Stem Cells and Cellular Origins of Breast Cancer: Updates in the Rationale, Controversies, and Therapeutic Implications

Breast cancer stem cells have been known to contribute immensely to the carcinogenesis of the breast and therapeutic resistance in the clinic. Current studies show that the population of breast cancer stem cells is heterogeneous, involving various cellular markers and regulatory signaling pathways. In addition, different subtypes of breast cancer exhibit distinct subtypes and frequencies of breast cancer stem cells. In this review, we provide an overview of the characteristics of breast cancer stem cells, including their various molecular markers, prominent regulatory signaling, and complex microenvironment. The cellular origins of breast cancer are discussed to understand the heterogeneity and diverse differentiations of stem cells. Importantly, we also outline the recent advances and controversies in the therapeutic implications of breast cancer stem cells in different subtypes of breast cancer.


INTRODUCTION
Breast cancer is the most commonly reported type of cancer in women worldwide, and a leading cause of morbidity and mortality (1). With the morbidity and mortality associated with breast cancer, more newer therapeutic approaches are warranted. The complexity and heterogeneity of breast cancer render its treatment challenging. Gene expression profiling comprehensively demonstrated molecular portraits of breast cancer (2, 3), defining tumors as luminal-like, HER2-positive, or basal-like. Clinically, the heterogeneous breast cancer is categorized into four distinct molecular subgroups (2,4): Luminal A and luminal B breast cancers are broadly defined as those with estrogen receptor (ERs) positive expression, which response to the antiestrogen therapy. The HER2 positive breast cancer is the subtype with high amplification of HER2 gene. The triple negative breast cancer (TNBCs, usually basal-like), lacking the expression of ER, progesterone receptor (PR), and HER2 (5), always has an increased incidence of germline BRCA1/2 mutations (6,7). Compared to the other subtypes, TNBC is highly heterogenous, and usually have higher incidence of hematogenous metastasis. Based on a large cohort of 465 primary TNBC tumors, TNBC are now classified into transcriptome-based subtypes: luminal androgen receptor, immunomodulatory, basal-like immune-suppressed, and mesenchymal-like (8).
Two models have emerged to explain the heterogeneity of breast tumors (9). One is the clonal evolution model (10), which postulated that random mutation and clonal selection give rise to cellular heterogeneity within breast tumors. Another is the cancer stem cell model (11), which posited that the cellular diversity and tumor hierarchy are generated by the breast cancer stem cells (BCSCs). In both of the two models, tumor microenvironment plays an important role in equipping the evolution of breast cancer cells.
BCSCs (12) are defined as a small fraction of cells capable of self-renewal and propagation of the heterogeneous populations of breast cancer cells. The concept of BCSCs revealed the cellular origin, tumor maintenance, and progression of breast cancer. Clinically, BCSCs are considered to be responsible for the development of resistance to treatment and cancer relapse, through their virtue of relative resistance to radiation, cytotoxic chemotherapy and molecular targeted therapy (13).
In the present review, we provide an overview of the current advances in stem cells and the cellular origins of breast cancer. We summarized information regarding the various molecular markers of BCSCs, BCSC portraits among different subtypes of breast cancer, and associations between BCSCs, and the tumor microenvironment. Moreover, considering the contribution of BCSCs to the development of resistance to treatment and tumor recurrence, we also focused on regulatory pathways and the related therapeutic implications of BCSCs.

MOLECULAR MARKERS AND TUMORIGENIC FEATURES OF BCSCs
BCSCs are defined as a limited group of breast cancerinitiating cells, possessing properties of self-renewal, and differentiation potential (14). In cancer research, BCSCs are usually characterized as cells that are able to form the transplantable tumors and re-establish tumor heterogeneity (15). A panel of molecular markers was used to identify the BCSCs (Table 1). Among these, CD44 + /CD24 −/low (16,17) and ALDH1 + (aldehyde dehydrogenase1) (18) are the most commonly used markers. Indeed, the CD44 + /CD24 −/low and ALDH1 + phenotype cells are two distinct subpopulations of BCSCs (18). Different gene expression profiles (34) have shown that CD44 + /CD24 −/low marked a mesenchymal and quiescent type of BCSCs (EMT-BCSCs, EMT: epithelialmesenchymal transition) (35), while ALDH1 + labeled an epithelial and proliferative type of BCSCs (MET-BCSCs, MET: mesenchymal-epithelial transition). According to the fostered concept of transient EMT-MET switches in metastatic tumor cells (36), these two subgroups of BCSCs can be accepted as two dynamic states of BCSCs. It has been recently demonstrated that the transition between these two states of BCSCs (EMT and MET-BCSCs) can be reversely regulated by cytokine signaling such as Id1 (inhibitor of DNA binding 1) (37).
Considering the heterogeneity of breast cancer, the variety of BCSCs markers observed in different studies may be attributed to different levels of breast cancer hierarchy. Moreover, it may be explained as the results of dynamic states of BCSCs, regulated by the microenvironment. Further in vivo and patient-derived xenograft studies are required for the definitive identification of BCSCs.

SIGNALING PATHWAYS REGULATING BCSCs
Given their self-renewal and tumor-initiating properties, BCSCs have emerged as the "ringleader" for the development of therapeutic resistance in breast cancer (38). Therefore, BCSCrelated therapeutic options, such as targeting the main regulatory signaling pathways in BCSCs, have recently been developed for the treatment of breast cancer ( Table 2), especially in case with therapeutic resistance. The major signaling pathways (39) regulating BCSCs include Wnt (40), Notch (41), and Hedgehog (42) (Figure 1). Inhibitors blocking these signaling were developed as the BCSC-targeting therapies (Figure 1). (1) Increased activation of the Wnt pathway is usually found in BCSCs, leading to the nuclear translocation of cytosolic β-catenin to activate the Wnt-targeted genes,

MICRORNAS REGULATING BCSCs
Epigenetic regulation of microRNAs is important in BCSCs (53). MiR-200 family was proved to be a critical regulator for BCSCs growth and function. MiR-200c can strongly suppress the clonogenicity of BCSCs by targeting BMI1 (BMI1 protooncogene, polycomb ring finger) (54). MiR-200b directly acts on Suz12 (SUZ12 polycomb repressive complex 2 subunit), and the miR-200b-Suz12-cadherin pathway functions on BCSC growth (55). MicroRNA-200c/141 was found to regulate the heterogeneity of the BCSCs and promote the EMT-like BCSC generation, by targeting HIPK1 (homeodomain-interacting protein kinase 1)/β-catenin axis (56). Let-7 was found to regulate BCSCs by silencing H-RAS (HRas proto-oncogene, GTPase) and HMGA2 (high mobility group AT-hook 2), resulting in reduction of self-renewal and enhancement of differentiation of BCSCs (57). Similar to let-7, reduction of microRNA-30 in BCSCs contributes to the maintenance of self-renewal capacity in BCSCs, by targeting UBC9 (ubiquitin-conjugating enzyme 9) and ITGB3 (integrin b3) (58). In addition, in BCSCs, miR-600 acts as a bimodal switch that regulates WNT Signaling through SCD1 (stearoyl desaturase 1), balancing the self-renewal, and differentiation of BCSCs (59). More microRNAs and their clusters need to be investigated in BCSCs, especially for their roles in stem cell maintenance of self-renewal, differentiation, and EMT transition. Clarifying the microRNA regulation of BCSCs can further advance or understanding of the roles of BCSCs in breast cancer progression.

BCSCs AND THE TUMOR MICROENVIRONMENT
BCSCs are located in the tumor microenvironment, which is also called as BCSC "niche." The BCSC niche plays a vital role in sustaining the function of BCSCs. It is a complex

network, containing stroma cells [such as mesenchymal stem cells (MSCs), cancer associated fibroblasts (CAFs), adipocytes, endothelial cells, and immune cells], extracellular matrix (ECM) components, cytokines, growth factors, and physical factors (such as hypoxia).
MSCs are the multipotent mesenchymal stomal cells which can be recruited from the bone marrow or normal breast stroma. It has been shown that bone marrow-derived MSCs can expand the BCSC population through cytokine loops involving IL6 (interleukin 6) and CXCL7 (pro-platelet basic protein) (60). It was also found that MSCs can propagate traits of BCSCs by promoting the contact-dependent upregulation of microRNA-199a and subsequent repression of FOXP2 (forkhead-box P2) (61). CAFs are activated fibroblasts in the tumor-hosting niche, which can promote cancer progression, especially that of breast cancer. It was reported that CAFs can regulate BCSCs through factors such as CCL2 (monocyte chemotactic protein-1) (62), IL-6 and IL-8 (63). Moreover, autophagic CAFs can promote stemness of luminal breast cancer cells by releasing HMGB1 (high-mobility group box 1) (64). Conversely, BCSCs can also regulate CAFs via signaling such as Hedgehog, in which CAFs subsequently promote the expansion and self-renewal of BCSCs (49,65). Immune cells, especially the tumor-associated macrophages (TAMs)-are closely associated with tumor propagation. Tumor cells produce M-CSF (macrophage colony-stimulating factor) to expand TAMs, while TAMs produce TNFα and TGFβ to facilitate CSCs (66). In breast cancer, TAMs can promote BCSCs through a paracrine EGFR/Stat3/Sox-2 signaling pathway (67), while upregulation of HAS2 (hyaluronan synthase 2) in CD44 + /CD24 − /ESA + BCSCs can enhance the interaction between BCSCs and TAMs, resulting in the BCSC growth (68). Other immune cells, such as tumor-infiltrating lymphocytes (TILs, including CD4 + , CD8 + , and FOXP3 + TILs) are also closely correlated with BCSC phenotypes, therapeutic response, and prognosis in breast cancer (69). Recent research by machine learning identified that immune microenvironment content and PD-L1 levels associated with the stemness of breast cancer (70). Adipocytes constitute a major component of the breast stroma which provide protumorigenic signals in breast cancer (71). In the Goto et al. study (72)-using a PDX (Patient derived xenograft) model-it was found that adipose tissue secreted adipsin to enhance BCSC properties in breast cancer. Endothelial cells are necessary in tumor angiogenesis, which is important for nutrient and oxygen supply in the tumor microenvironment. Independently of their vascular functions, Ghiabi et al. (73) found that endothelial cells can enrich the CD44 + /CD24 − stem population in breast cancer.
The ECM is a three-dimensional network of extracellular macromolecules that confines tumor cells in the microenvironment, thereby sustaining tissue homeostasis. It was illustrated that stiffness of the ECM influenced breast cancer cells through the YAP/TAZ (Yes-associated protein/Transcriptional coactivator with PDZ-binding motif) (74). Hpoxia is very important in sustaining the quiescent state of stem cells by activating the hypoxia-inducible factor (HIF). In breast cancer, hypoxia induces the BCSC phenotype through HIF-dependent and ALKBH5-mediated m(6)-demethylation of NANOG mRNA (75).

CELLULAR ORIGINS OF BREAST CANCER
The concept of "cellular origins" is closely associated to-but quite distinct-from the notion of "cancer stem cells" (76). The BCSC concept highlights a tiny population of the breast tumor-initiating cells that can maintain tumorigenesis and seed metastases, while the notion of cellular origins emphasizes the original normal cell types in the breast which generate a fullblown tumor. The diversity of the phenotypes displayed by breast cancer cells stimulates interests in investigating the cellular origins of this disease.
Historically, in breast cancer, the name "luminal" of luminal-A/B subtypes and the name "basal-like" of basal-like subtype are derived from similarities in transcriptomes between breast tumors and the corresponding normal mammary luminal or basal epithelium. However, the real cellular origins of luminal and basal-like breast cancer are greatly different from their naming rules. Oncogenic events in different types of mammary cells lead to distinct types of breast tumors (Figure 1). PIK3CA (α-catalytic subunit of PI3K) mutations occur in 30% of breast cancers, including both luminal and basal-like tumors. However, Meyer et al. (77) reported that the mutant PIK3CA in mammary luminal progenitors generated heterogeneous tumors of both luminal and basal differentiation. Van Keymeulen et al. (78) also found that expression of the PI3KCA mutant in luminal cells-marked by CK8-induced the luminal or basal-like breast tumors, while its expression in basal cells-marked by CK5-gave rise to the luminal tumors. BRCA1 basal-like breast cancers may originate from basal stem cells. However, interestingly, Molyneux et al. (79) demonstrated that deletion of BRCA1 in mammary luminal epithelial cells-targeted by Blg-can generate basal-like breast tumors, phenocopying the human BRCA1-associated breast cancers, while the deficiency of BRCA1 in basal cells-targeted by CK14 − can only generate the malignant adenomyoepitheliomas which are rare in human BRCA1-associated breast cancer. Furthermore, Tao et al. (80) depicted that the CK8 + luminal cells carrying the Etv6-NTRK3 fusion oncogene can induce the heterogeneous tumors with the expression of luminal and basal markers. Compelling evidence showed that luminal progenitors can serve as the cellular origins of both luminal-and basal-like human breast cancers, while the distinct genetic mutationsoccurring in the transformation of luminal progenitors-are probably determinant of the eventual luminal-like or basallike tumor phenotypes (81). Genetic sequencing results have illustrated different mutation profiles between luminal-like and basal-like tumors. The luminal-like tumors present distinct mutations, such as PIK3CA (82,83), GATA3 (84,85) and FOXA1 (84), while the basal-like tumors exhibit high rates of p53 and BRCA1 mutations. Using conditional mouse models, Liu et al. showed that somatic loss of both BRCA1 and p53 did result in the development of basal-like breast cancer (86).
The cellular origins of the rare type of breast cancers, such as metaplastic carcinoma should also be mentioned. Molecularly, this metaplastic subtype is similar to claudin-low breast cancer. Keller et al. (87) found that the transformation of CD10 + basal cells gave rise to rare metaplastic tumors. McCarthy et al. (88) demonstrated that these metaplastic tumors frequently harbored the p53 mutation and aberrant BRCA1 expression.

BCSCs AMONG DIFFERENT SUBTYPES OF BREAST CANCER AND THEIR THERAPEUTIC IMPLICATIONS
Different subtypes of breast cancer exhibit different abundances of BCSCs, as well as varying proportions of epithelial or mesenchymal BCSC subtypes. It is commonly recognized that BCSCs are much more enriched in the TNBCs and HER2 subtypes vs. luminal breast cancer. In the model illustrated by Brooks et al. (9), claudin-low TNBCs are characterized by a high proportion of mesenchymal BCSCs with CD44 + /CD24 −/lo expression, while basal-like TNBCs contain a subcomponent of mesenchymal BCSCs and a higher proportion of epithelial ALDH1 + BCSCs. Liu et al. (89) compared the transcriptional profiles of epithelial or mesenchymal BCSC subtypes in TNBCs and found that the bi-BCSC subgroup (i.e., ALDH1 + and CD44 + /CD24 −/lo ) was highly purified, with expression of prognostic genes such as P4HA2, PTGR1 and RAB40B.
TNBCs harbor the highest proportion of BCSCs compared with other subtypes, contributing to the poor prognosis associated with this subtype (90). Currently, the only established treatment against TNBCs is the cytotoxic chemotherapy; however a considerable number of patients develop resistance (5,91). Recently, a PARP (poly-ADP-polymerase) inhibitorthrough the underlying mechanism of inhibiting DNA repair, has demonstrated good efficacy against BRCA1-associated breast cancer, which usually refers to TNBCs. In breast cancer cell lines, PARP inhibitor (Olaparib) can significantly decrease the proportion of BCSCs with CD44 + /CD24 −/low /ESA + cell surface marker, indicating the potential activity of PARP inhibitor in anticancer stem cells (92) (Figure 2, Table 2). However, Liu et al. (93) recently reported that there were some BCSCs in BRCA1mutant TNBCs which were relatively resistant to PARP inhibitor (Olaparib), and reduction of RAD51 can sensitize these BCSCs to Olaparib treatment. Clinical benefits from the EGFR inhibitor (94) and a pan-PI3K inhibitor (95) have been reported in TNBC patients, while Fu et al. (96) further reported that EGFR/Notch bispecific antibody PTG12 in combination with pan-PI3K inhibitor GDC-0941 exerted a stronger antitumor effect in TNBC tumors by inhibiting the stem cell-like subpopulation and reducing tumor-initiating cell frequency (Figure 2, Table 2). Up to half of all TNBCs express androgen receptor (AR) (97). AR-targeted therapies (Enzalutamide) can decrease a BCSC-like population in TNBC cell lines (98), indicating Enzalutamide may enhance the efficacy of chemotherapy by targeting a BCSC like cell population (Figure 2, Table 2). In TNBCs, hypoxia-driven BCSCs abated the effectiveness of paclitaxel-based chemotherapy and antiangiogenic agents (e.g., VEGF inhibitors bevacizumab), through HIF-1α (hypoxia-inducible factors 1α). In pre-clinical models, the co-administration of HIF-1α inhibitors was able to overcome BCSC-related resistance (99,100). HIF-1α vaccination can also inhibit tumor growth in TNBC models of C3(1)Tag mice and decrease SCa-1 marked BCSCs (101) ( Table 2). Recently, Cho et al. (102) reported that L80, which is the C-ring truncated deguelin derivative as a C-terminal HSP90 inhibitor, can effectively target BCSC-like trait in TNBCs, together with obvious reduction in CD44 + /CD24 − cancer cell population, ALDH1 activity and mammosphere forming-ability ( Table 2).
Some drugs were repurposed as the BCSC inhibitors in TNBCs ( Table 2). Disulfiram, an antialcoholism drug, was found to target BCSCs and reverse the acquired pan-chemoresistance in TNBC cell lines (103) ( Table 2). Mifepristone, a progesterone antagonist for abortion, was reported to suppress BCSCs in TNBC tumors by down-regulating Krüppel-like factor 5 (KLF5) expression, which is a stem cell transcription factor overexpressed in basal type TNBC (104) ( Table 2). Metformin, a first-line drug for type 2 diabetes mellitus, was found by the same group that can decrease the BCSC population in TNBC, also by targeting KLF5 for degradation (105) ( Table 2). Besides, sulforaphane, a cancer prevention agent, was found to preferentially eliminate BCSCs by inhibiting NF-kB p65 subunit translocation and downregulating p52 (106) ( Table 2). And Sulforaphane can also reverse taxane-induced ALDH + BCSC enrichment (106). Interestingly, novel medicines such as nanomedicines, Gd@C 82 (OH) 22 nanoparticles, were found possess intrinsic inhibitory activity against BCSCs in claudin-low TNBC cell lines (107).
In Brook's model, HER2 breast cancer is characterized by a high proportion of epithelial ALDH1 + BCSCs (9). HER2 is a crucial regulator in BCSCs (108). Recent studies indicated that the regulation of BCSCs by HER2 was not observed only in HER2 breast cancer but extended to all the subtypes of breast cancer (109). To some extent, the remarkable clinical efficacy of HER-2 inhibitors (e.g., trastuzumab, pertuzumab, lapatinib, and TDM-1) was attributed to target BCSCs (109) (Figure 2, Table 2). Nevertheless, a proportion of HER-2 amplified breast cancers continue to eventually develop drug resistance, probably due to the PTEN (phosphatase and tensin homolog) loss or activation of PIK3CA mutation (110). A recent study performed by Sun et al. (111) showed that MEOX1 (mesenchyme homeobox 1) may be a novel target in BCSCs of PTEN-deficient trastuzumabresistant breast cancers. However, the precise role of BCSCs in the development of resistance to HER-2 inhibitors remains elusive. Thus, further investigation is required to elucidate this process.
Luminal A breast cancer had the lowest proportion of BCSCs among all the breast cancer subtypes, presenting the best prognosis (9). Luminal B breast cancer displayed a certain proportion of BCSCs, which was lower than those observed in TNBC or HER2 breast cancer. A proportion of luminal-B patients have poor prognosis, supposed to be the result of containing a proportion of the BCSCs (112). The presence of the BCSCs is also regarded as the main cause of resistance to hormonal therapy in luminal breast cancer (113,114), which can be regulated by the CyclinD-CDK4/6 (cyclin-dependent kinase 4/6) complex (115,116) and mTOR signaling (117) (Figure 2, Table 2). Use of CDK4/6 (i.e., palbocilib) (118) or mTOR inhibitor (i.e., everolimus) (119) significantly improves the survival of patients who develop endocrine resistance. Recently, the PI3K inhibitor (i.e., alpelisib) was also proved to have great clinical activity in PIK3CA-altered luminal breast cancers, including the endocrine therapy-resistant cases (120).
As described above, BCSCs are of great therapeutic relevance, particularly in overcoming resistance to treatment. Although pre-clinical models and clinical trials yielded promising results in targeting BCSCs, the efficacy and safety of BCSC-targeted therapy requires further evaluation. BCSCs share the similar markers and signaling pathways with mammary stem cells. Currently, it is unclear whether BCSC-targeting agents may also target normal mammary stem cells, resulting in severe treatment-related side effects. Besides, the clinical trial data have shown that addition of BCSCs-targeting agents to chemotherapy or hormonal therapy improved treatment efficacy.
However, the combination of BCSC-targeting agents and different conventional therapies requires further investigation.

CONCLUSIONS AND PERSPECTIVES
Impressive advances have been witnessed in understanding the carcinogenesis of breast cancer, in which BCSCs hypothesis provided very important models. BCSC-targeted therapy may eradicate the cancer stem cells, which are regarded as the "the seed of tumor." Therefore, indubitably, the clinical relevance of BCSCs is a primary concern. However, the BCSC population is heterogeneous, involving various cellular markers and regulatory signaling pathways. In addition, the BCSC microenvironment is complex. These facts render BCSC-targeted therapy difficult. Also, different subtypes of breast cancer exhibit distinct BCSC subtypes and abundances. This evidence calls for appropriate BCSC-targeted regimens against different subtypes of breast cancer. Moreover, considering that BCSCs only constitute a small fraction of the tumor cells, the classic clinical endpoints of tumor shrinkage may be inappropriate to assess the efficacy of BCSCtargeted therapy. Recent studies have indicated that the detection of BCSCs markers in tumor biopsies or even in circulating tumor cells may be a favorable option (121,122).
In conclusion, despite challenges ahead, the comprehensive understandings of stem cells and the cellular origins of breast cancer have assisted and will continue to assist treating physicians in ultimately overcoming the stubborn aspects of breast cancer.

AUTHOR CONTRIBUTIONS
JZ, YZ, and QC mainly drafted the manuscript. JZ and YC supervised and edited the manuscript. YZ, HC, and LQ participated in drawing the pictures. QC edited the language of the manuscript. All authors read and approved the final manuscript.