressing the Hoechst 33342 dye or ALDH1, were singly detectable in breast cancer cell lines or xenografts that rarely produce mammospheres [21].

Initially implicated in mammary cancer development and maintenance of the stem cell/progenitor pool in the mouse mammary gland, a Wnt cross-talk with steroid receptor pathways has been recently shown, and Wnt signaling down-regulation by Wnt antagonists or small interfering RNAs (siRNA) proved to increase the expression of differentiated markers in human breast cell lines [31]. This suggests that the molecular targeting of Wnt signaling could represent a potential therapeutic approach. In clinical breast cancer, contrasting results have been reported for the association between Wnt5a expression and prognosis [32–34], whereas intracellular β-catenin expression localization [35] is differently associated with clinical outcome (favorable for cytoplasmic, unfavorable for nuclear), and promoter methylation of the secreted frizzled-related protein 1 appears as an independent factor for adverse patient survival [36,37].

The Oct-4 transcription factor plays a pivotal role as a key regulator of pluripotency in the earliest stages of mammalian development. Because of its ability to maintain the stem cell state and thus to be refractory to xenobiotic agents, it was hypothesized that Oct-4 aberrant expression may contribute to the activation and maintenance of neoplastic process in normal and tumor cells [19,38]. A possible role for Oct-4 in tumorigenesis is supported by several additional lines of evidence and it is therefore desirable to develop experimental molecules capable of selectively targeting this pathway.

Nanog is a newly identified homeodomain-bearing transcriptional factor and it represents a key molecule involved in the signaling pathway for maintaining the capacity for self-renewal and pluripotency. Several studies have suggested that the molecular stem-maintenance circuitry controlled by Nanog may be active in breast tumors. Thus it is one of the molecular markers suitable for recognizing and targeting the undifferentiated state of cells in malignant human tissue [38].

Hedgehog cascade has been implicated in the development of several cancers including breast cancers. Sonic hedgehog contributes to a molecular signature segregating inflammatory breast cancers at different prognosis [39], whereas exposure to cyclopamine, an alkaloid blocking hedgehog pathway, suppresses Gli1 expression and breast cancer cell growth [40].

BMI-1 is a component of the Polycomb complex induced through the hedgehog signaling pathway, which is responsible for self renewal of normal and leukemic stem cells, represses genes inducing cell senescence and death, and immortalizes human mammary epithelial cells. BMI-1 is overexpressed in breast cancer cells growing as non-adherent mammospheres [41], and a BMI-1-based 11-gene signature is a powerful, therapy-independent predictor of recurrence, distant metastasis and death in 11 epithelial and non-epithelial cancers [6]. Such findings suggest that the targeting of BMI-I could also constitute a potential therapeutic strategy.

Notch cascade is known to play a role in several cancers [42,43], and Notch 1 and 4 are involved in tumor induction in mouse and in increased proliferation in mammospheres [44,45]. Notch 1 and 2 are directly and inversely associated to an unfavorable outcome in breast cancer patients, respectively [44–47].

HER-2, a member of the EGFR family whose amplification is present in about 25% of human breast cancers, correlates with a distinct molecular profile and unfavorable outcome and has been shown to regulate the mammary stem/progenitor cell population driving tumorigenesis and invasion [25]. In fact, in normal mammary epithelial cells, HER2 overexpression increases the proportion of stem/progenitor cells (in vitro mammosphere assay), the expression of the stem cell marker ALDH1, and the generation of hyperplastic lesions in the humanized mammary fat pad of NOD/SCID mice. Similarly, in breast carcinoma cell lines, HER2 overexpression increases the ALDH-expressing BCIC population, which displays enhanced expression of stem cell regulatory genes and exhibits increased invasion in vitro and increased tumorigenicity in the humanized mammary fat pad of NOD/SCID mice [48].

Epithelial to mesenchymal transition is an essential process that enables reprogramming of polarized epithelial cells towards a mesenchymal motile phenotype during physiological remodeling in normal adult tissues. In fact, the typically dormant EMT program is reactivated during wound repair and tissue regeneration. Recent studies have demonstrated that EMT is associated with the expression of many stem cell markers and phenotypes in human mammary epithelial cells in parallel with loss of E-cadherin expression, a key component of adherent junctions [49]. These studies suggest that induction of EMT by extracellular stimuli and microenvironment factors could confer both self-renewal and metastatic properties to breast cancer cells resulting in de novo generation of cancer stem cells population from differentiated tumor cells, thus making this cellular mechanism an alternative driving force in tumor progression. Overall, the findings that link EMT and cancer stem cell hypothesis indicate that targeting both pathways simultaneously may hold considerable therapeutic promise [29,49].

The resistance of tumor-initiating cells to anticancer drugs has been hypothesized based on their unique properties to reduce drug uptake and effect because of repair and anti-immune mechanisms, which are in keeping with inherent resistance to a variety of conventional agents. Such findings have been shown in different in vitro experimental systems and indirectly by the retrospective observation that residual post-treatment tumors are enriched for cells with tumor-initiating capability [30]. Importantly, the intrinsic resistance of BCIC to chemotherapy has been recently directly demonstrated also on a series of 31 primary breast cancers submitted to neoadjuvant chemotherapy with docetaxel or doxorubicin/cyclophosphamide [55]. Chemotherapy treatment increases the fraction of CD44⁺/CD24^low/− and the mammosphere formation efficiency in residual tumor cells after treatment, thus providing strong evidence from a clinical setting for a subpopulation of chemotherapy-resistant BCIC. Conversely, in a parallel series of HER2⁺ tumors from 21 patients treated with lapatinib, the residual tumor cell population following neoadjuvant treatment did not show an increase in tumor cells with putative stem cell features, thus providing support for the hypothesis of breast cancer stem cells, since HER2 is a BCIC driver and the evidence that specific inhibitors of signaling pathways involved in self renewal (EGFR/HER2) may provide a therapeutic strategy for eliminating BCIC [25].

Another analysis carried out on a series of 108 primary breast cancer patients treated with neoadjuvant therapy consisting of sequential paclitaxel and epirubicin-based regimens has not indicated an association between increased fraction of CD44⁺/CD24^low/− cells and chemotherapy resistance but showed ALDH1 overexpression in non-responding tumors as well as in the residual tumor cell population [54]. Other studies challenged the role of a stem-cell phenotype (as defined by the fraction of cells CD44⁺/CD24⁻ or expressing ALDH1) in tumor resistance to chemotherapy. In fact, in a series of 66 breast cancers subjected to primary systemic treatment with regimens including epirubicin/cyclophosphamide or doxorubicin/pemetrexed, the post-treatment residual tumor cell population was not enriched by CD44⁺/CD24⁻ cells [57]. Similarly, a recent investigation on a substantial series of neoadjuvantly treated cases (with paclitaxel and epirubicin-based regimens plus herceptin if HER-positive tumors) and triple-negative breast cancers did not confirm the association between ALDH1 expression in tumor cells and clinical outcome, whereas expression of ALDH1 in stromal cells was increased following neoadjuvant treatment and was significantly associated with disease-free and overall survival in the triple-negative cases. These observations suggest a possible role of the tumor microenvironment in affecting the prognostic relevance of putative stemness markers [13].

We also analyzed the expression of ALDH1 and CD44⁺/CD24^low/− in a series of 20 matched primary tumors/metachronous metastasis (mainly from soft tissues). The analysis showed an increased frequency of CD44⁺/CD24^low/− cases from the primary to the metastatic lesion [58].

The non-univocal results in these preliminary translational studies emphasize the importance of an experimental design requiring a large number of clinical specimens (possibly derived from patients entered in clinical trials, to improve the level of evidence of results) from tumors homogeneous for stage and biomolecular subtype and of standardized technical and analytical approaches.

Recent findings in different tumor types indicate that the combined targeting of molecules of self renewal and developmental pathways, associated with conventional agents, could represent a promising treatment strategy in patients with advanced disease [59,60]. Several lines of evidence support these findings in experimental preclinical systems in which targeting such key molecules may interfere with stemness hallmarks even in malignant stem-like tumorigenic cells. In vivo assays still represent the gold standard for identifying and studying novel therapeutic approaches against CSC, but they are time consuming (several months are needed for serial transplantation experiments) and thus make a high-throughput screen to validate compounds difficult. In addition, for some tumor types like breast cancer, the taking of a direct engraftment of tumor specimens into immunocompromized mice is too low to allow for a large-scale screening of novel therapeutic agents. Conversely, the propagation of in vitro cultures of BCIC represents a challenge for dissecting the effect of investigational agents directly on stem cell biology and an opportunity for the possibility to also investigate differentiation as an end point. However, it does not guarantee unbiased results, for being “niche-independent” since interactions with the tumor microenvironment are generally not considered in the assay conditions, as well as for the necessity to control for genetic instability, which might be induced by serial in vitro cell propagation cycles. Indeed, additional and more complex chromosomal rearrangements have been observed in cancer cells from non-adherent spheres compared to the parental populations [20]. Moreover, in vitro assays—which are quantitative and rapid—should ideally also be specific, measuring only the cells of interest, and sufficiently sensitive to measure candidate CSC when present at a low frequency. The possibility to couple in vitro assays and in vivo injection of cells in immunocompromized mice might overcome limitations related to each single test and provide mechanistic insight into the molecular cross-talks, while still representing a valuable pre-clinical system for screening and validation of novel therapeutic approaches.

In the last few years, BCIC isolated on the basis of the capability to grow as non-adherent spheres or the presence of stemness-related markers (CD44 /CD24^low/− and ALDH1, side population) from cell lines and clinical specimens have been challenged against different molecules to investigate the potential of directly targeting BCIC for new therapeutic strategies (Table 2). We reviewed the inhibitors of self-renewal pathways relevant to breast cancer development and differentiating agents as potential strategies to target BCIC and their progenies.

It has been shown that therapeutic approaches targeting the Notch pathway (both selectively, using monoclonal antibodies or RNA interference, and non-selectively, using inhibitors of enzymes involved in glycosylation) were able to suppress the self-renewal ability of BCIC. Specifically, the mammosphere-forming efficiency of non-adherent cultures derived from in situ ductal carcinomas was found to be reduced following treatment with a Notch4-neutralizing antibody or the γ-secretase inhibitor DAPT [47]. Similar results were obtained in mammosphere cultures derived from HER-2 overexpressing carcinoma cell lines following siRNA-mediated Notch1 silencing or γ-secretase inhibitor I (GSI) treatment [48]. It has also been reported that clinically used inhibitors of receptor tyrosine kinases such as gefitinib, a small molecule inhibitor of EGFR, and trastuzumab, the humanized monoclonal antibody directed against HER-2, were also able to reduce tumor-sphere-forming capability [25,47,48]. In addition, the loss of serial transplantability was also reported following treatment of HER-2 overexpressing xenotransplants with trastuzumab [48].

The possibility of specifically targeting self-renewal ability of cancer stem cells led to the use of Suberoylanilide Hydroxamic Acid (SAHA), a pan-histone deacetylase (HDAC) inhibitor [61]. Much evidence suggests that HDAC enzymes control reversible chromatin remodeling, which is a key epigenetic mechanism that regulates transcription of genes associated with multiple activities of normal and malignant cells, including self-renewal, survival and resistance to apoptosis, cell cycle progression, aggregation, motility, invasion and metastasis. SAHA inhibited self-renewal of tumor spheroids from established breast cancer cell lines and pleural effusion aspirates from patients with advanced breast cancers, as assessed by decreased clonogenic growth, and markedly affected mammosphere formation and their 3-dimensional structure, which was associated with translocation of E-cadherin protein from the plasma membrane to the cytoplasm [61].

Interestingly, the lentivirus-mediated reconstitution of let-7, a family of miRNAs regulating several stemness- and EMT-related pathways, was able to reduce proliferation, tumor-sphere formation and the fraction of undifferentiated BCIC in vitro, as well as to affect tumor and metastasis formation in NOD/SCID mice [30]. Increased let-7 was paralleled by the reduction of two known let-7 targets, such as RAS (important for self renewal) and HMGA2 (high mobility group AT-hook 2, important for multipotency). This suggests the possibility of a therapeutic use of let-7 mimics with minimal toxicity to normal tissues that already express this miRNA family [30].

By coupling gene expression analysis with functional studies, Ginestier et al. [62] showed that ALDH1-expressing breast cancer cells also overexpress the IL-8 receptor CXCR1, whose blocking by a specific antibody and/or by the small molecule inhibitor repertaxin reduced the fraction of cells with stemness features in the ALDH1-positive subpopulation from human breast cancer cell lines. This effect was followed by an induction of apoptosis also in the bulk population via a bystander effect mediated by FASL/FAS. In association with docetaxel, repertaxin was able to delay tumor growth and development of metastasis in NOD/SCID mice, thus suggesting novel strategies based on the interference with cytokine regulatory loops to target BCIC. Another study demonstrated that treatment of the side population of breast cancer cell lines with quinoline compound dofequidar fumarate (able to reverse the multidrug resistance phenotype by inhibiting the function of the ABC transporter ABCG2/BCRP) reduced the fraction of cells with stemness properties and sensitized these cells to treatment with irinotecan in vitro and following xenotransplantation into nude mice [63].

The possibility to induce differentiation of breast cancer cells characterized by stemness features was demonstrated by treatment with the potassium ionophore salinomycin [64]. Specifically, treatment of mice with the compound was found to inhibit mammary tumor growth and to induce increased epithelial differentiation of tumor cells. The experimental model used to investigate salinomycin activity is of utmost interest in the field of assays challenging the activity of compounds on putative BCIC, which accounted for major concerns either in in vitro (with the lack of standardized approaches to isolate and propagate cancer stem cells) and in vivo testing. Indeed, the human cancer cells implanted into immunocompromized mice might not realistically recapitulate what happens in patients during cancer initiation and progression. Conversely, the approach proposed by Gupta et al. [64] exploits the use of telomerase-immortalized human mammary epithelial cells, in which targeting E-cadherin by short hairpin RNA promotes EMT and the expression of cellular/molecular features characterizing BCIC. These genetically modified cells, and their parental cell line that does not exhibit stemness-related features, represent an experimental model ideally suitable for the high-throughput screening for compounds specifically targeting BCIC. By using this approach, the authors were able to demonstrate that salinomycin reduces the proportion of BCIC by more than 100-fold relative to a conventional anticancer agent such as paclitaxel [64].

Recent findings also provided support to the possibility to develop chemopreventive strategies targeting breast stem cells in high-risk subjects or BCICs. In fact, it has been recently shown that metformin, a standard drug for diabetes, was able to inhibit cell transformation in an inducible model consisting of non-transformed human mammary epithelial cells (MCF-10A) containing ER-src, a fusion of the v-src oncoprotein with the ligand-binding domain of ER, as well as to selectively kill BCIC in three other mammary adenocarcinoma cell lines derived from genetically and phenotypically different tumors [65]. Very recently, it was reported that tranilast, a non-toxic aryl hydrocarbon receptor (AHR) agonist, was able to inhibit mammosphere formation by BCIC from triple-negative breast cancer cells selected in vitro for mitoxantrone resistance. The drug was also effective in vivo since it prevented lung metastasis in mice injected i.v. with MDA-MB-231 mitoxantrone-selected cells [66]. Again, a novel Gemini vitamin D analogue, BXL0124, was found to inhibit the expression of CD44 in MCF10DCIS.com human breast cancer cells in vitro and in vivo, as well as to down-regulate the transcriptional activity of the CD44 promoter, suggesting a possible role of the compound for BCIC treatment [67].

Recent studies even on tumor types other than breast cancer, showed the possibility to interfere with CSC properties by novel approaches such as telomerase antagonists [68] and an oncolytic reovirus [69], both active also on tumor bulk.