Cancer stem cells and strategies for targeted drug delivery

Abstract

Cancer stem cells (CSCs) are a small proportion of cancer cells with high tumorigenic activity, self-renewal ability, and multilineage differentiation potential. Standard anti-tumor therapies including conventional chemotherapy, radiation therapy, and molecularly targeted therapies are not effective against CSCs, and often lead to enrichment of CSCs that can result in tumor relapse. Therefore, it is hypothesized that targeting CSCs is key to increasing the efficacy of cancer therapies. In this review, CSC properties including CSC markers, their role in tumor growth, invasiveness, metastasis and drug resistance as well as CSC microenvironment are discussed. Further, CSC-targeted strategies including the use of targeted drug delivery systems are examined.

Graphical Abstract

1. Introduction

Stem cells are specialized cells that possess the ability to differentiate into varied cell types. In a tumor, all tumor cells are not alike; some tumor cells possess greater differentiation capabilities than others. The notion that tumors have stem cells — cancer stem cells (CSCs) — has stirred a lot of scientific interest and greatly changed how we think about treating cancer. These self-sustaining stem cells possess characteristic self-renewal ability, enabling them to give rise to heterogenous tumor lineages. Since these cells are capable of forming an entire tumor, they are also called tumor initiating cells (TICs). The TICs differentiate into progenitor cells and further into other kinds of cells found in the tumor. Two models have been used to describe the origin of CSCs. The stochastic model (Figure 1) envisages that every tumor cell has a low but equal chance of limitless proliferation, behaving as a stem cell. As naturally acquired cell mutations are selected over time, the most aggressive tumor cell populations are selected. These populations are responsible for tumor progression and maintenance. The hierarchal model, on the other hand, proposes that only a distinct subset of tumor cells possess the potential of limitless proliferation. These cells possess the ability of self-renewal and form the top of the tumor pyramid [1]. They have the capacity for asymmetric cell division, that is, they can divide into daughter CSCs or into progenitor cells. The rapidly dividing progenitor cells form the middle of the pyramid, while the differentiated, heterogenic, non-tumorigenic cancer cells that make the bulk of the tumor are at the bottom of the pyramid (Figure 1a) [2].

Fig. 1.

Proposed models for cancer stem cell evolution. Hierarchically organized tumors (a) have a subpopulation of self-renewing cells maintaining the clonal population at the top of the pyramid while giving rise to larger number of progenitor cells ultimately differentiating into other cells in the tumor. The tumor formed has low heterogenicity. Clonal model (c) suggests that selective pressure and continued mutations give way to a highly aggressive population of cells which are tumor initiating and can also self-renew. With different cell phenotypes present, the tumor produced has a high heterogenicity. The unified model (b) suggests stem cell plasticity where cancer stem cells can transition between non-CSCs and CSCs given the right conditions. The resulting tumor is complex, has high heterogenicity and increased resistance to most anti-cancer therapies

While the models of CSC origins were initially postulated in the late nineteenth century, it was in 1997 that Bonnet and Dick identified CSCs by demonstrating that acute myeloid leukemia (AML) arose from preexisting cells with stem-cell like features, emphasizing the hierarchical origin of CSCs. They extracted hematopoietic cells from patients with AML and serially transplanted into NOD/SCID mice to model progression of leukemia in vivo [3]. It was shown that a core population of cells expressing CD34 (similar to the cell-surface phenotype of normal SCID-repopulating cells) were responsible for generation of leukemia in xenografts. With the discovery that CD34⁺ cells could initiate leukemia, it was suggested that CSCs could play a role in cancer development in other cancers as well. This was later realized in solid tumors when Al-Hajj et al. identified a small portion of CD44⁺CD24⁻ breast cancer cells, which when reinjected into immunodeficient mice, led to cancer incidence in 8 of 9 mice [4]. As little as 100 of the CD44⁺ cells resulted in a new tumor whereas a variety of other tumor cells when isolated and reinjected did not result in tumors. While the hierarchical model explains the tumor origin better, it has been additionally demonstrated that CSCs possess bi-directional differentiation ability with any cancer cell capable of becoming a CSC (or vice versa) under the right conditions [5]. This is the basis for unified CSC model widely accepted today, and extended to various human tumors with additional therapeutic implications [6]. Finally, CSCs have been defined based on the following three properties:

1. Differentiation – ability to form a heterogenous population, which constantly diversifies and specializes according to a hierarchical setup

2. Self-renewal – ability to regenerate into new identical stem cells with the same ability to proliferate, thus maintaining the stem cell population

3. Homeostasis control – ability to modulate self-differentiation and proliferation according to environmental stimuli and genetic obstacles

Like the hematopoietic system, other tissues need to undergo cyclic renewal for maintenance. The development of in vitro culture systems to identify various cancer cell populations has led to the observation of a hierarchical system in various tumors including those of breast, colon, pancreas, lung, prostate, brain, and skin [4, 7, 8, 9, 10]. The CSC working model is being further extended to lymphoid tumors including multiple myeloma, where studies reported a small population of CD138⁻ cells possessing engraftment capability in NOD/SCID mice [11, 12]. The number of CSCs range from undetectableto highly abundant in tumors, irrespective of the stage of tumor development. This could be due to the limitations in identifying a set of surface markers relatively unique to CSCs. The other commonly used techniques to isolate or enrich CSC populations include detection of a side population of cells with distinct ability for Hoechst 33342 exclusion, assessment of sphere forming ability in suspension culture, and measurement of aldehyde dehydrogenase (ALDH) activity. Different techniques for detection and isolation of CSCs is discussed in detail elsewhere [13]. Regardless of the techniques utilized, the isolated or enriched CSC populations must be tested by serial transplantation/xenografting in secondary recipient/immunodeficient mice to span the full spectrum of tumor progression. The tumors that develop must represent the tumor heterogeneity of the parent tumor for the injected cells to be considered a reliable CSC population.

Current treatments for cancer largely include a combination of surgical resection, chemotherapy, targeted therapy, immunotherapy and radiotherapy. Cancer cells eventually develop resistance to most therapies. Cytotoxics also induce tumor heterogeneity further resulting in resistance. Several recent reports suggest that CSCs contribute significantly to the development of therapy resistance. Chemoresistance due to alterations in apoptotic processes, epigenetic modifications, and increased expression of anti-apoptotic proteins (all of which have been observed in CSCs) has been reported in many cancers including colorectal cancer, ovarian cancer, and glioblastoma [14]. Even after surgically removing the tumor, the presence of a very small population of CSCs may cause tumor reoccurence [15]. While treatments targeting CSCs could be an effective approach to eradicate cancer, CSCs are inherently resistant to various treatments [16]. This review examines the properties of CSCs that contribute to therapy resistance and discusses potential approaches to overcome these mechanisms.

2. CSCs and anti-cancer therapy resistance

CSCs adapt to changes in tumor microenvironment (TME) making them more resistant to anti-cancer therapies than bulk tumor cells. Mechanisms of resistance include a combination of intrinsic and extrinsic factors. CSCs are inherently resistant to radio- and chemotherapy due to their highly active anti-apoptotic machinery, protective autophagy, efficient cell cycling and DNA repair, along with highly qualified epithelial-mesenchymal transition (EMT) regulators, reactive oxygen species (ROS) scavengers, and drug transporters [17, 18]. CSC resistance is also reinforced by TME signals, which are exceptionally powerful in advanced tumor stages, thus additionally complicating tumor-targeted therapies [19]. Stressful TME (caused by hypoxia, increased EMT, etc.) promotes pluripotency and increases CSC stemness [20]. Further, a multitude of signaling pathways (including TGF-β, PI3K/AKT, NF-κB, Wnt/β-catenin) which independently or collectively induce EMT are observed to suppress the immune system, promote cell stemness and induce stem cell quiescence [21]. Various mechanisms of therapy resistance in CSCs are presented in Figure 2 [17].

Fig. 2.

Mechanisms of therapy resistance in CSCs. ABC transporters: ABC transporters reduce the effective concentration of cytokines and administered chemotherapeutics in the cytoplasm. Microenvironment/Hippo: The tumor microenvironment (increased EMT, hypoxic conditions, inhibited Hippo pathway) contributes to increased stemness of CSCs. Epigenome: Instability of chromatin and changes in epigenetic modulators allow for phenotypic differentiation of CSCs. Exosomes: Exosomes promote a CSC phenotype in normal tumor cells by transporting MDR transport proteins to recipient drug sensitive cells. DNA damage repair (DDR): An increased DDR is observed in CSCs due to overexpression of proteins involved in repairing DNA damage (PARP-1, BRCA1, RAD 50, etc.). UPR/ERS: Endoplasmic reticulum stress (ERS) is characterized by presence of misfolded protein in the ER lumen. The unfolded protein response (UPR) signal transduction pathway which balances the ER protein folding is disturbed in the presence of certain physiological conditions (hypoxia, TME, unstable genome). Autophagy: Autophagy has been shown to increase CSC survival and aggressiveness by improving stress tolerance. Metabolism: CSCs can efficiently adapt metabolic pathways or switch pathways to maintain an energy source. This makes them thrive in adverse conditions and imparts resistance to most conventional anti-cancer treatments. Apoptosis evasion: CSCs evade apoptosis by regulating (or altering) apoptotic genes during transcription, translation or even post-translation. Figure adapted from [17]

Rapid cell proliferation is energetically taxing. Tumor cells have higher energy requirements and must adapt to survive in oxygen- and nutrient-deficient environments. One of the intrinsic adaptive measures utilized by tumor cells to survive these conditions is Unfolded Protein Response (UPR). Endoplasmic reticulum (ER) stress is generally caused when misfolded and unfolded proteins concentrate in the ER. Unresolved ER stress could lead to programmed cell death. CSCs present themselves with an adaptive UPR as the ER receptors (Inositol Requiring Enzyme 1α, Proteinkinase RNA-activated-like ER Kinase and Activating Transcription Factor 6) signal to relieve ER stress and reestablish ER homeostasis. An adaptive UPR is known to support tumor growth and increase drug efflux during chemotherapy [22].

Cancer cells, like normal cells, suffer from minor DNA damage as a result of regular wear and tear. These DNA damage events may modify transcription or replication, result in mutations, or alter chromatin arrangements. When the DNA damage is higher, the normal response is for the cell to undergo apoptosis. For CSCs, a DNA damage response (DDR) characterized by the overexpression of proteins involved in repairing DNA damage (PARP-1, BRCA1, RAD 50, etc.) addressees this problem. A highly active DDR can attenuate the effectiveness of DNA damaging anticancer drugs or radiotherapy. With a highly active DDR, CSCs may be spared apoptosis but end up with severe mutations, which further increases cancer cell heterogeneity and CSC stemness. A higher DDR and repair gene expression have been observed in CD133⁺ CSCs in lung carcinoma cells, breast TICs, pancreatic CSCs, glioblastoma and non-small-cell lung cancer [23].

Epigenetic modulation entails turning on and/ or off genes without altering DNA sequences. These changes in rate and frequency of genetic expression are heritable and are caused by a variety of lifestyle factors. Epigenetic modulation includes DNA methylation (capping a methyl group to part of the DNA molecule, preventing expression of certain genes), histone modification (reversible change to the N terminal of histone that affects chromatin packing) or chromatin remodeling. Overexpression of DNA repair proteins imparts effective anti-apoptotic nature to CSCs. As tumor progresses, alterations of epigenetic modulators (TP53, STAT3, STAT 1, etc.), chromatin modifications, and DNA methylation promote cell reprogramming, producing different CSC phenotypes with increased stemness [24]. Tumor progression is favored by mutated epigenetic regulators such as chromosomal rearrangement of gene KMT2A/MLL (which encodes a histone methyltransferase) in mixed lineage leukemia. Mutations in DNA methyl transferase (DNMT)1 and DNMT3A have been found in colorectal cancer and AML, respectively [25, 26]. DNMT3A mutations appear to reduce the enzyme activity, resulting in the expansion of pre-leukemic stem cells [27].

Autophagy, or cell self-digestion mechanism, is important to maintain cell homeostasis by keeping a quality check on cell organelles, especially during stressful conditions. CSCs are characterized by a high level of autophagy due to demanding conditions like hypoxia. Autophagy plays a role in various CSCs such as those in cancers of breast, pancreas, liver, ovaries, bone and brain [28]. Expression of CSC markers such as CD44 and mesenchymal markers like vimentin increases with autophagy [29]. Autophagy increases CSC survival and stemness by improving stress tolerance and supplying nutrients to meet the increased metabolic needs of CSCs.[30] Autophagy is induced with chemotherapy, radiotherapy, inhibition of mTOR, AKT or PI3K, but tumor cells evade apoptosis, as autophagy increases FOXO3A. This limits FOXO3A-dependent induction of Puma (p53 upregulated modulator of apoptosis), a Bcl-2 family pro-apoptotic protein [29].

The tumor works closely with various inflammatory molecules, signaling pathways, fibroblasts, blood vessels and extracellular matrix. The CSC TME is altered with upregulated signaling pathways and hypoxic conditions favoring tumor progression and maintenance. Understanding the complex working of CSCs is critical to designing effective therapies that target and eliminate CSCs. The implications of drug transporters, CSC surface markers and TME in anti-cancer resistance are discussed further in the subsequent sections.

2.1. Drug transporters

Decreased expression of Reduced Folate Carrier (RFC) and polymorphisms in the gene lead to reduced uptake of commonly used chemotherapeutic agents such as methotrexate, a dihydrofolate receptor (DHFR) substrate [31]. Reduced expression of transporters or mutations in the genes encoding for a transporter can result in reduced uptake of drugs [32]. ATP-dependent efflux pumps are major contributors to multi-drug resistance (MDR). ATP Binding Cassette (ABC) transporters is a superfamily of transmembrane proteins consisting of 49 MDR transporters in 7 subfamilies (designated ABCA-G). ABC transporters form a pore like structure that spans the entire thickness of the membrane, with two transmembrane domains and two nucleotide binding domains (NBD). When a substrate binds to the NBD, a molecule of ATP gets hydrolyzed to ADP while the conformational structure of the protein changes to allow movement of the substrate out of the cell. ABC transporters are involved in the movement of a variety of materials including sugar, proteins, xenobiotics and polysaccharides. Different ABC transporters (ABCB1, ABCC1 and ABCG2) are found to be overexpressed in cancers of prostate, lung, and breast, as well as in childhood neuroblastoma. [33, 34, 35, 36, 37]. It has been established that ABC transporters keep the cytoplasmic concentrations of drugs and other signaling molecules low, rendering anti-cancer treatments ineffective [37]. This also leads to MDR as cancers become resistant to other unrelated chemotherapeutics after exposure to one drug. MDR transporters are also overexpressed on CSCs compared to bulk cancer cells, making them more resistant to treatments and helping tumor propagation. Targeting CSCs using specific inhibitors of ABC transporters is an attractive strategy to target CSCs. In addition to contributing to chemoresistance, ABC transporters work in tandem with various other components of the tumor system, tumor signaling pathways and TME. ABC transporters also regulate redox cell status. Increased Nrf-2 in blood brain barrier (BBB) during oxidative stress increases protein expression and activity of ABCB1, ABCG2, and ABCC2. This decreases drug perfusion across CNS barriers [38]. Further, during oxidative stress, ABC transporters (ABCC1–4, 4, 7 and ABCG2) transport GSH and GSH conjugates, which are important for maintaining redox homeostasis. ABC transporters are also involved in lipid and cholesterol transport. ABCA1, which transports cholesterol and phospholipids to plasma membranes, is believed to be involved in cancer progression in prostate and ovarian cancers by transporting lysophosphatidylinositol and sphingosine 1-phosphate [39, 40, 41]. ABC transporters are also implicated in tumor metabolism. Mitochondrial metabolism, which is critical for tumorigenesis and ATP production, could meet the energy requirements of overexpressed ABC transporters in CSCs.

2.2. CSC surface markers

Several cytosolic and cell surface markers present on CSCs have been identified and characterized. Cell surface markers are proteins or carbohydrates on or within the cell’s plasma membrane that support intercellular communication and cell recognition. ‘Cluster of differentiation’ (CD) is the nomenclature that is used to identify cells based on a set of surface markers. CD markers can be used to differentiate cancer cells from normal cells and to identify CSCs within a cancer cell population. Surface macromolecules also present themselves as receptors or antigens for targeted drug delivery. These molecules can be recognized by magnetic micro-bead isolation techniques and detected by flow cytometry. Knowledge of unique CD markers is pivotal for diagnosis and subsequent development of therapeutic strategies targeting CSCs. Bonnet and Dick, for the first time, reported that the SCID-leukemia initiating cells (SC-ILs) from all subtypes of AML analyzed were exclusively CD34⁺ CD38⁻ [3]. This seminal work led to the search for other CSC surface markers in related blood cancers as well as solid tumors. Jordan et al. demonstrated that CD123 is a unique marker for primitive leukemic stem cells [42]. As reported by Hauswirth et al., CD33, a sialic acid-dependent single transmembrane adhesion protein triggered by phosphorylation, is expressed on leukemic stem cells in AML and is unique to CSCs [43]. Thus, the combination of CD34⁺/CD38⁻/CD123⁺/ CD33⁺ is a general marker phenotype for AML CSCs. Hosen et. al. identified that CD96, which belongs to the immunoglobulin superfamily, is an important marker of AML CSCs [44]. CD96 was demonstrated to be overexpressed on a majority of CD34+/ CD38− AML cells compared to the basal expression level of CD96 on normal HSCs. A list of various CSC surface markers recognized in different cancers are listed in Table 1. Readers are directed to a detailed discussion of the structure, function and implications of CSC markers in cancer elsewhere [45].

Table 1.

Commonly expressed cancer stem cell surface markers in different cancers.

CSC surface marker	Cancer	Reference
CD33	Hematological	[46]
CD34	Hematological, liver	[47]
CD38	Hematological	[48]
CD123	Hematological	[49]
CD19	Hematological	[50]
CD13	Liver	[51]
CD26	Hematological, colorectal, lung	[52–54]
CD24	Breast, colon, pancreas, prostrate, esophagus, nasopharynx, ovary, kidney, lungs	[55–57]
CD44	Breast, prostate, pancreas, ovary, colon, liver, lung, head and neck	[55, 58, 59]
CD133	Breast, colon, brain, liver, lung, pancreas, prostate, stomach, melanoma	[60]
Ep-CAM	Liver, pancreas	[61, 62]
CD90	Brain liver esophagus	[63]
CD20	Skin	[64]
CD271	Head and neck, esophagus, skin	[65, 66, 44]
CD96	Hematological	[44]

As seen in Table 1, the cell surface markers for solid tumors differ from those in hematological malignancies. CD44 is one of the most commonly expressed CSC markers found in various solid malignancies including those of the bladder, colon, liver, ovaries, pancreas, and prostate. These markers are expressed individually or together with CD24, CD133, CD34, or c-Met markers [67, 68, 69, 70, 71, 72]. Al-Hajj et al., prospectively identified CD44⁺/CD24^−/low Lin ⁽⁻⁾ phenotype to have breast cancer initiating potential. This phenotype is now correlated with high resistance to standard anti-cancer therapies [4]. CD44⁺ cells were also identified to be gastric colon cancer initiating cells [73, 74]. CD44 is particularly important for CSCs as it regulates cancer cell adhesion and homing, cell extravasation and migration, crosstalk between stem cells and their niche, EMT, and the suppression of cell apoptosis [45]. Further, the unique structure of CD44 (a link domain and a hyaluronic acid binding domain on its globular structure) makes it an important target for drug delivery applications.

CD133 is one of the most prominent CSC markers and was initially categorized as a marker of primitive hematopoietic and neural stem cells [75]. CD133⁺ subset was the first reported putative CSC population for hepatocellular carcinoma [76]. It is a transmembrane protein with 5 subunits and 865 amino acid residues. Commonly located on cell membrane protrusions, it has been implicated in membrane reorganization [77]. CD133 has now been shown to be present on tumor initiating cells in a variety of cancers including breast, brain, prostate, ovarian, lung, liver, and colon cancer [78, 79, 80, 81, 82, 83, 84]. Under stressful conditions, such as hypoxia, CD133 is significantly upregulated in human glioma cells [85]. Other surface markers, which promote CSC progression and maintenance include CD13, CD47, and epithelial cell adhesion molecules (EpCAM). EpCAM regulates Ca2+-independent homotypic cell–cell adhesion, is limited to epithelial cancers and could be either tumor promoting or tumor suppressive based on the type of cancer in question [86]. In the seminal work by Hajj-Al et al., it was shown that EpCAM⁺ lineage fraction had a significantly higher frequency of TICs than the EpCAM⁻ lineage fraction. EpCAM is now a well-established marker for TICs, CSCs and cancer cells, and an attractive target for CSC targeted therapies. CD13 has been identified as a novel surface marker in liver CSCs. Using human liver cancer cell lines and clinical samples, Haraguchi and coworkers demonstrated that CD13⁺ cells were present majorly in the G₀ phase of the cell cycle and formed cellular clusters within the tumor. Upon treatment with 5-fluorouracil, these cells survived and amplified along the fibrous capsule where liver cancers usually relapse [51]. CD47 has found to be overexpressed in many cancers including small cell lung cancer, leukemia, non-Hodgkin’s lymphoma, and breast cancer. As overexpression of CD47 helps CSCs avoid phagocytosis by macrophages, CD47 blocking treatments offer a targeted treatment approach for many cancers. [87].

Proteoglycans (PGs) are molecules formed through the covalent interaction of protein core with glycosaminoglycan (GAG) chains. The GAG chains are subdivided into (a) heparan sulfate (HS), (b) chondroitin sulfate, (c) keratin sulfate, and (d) hyaluronic acid. PGs play a key role in cancer biology as they have an affinity to bind both with ligands and receptors that modulate tumor growth and neovascularization. They can be broadly classified into pericellular, cell-surface, intracellular, and extracellular matrix PGs based on their location in the cellular environment. HS-carrying perlecan and agrin along with Collagens XVIII and XV constitute the pericellular PGs that are associated with the assembly of basement membranes. Localization of HSPGs in the basement membrane controls their barrier functions and coordinates cell-cell/ECM-cell associations. Among the cell-surface PGs, syndecans influence the growth factor receptor signaling while the glypicans can effectively modulate Hedgehog and Wnt cell signaling pathways. The intracellular PG, serglycin, has a role in inflammation and tumor progression and is overexpressed in a variety of hematological and non-hematological cancers [88, 89, 90]. It physically binds to CD44 and functions in cell signaling, lymphocyte activation, hematopoiesis and tumor metastasis. Interaction of serglycine with CXCL4 chemokine receptor contributes to angiogenesis. Perlecan, a highly glycosylated HSPG component of the extracellular matrix (ECM) plays a vital role in the presentation of growth factors to receptors. It contributes to cancer progression by establishing connections between cells and signaling molecules in the process of ECM dysregulation, angiogenesis, and invasion. Both tumor cells and the host stromal cells are known to synthesize perlecan. Interaction of perlecan with angiogenic growth factors helps in the generation of new blood vessels during cancer proliferation. Elevated expression of perlecan is observed in cancers of colon, pancreas, breast, prostate, brain and skin [91, 92]. Suppression of perlecan has been shown to downregulate invasive nature of melanoma cells and block tumor growth and angiogenesis in vivo [93]. We observed significant cell surface expression of HSPG2 in both primary tumors and metastatic lesions in triple negative breast cancer (TNBC).

2.3. CSC microenvironment and its “niche”

The CSC theory is consistent with the presence of cell communities that retain tumor growth through self-renewal and generating the bulk tumor, on the basis of the idea that tumors are organ systems regulated by aberrant development and homeostasis [94]. The TME consists of the tumor stroma that constitutes most of the tumor mass, including the ECM, mesenchymal stem cells (MSCs), endothelial cells, immune cells, as well as cytokine networks and growth factors [95, 96].’Niche’ is a term used to describe the microenvironment where CSCs reside. The role of the niche has been demonstrated to be relevant at each stage of tumorigenesis. Accumulating evidence has suggested that alterations in niche factors contribute to tumorigenesis [97]. As the niche gets established, CSCs employ tumor-associated macrophages (TAMs), cancer-associated fibroblasts (CAFs) and other stromal cells to create paracrine loops for supplying CSCs with tumor necrosis factor α (TNF-α), transforming growth factor β (TGF-β) and interleukins (ILs) for CSC maintenance. CSCs also contribute to metastatic progression by establishing perivascular niches enhanced in angiocrine factors. The adjacent stromal cells secrete ECM, which in turn frees TGF-β and other growth factors such as vascular endothelial growth factor A (VEGF-A), in order to enable further tumor growth [98]. These supportive niches also retain the characteristics of CSCs, sustain their phenotypic plasticity, safeguard them against drug-induced apoptosis, promote their metastatic ability and are critical for stemness maintenance [96, 98, 99, 100]. In order to preserve their stemness, CSCs identify and communicate with certain vascular niches. Accelerated formation of the vascular niche is assisted by vascular endothelial cells, one of the CSC niche elements [101]. Tumor neovasculature is often deficient in the basement membrane and supplied with endothelial CSC (ECs) and other related cells that may encourage tumor invasion and metastasis. Vascular niches can enhance stem-like properties of CSCs and can help CSCs escape radio-/chemo- therapy [99]. Because tumor angiogenesis is necessary for tumor development and metastasis, drugs that affect angiogenesis and vascular permeability can be effective in stopping the formation of a pre-metastatic niche, leading to the elimination of metastatic CSCs [102].

The ECM is responsible for maintaining tissue architecture and homeostasis and maintaining cell populations including immune cells, capillaries, and fibroblasts. The ECM in the CSC niche raises two problems from a therapeutic point of view. First, together with increased intratumoral pressure, ECM inhibits drug transport and shields CSCs from chemotherapy. Secondly, the active ECM environment encourages the growth of tumors through enhanced signaling and may favor cancer cell migration [103]. Therefore, modifying ECM to enhance the penetration of drugs and drug nanocarriers can improve targeting of CSCs and thus provide superior anti-tumor effects [104].

Hypoxia is a key component of the TME and potentially contributes to improved CSC survival. Activation of a subset of proteases that promote metastasis accompanies the acidic microenvironment of hypoxic cells. Also, hypoxic cells are generally located deep in the tumor and are hard to reach. Due to the inaccessibility of these areas along with anomalous angiogenesis and acidic microenvironment, therapeutically effective drug concentrations are not achieved, leading to resistance to therapy. Hypoxia inducing factors (HIFs), a family of transcription factors, are responsible for maintaining homeostasis under hypoxic conditions. Stability of HIF-α expressed in many tumors is necessary for tumor maintenance. Targeting HIFs for therapeutic elimination of CSCs and reprograming of the TME as well as targeting of the hypoxic CSC niche in conjunction with chemotherapy are attractive approaches to eradicate CSCs [105, 106, 107].

Recent studies suggest that EMT plays a significant role in disease progression and generation of distant metastasis in many solid tumors [108, 109]. EMT is a regulatory development program essential for the creation and maintenance of CSCs [110]. Acquisition of CSC characteristics is strongly linked to the progress of EMT transdifferentiation [111]. Signals from the TME tightly regulate primary tumor growth and metastasis processes. Through activation of transcription factors (TFs) such as Snail, Slug and Zeb, the malignant epithelial cell’s transcriptional program is modified and results in the loss of cell-cell attachment and the gain of mesenchymal phenotype and motility. The transition from epithelial to mesenchymal and more motile phenotype in CSCs promotes dissemination [112]. Previous studies show that EMT-related transcriptional signals have an overlapping pattern with those found in CSCs, involving TGFβ, Wnt, Hedgehog, Notch, miR-200 and miR-205 [112, 113]. The relationship between EMT and CSCs could be exploited to design novel therapeutic strategies to reduce tumor recurrence, metastasis and therapy resistance [114].

2.4. CSC-targeted treatment: the importance and need

Because of the peculiar properties of CSCs, including high tumorigenic activity, self-renewal ability, and multilineage differentiation potential [115], developing therapeutic strategies to combat CSCs has gained increasing attention. Standard cancer therapies including chemotherapy, radiation therapy, and molecularly targeted therapy aim to eliminate as many tumor cells as possible, but because they are not effective against CSCs, they fail to eliminate the tumors completely [116, 117]. CSCs are an undifferentiated cancer cell subpopulation but possess “robustness”, which has been blamed for a slow cell cycle, the ability to efflux cytotoxic agents, resistance to oxidative stress, and rapid response to DNA damage [115]. Hence, standard therapies lead to enrichment of CSCs, further promoting cancer relapse [118, 119, 120, 121]. CSC-targeted therapies are thus considered an effective and promising treatment strategy to eradicate cancer [100]. Approaches that simultaneously kill both the bulk of cancer cells and CSCs to prevent metastasis and recurrence will improve treatment outcomes. [122]. Furthermore, it is necessary to develop delivery systems for more efficient targeting and eradication of CSCs and overcoming drug resistance [123, 124, 125, 126].

3. Strategies to target CSCs

3.1. Targeting ATP-driven efflux transporters

The ABC superfamily in humans consists of genes encoding membrane proteins involved in energy-dependent transport of various substances across cell membranes. Many tumors after cancer relapse exhibit resistance to multiple drugs. ABC transporters are overexpressed in MDR tumors and lead to poor treatment outcomes [127]. Over the last three decades, expression of various transporters in tumors and their relevance in anti-cancer therapy have been examined. MDR inhibitors have been reported to target CSCs and improve chemotherapy outcomes [128, 129, 130, 131]. It is important to understand the molecular mechanisms of MDR transporters to effectively target them. The ABCB1 gene encoding P-glycoprotein (P-gp; MDR1) is the first identified and most well characterized ABC transporter that is overexpressed in MDR tumor cells [132, 133, 134]. Several of the ABCB family (also known as the MDR family of ABC transporters) members are responsible for intracellular peptide transport and confer MDR to tumor cells [135]. Most cells expressing MDR-1 are resistant to important chemotherapy agents including taxanes, doxorubicin and vinca alkaloids among others [136]. Several strategies are being pursued to overcome MDR mechanisms. These include discovery of new anticancer drugs that bypass drug efflux, MDR modulators (chemosensitizers or MDR inhibitors), monoclonal antibodies, multifunctional nanocarriers and RNA interference (RNAi) based therapies (Figure 4) [137].

Fig. 4.

Potential approaches to combat multidrug resistance (MDR) in cancer stem cells. ABC – antigen binding cassette, RNAi – RNA interference, miRNA – microRNA, ASO – antisense oligonucleotides, CRISPR/Cas9 – Clustered randomly interspaced short palindromic repeats/Cas9

Glioblastoma stem-like cells (GSCs) are a major contributing factor to poor response in glioblastoma treatments. Torres et al. investigated the efficacy of tacrolimus as a regulator of multidrug resistance-associated protein 1 (MRP-1) protein and improved the efficacy of vincristine treatment by attenuating MRP1-mediated chemoresistance in GSCs [138]. A carboxyfluorescein diacetate (CFDA)-retention assay was used to assess MRP1 activity in GSCs and normal glioblastoma cells from U87MG (human) and C6 (rat) glioma cell line, while MTT and apoptosis were utilized to evaluate the cytotoxicity of vincristine treatment. Vincristine suffers from poor permeability across BBB because of P-gp-mediated drug efflux. Tacrolimus was seen to promote apoptosis, decrease MRP-1 activity in GSCs, and improve chemosensitization towards vincristine treatment in GSC derived tumors. Minko et al., reported that N-(2-hydroxypropyl) methacrylamide (HPMA) copolymer-adriamycin conjugate inhibited the expression of the MDR1 and β₂m genes in multidrug resistant A2780/AD cells [139]. Other block co-polymers such as Pluronic P85 were studied by Alakhov et al., for their chemosensitizing effect on multi-drug resistant ovarian cancer cell lines. The cytotoxic effects of other P-gp substrate drugs (doxorubicin, epirubicin, vinblastine, and mitomycin C) increased by a factor of 20−1000 and of non-substrate drugs (methotrexate and cisplatin) increased by a factor of 2–5.5 in the presence of the co-polymer [140]. In a study reported by Venne et al., Pluronic L61 increased the sensitivity of multidrug resistant CH^RC5 Chinese hamster ovary cells and MCF-7/ADR human breast carcinoma cells to the cytotoxic action of doxorubicin [141]. Taxanes are unable to kill cancer cells that overexpress P-gp. As taxanes work by stabilizing microtubule assembly resulting in cell arrest before anaphase, they must accumulate inside the tumor cells at sufficiently high concentrations to kill the cells. Several studies have shown that administering paclitaxel with P-gp inhibitors can be effective in increasing intracellular paclitaxel concentrations [142, 143]. Curcumin, a natural polyphenol has been widely studied for its chemoprotective and anti-cancer properties. One of the ways curcumin helps fight cancer is by downregulating intracellular levels of P-gp, MRP-1 and ABCG2. As a result, therapies combining various anti-cancer drugs with curcumin have been studied. One of the drawbacks with use of curcumin is its high hydrophobicity (particularly important if the anti-cancer drug in question is also hydrophobic) which limits efficient loading and release from delivery systems. Ganta et al. developed a flaxseed oil nanoemulsion to effectively load and deliver paclitaxel and curcumin in combination to wild-type SKOV3 and drug resistant SKOV3_TR human ovarian adenocarcinoma cells [144]. Another study investigated curcumin delivery to sensitize breast CSCs to mitomycin C. Curcumin also improved the sensitivity of cisplatin, paclitaxel and doxorubicin in MDAMB-231 and MCF-7 cells (>2-fold reduction in half-maximal inhibitory concentration) by reducing the expression of ABCG2 [145]. ABCB5 has also been shown to be implicated in chemoresistance of CSCs. Frank et al. demonstrated that targeting ABCB5 using a specific blocking monoclonal antibody reestablishes sensitivity to doxorubicin in melanoma cells [146]. In another report epirubicin-loaded lipid microbubbles conjugated with anti-ABCG2 monoclonal antibody was shown to induce apoptosis in myeloma stem cells [147]

Nanoparticles are an efficient platform to target MDR tumors due to their passive and active targeting capabilities. Rapidly growing tumors have a leaky vasculature and poorly developed lymphatics that allows for the EPR effect. Nanoparticles may also avoid drug efflux by ABC transporters as they are internalized via either non-specific or specific endocytosis, resulting in a higher intracellular accumulation of the drug. Chen et al. studied poly(lactide-co-glycolide)/d-alpha-tocopherol polyethylene glycol 1000 succinate (PLGA/TPGS) nanoparticles loaded with doxorubicin and elacridar for targeting liver CSCs and bulk cancer cells. Elacridar, an acridone imidazole amide derivative inhibits ABCB1 and ABCG2. Synergistic cytotoxic effects with nanoparticles containing elacridar and doxorubicin (molar ratio 1:1) were demonstrated in HepG2 cells and HepG2 tumor spheroid model. Nanoparticles targeted the tumor better and showed high antitumor efficacy and low systemic toxicity in vivo [148]. Chemotherapy combined with ABC inhibition is an effective way to target both CSCs and the bulk of tumor cells. Specific inhibitors of these transporters have been established and more are being continuously developed [149]. However, an effective drug delivery system may be needed for efficient targeting of ABC transporters on CSCs. Since these transporters are also responsible for maintenance of function in normal stem cells and other mission critical barriers, their doses must be carefully titrated to limit any possible side effects.

Genetic approaches to target ABC transporters include using RNA interference (RNAi) and antisense oligonucleotides (ASO). RNAi is a powerful biological process that has been used to target MDR tumors. RNAi works by silencing specific gene sequences by destroying targeted mRNA molecules. RNAi based MDR modulation in cancer was first attempted in 2003 where synthesized siRNAs were used for MDR modulation by transient downregulation of P-gp expression. Systemic delivery of naked siRNA, however, is inefficient because of rapid degradation of the siRNA and thus siRNA is often delivered using nanocarriers. Kato et al. delivered small interfering RNA (siRNA) for O6-methylguanine-DNA methyltransferase (MGMT) encapsulated in cationic liposomes. Liposome mediated delivery bypassed the ABC transporters, thereby overcoming drug resistance conferred by these transporters [150]. Hashemitabar et al. developed an ABCG2-targeted aptamer complex that increased the efficacy of doxorubicin against ABCG2 overexpressing, mitoxantrone-resistant breast cancer cell line (MCF7/MX) [151].

MicroRNAs (miRNAs) have been found to play a role in tumorigenesis. microRNAs are regulatory noncoding short RNA sequences capable of altering gene expression. miRNAs contribute to drug resistance by modulating the expression of the genes coding for drug transporters. MicroRNA regulation has directly been correlated with chemosensitivity in HCC. Reduced miR-33a expression, which directly targets ABCA1 expression, leads to chemoresistance in HCC [152]. Overexpression of miRNA-328 was seen to increase the sensitivity of breast cancer cells to mitoxantrone by downregulating the BCRP transporter [153, 154]. In MCF-7 breast cancer cells, expression of miRNA-21 decreases the responsiveness of cancer to doxorubicin treatment. Similar effect was seen with trastuzumab in breast cancer. Role of miRNAs is further observed in a wide range of tumor types including colorectal, ovarian, prostate, lung, and hematopoietic cancers. Use of ASOs and ribozymes to overcome MDR have also been reported with limited efficacy. ASOs are small DNA sequences designed to be complementary to the genetic sequence of interest. However, ASOs were incapable of completely reversing P-gp in cell lines overexpressing high levels of P-gp [155]. The ribozyme approach also suffers from similar problems [156]. Over the last five years, genome editing using CRISPR/Cas9 has been investigated for targeting MDR mediated by ABCB1 and has shown potential, but further studies are needed to demonstrate the translational potential of this approach [157, 158, 159].

Since CSCs have various pathways to survive toxic drug concentrations, only ABC inhibition is not enough to accomplish complete tumor eradication. Further, quiescence makes CSCs less sensitive to conventional therapies, with cells being resistant to antitumor drugs even in the absence of drug transport inhibitors. Also, it is important to maintain the efficacy-safety balance by limiting any undue toxicity to normal stem cells.

3.2. Targeting surface markers

While using efflux pump inhibitors is a promising strategy to sensitize CSCs to chemotherapeutics, a more targeted approach to kill CSCs is to exploit the presence of specific surface markers on these cells. Delivery systems that utilize an antibody targeting a surface marker on CSCs have shown promising activity [160]. Existence of a molecular link between stem cell markers such as MDR-1 and the hyaluronan receptor, CD44, has been shown in breast and ovarian cancers [161]. Thus, hyaluronic acid (HA)-based drug conjugates have been investigated for targeting CSCs [162]. Once HA binds with the CD44 receptor, the receptor–polymer system is internalized, followed by degradation of HA. Further, because HA is a hydrophilic polymer, HA conjugation with anticancer drugs may improve their solubility [163]. The earliest study reporting the use of HA targeting was in 2001 when Elias et al. demonstrated that HA-targeted liposomes bound to B16F10 murine melanoma cells (which express CD44 at high levels) at greater levels than against CV-1 African green monkey kidney cells, which express low levels of CD44 [164]. Paclitaxel and curcumin loaded PLGA nanoparticles coated with a HA-lipoid were investigated for targeting breast CSCs. The nanoparticle encapsulated drugs synergistically inhibited the growth of both breast CSCs and bulk breast tumor cells in mice bearing MCF7 tumors [165]. Goodarzi et al. developed a CD44 targeted macromolecular conjugate of docetaxel linked to HA via a pH sensitive linkage. HA–docetaxel conjugates showed specific toxicity only in CD44-expressing cells in vitro, along with a decreased risk of neutropenia and dose-dependent mortality in vivo [166]. In addition to HA conjugates, anti-CD44 antibodies and aptamers have also been utilized to target CD44⁺ tumor cells [167–169]. Several approaches to target EpCAM have recently been studied, including the use of antibodies Catumaxomab, MT110 (Micromet GmBH) and Edrecolomab. Notably, MT110 underwent Phase I clinical trial on patients with refractory solid tumors. Dose escalation to potential therapeutic effects could not be achieved due to severity of adverse effects [170]. CD33 is expressed on leukemic blasts in 85–95% of AML patients. Mylotarg® (gemtuzumab ozogamicin, developed by Wyeth and Pfizer) is a small molecule antibody conjugate used as a targeted treatment for CD33⁺ AML [171]. The small molecule component of the drug is an antitumor antibiotic, N-acetyl-gamma calicheamicin while the antibody component is an anti-CD33 hP67.6 antibody. The two components are connected by a bifunctional linker which is cleaved in the acidic lysosomes of the tumor cells releasing calicheamicin. The drug forms a DNA anti-metabolite, resulting in cell cycle arrest and/or apoptosis.

Antibodies targeting specific surface proteins have been used alongside other cancer therapies for effective CSC targeting. Liu and coworkers designed multifunctional silica shell coated iron oxide nanoparticles loaded with alvespimycin (heat shock protein inhibitor and semi-synthetic derivative of the antibiotic geldanamycin) and tagged with anti-CD20 antibody [172]. Combining CSC targeting, chemotoxicity, and hyperthermia, these particles produced a synergistic response in vitro in lung CSCs (compared to individual nanoparticle components). Distribution studies in a metastatic mouse model (with lung CSCs) revealed higher accumulation of CD20 targeted nanoparticles in the lung against particles without anti-CD20 antibody. Further investigations in a mouse xenograft model showed dramatically inhibited tumor growth and no apparent regrowth (with hyperthermia + multifunctional nanoparticle treatment). In another study, CXCR4-targeted self-assembling toxin nanoparticles were investigated for their potential to selectively kill CXCR4⁺ human colon-CSCs compared to 5-fluorouracil and oxaliplatin treatments [173]. CXCR4 plays a critical role in the poor response to chemotherapy as CXCR4 signaling regulates the expression of CD20 on B cells. These studies revealed selective CSC targeting with CXCR4-targeted self-assembling toxin nanoparticles and pyroptosis in the absence of apoptosis, demonstrating the potential for this approach.

CD133 is another commonly targeted epitope found on the surface of CSCs in many cancers. Majority of anti-CD133 antibodies only recognize the glycosylated epitope of the CD133 protein, potentially limiting their reliability. We developed a scFv that could recognize both glycosylated and non-glycosylated forms of human CD133 [84]. The scFv immunostained Caco-2 cells (CD133⁺) but did not stain U87 cells (CD133^−/low). Notably, the anti-CD133 scFv possessed ability to cross-react with mouse CD133 protein. This was demonstrated by co-immunostaining studies in mouse bone marrow cells, using anti-CD133 scFv-FITC and anti-mouse CD133-PE (clone 13A4) commercial antibody. Given their smaller size than the parental monoclonal antibodies, scFvs are an excellent choice for targeted drug delivery to CSCs [174]. Using the anti-CD133 scFv, our team developed a deimmunized immunotoxin (CD133KDEL). This CSC-targeted immunotoxin selectively inhibited head and neck squamous cell carcinomas (SCC), which are rich in CD133⁺ cells. Complete eradication of tumors in immunocompetent mice was observed with CD133KDEL [175]. Further, systemic delivery of CD133KDEL was effective in inducing tumor regression in a mouse metastatic, intrasplenic MDA-MB-231 tumor model.[176] Using the CD133 scFv as the platform, our team then developed a self-sustaining trispecific NK cell engager (TriKE) to trigger natural killer cell-mediated antibody-dependent cellular cytotoxicity (ADCC). The TriKE molecule contains an anti-CD16 scFv to induce NK cell activation and an IL-15 encoding cross linker that increases anti-cancer reactivity [177]. We also developed a novel anti-CD133 monoclonal antibody (hybridoma clone 7) with highly immunogenic amino acid residues selected from the native CD133 protein as the immunogen [77]. This antibody was able to specifically recognize CD133 protein in a range of immunological applications. We further investigated targeted paclitaxel delivery using anti-CD133 antibody conjugated PLGA nanoparticles. CD133 targeting resulted in 9-fold higher uptake of nanoparticles in Caco-2 cells, which express high levels of CD133. In vivo studies in MDA-MB-231 xenograft mouse model demonstrated significantly lower tumor regrowth with CD133 targeted nanoparticle treatment compared to that with free paclitaxel treatment [178]. Shigdar et al. identified RNA aptamers that specifically bind to CD133. They reported effective internalization of the RNA aptamers and demonstrated superior penetration into 3D tumor spheres [179]. Ma et al. prepared a CD133 aptamer modified docetaxel liposomal formulation and demonstrated improved therapeutic efficacy and reduced systemic toxicity in A549 lung cancer model [180]. Kim and coworkers demonstrated improved therapeutic efficacy with temozolomide encapsulating liposomes that were conjugated to both angiopep-2 peptide and CD133 monoclonal antibody and targeted to CD133⁺/ALDH1⁺ glioblastoma stem cells [181].

Overexpression of cell surface protein, perlecan, has been demonstrated in TNBC. Using phage display, we developed two human antibodies (clone 6 and AM6) that could bind to metastatic TNBC tumor cells. The affinity-matured antibody inhibited the growth of MDA-MB-231 tumors to a greater extent in nude mice than in NSG mice, suggesting natural killer cell-mediated antibody-dependent cell cytotoxicity. Development of therapies targeting perlecan could improve the management of metastatic TNBC. We further synthesized paclitaxel loaded PLGA nanoparticles covalently conjugated with Clone 6 and AM6 (discussed earlier) to investigate anti-tumor efficacy in TNBC [182]. Perlecan-targeted nanoparticles exhibited improved cell uptake, retention and cytotoxicity in vitro. In vivo studies in MDAMB-231-LM2 orthotopic mouse tumor model demonstrated enhanced tumor growth inhibition with AM6 nanoparticles against nanoparticles functionalized with isotype IgG.

3.3. Targeting signal cassette

Numerous signaling pathways in CSCs play specific roles in proliferation, differentiation, stemness induction and maintenance, self-renewal, and therapy resistance [183, 184, 185]. Many highly conserved signaling pathways, including the Hedgehog, Notch and Wnt/β-catenin pathways are deregulated in CSCs (Table 2) [112, 186]. Due to the importance of signaling pathways that drive oncogenesis and tumor metastasis, selective targeting of CSC signal cassette is of utmost clinical significance for drug discovery and development [184, 187]. Various ways to target CSC signaling pathways are presented in Figure 5.

Table 2.

CSCs signal pathway targeting

Pathway	Agents	Delivery system	Reference
Hedgehog pathway	Anthothecol	PLGA nanoparticles	[202]
	Cyclopamine	System based on HPMA	[203]
Notch pathway	γ-secretase inhibitor (GSI)	Imageable mesoporous silica nanoparticles (MSNPSs); engineered glucose functionalized MSNPs	[210, 211]
	Platinum (IV) prodrug and siRNA	Micellar nanoparticles (MNP)	[212]
Wnt pathway	Salinomycin	Self-assembled nanoparticle system based on iTEP	[221, 222]
	Peptides targeting Wnt/LRP5/6	Small magnetic iron oxide nanoparticles (IONP)	[223]
Epithelial-mesenchymal transition	miR-200c	Gelatinase-responsive nanoparticles	[225]
Multiple signaling pathways	Graphene oxide		[226]
	Ti0.8O2 nanosheets		[227]

Fig. 5.

Potential ways to target signaling pathways in cancer stem cells

3.3.1. Hedgehog pathway

The Hedgehog (Hh) pathway maintains the complexity and function of stem and progenitor cells from embryonic development to the upkeep of fully grown adult organs [188, 189]. This signaling pathway has been widely reported to regulate CSCs in various cancer types [190, 191, 192, 193, 194, 195, 196, 197, 198]. It is likely that Hh signaling acts through multiple modes and is involved in interactions between CSCs, differentiated tumor cells, and the microenvironment [189]. Studies also indicate that Hh signaling is activated in human tumor samples and irradiated mouse xenografts that are resistant to radiation [185]. Hh ligands may also support the repropagation of residual tumor cells after radiation therapy [199]. Furthermore, inhibition of Hh signaling has been shown to reduce CSC stemness and chemoresistance [200].

A few small-molecule Hh inhibitors, including the FDA-approved Vismodegib and Sonidegib, are selective antagonists of Hh pathway that act by binding to Smoothened (Smo) receptor. They have been investigated in clinical trials but their usage is limited by weak Smo receptor binding (due to Smo bind site mutation) and poor systemic bioavailability [187, 201]. Therefore, various drug delivery systems have been employed to deliver new generation of antagonists to target Hh pathways, especially in the downstream of Smo. For example, anthothecol, a limonoid derived from plant Khaya anthotheca (Meliaceae), was encapsulated in PLGA nanoparticles (Antho-NPs) to improve its efficacy against pancreatic cancer cell lines. The results indicated that Antho-NPs inhibited CSC growth via the Hh signaling pathway [202]. Yan Zhou etal, developed a delivery system based on N-(2-hydroxypropyl) methacrylamide (HPMA) to enhance drug solubility and reduce the systemic toxicity of cyclopamine, a Hh inhibitor. This system showed anti-CSC effectiveness in RC-92a /hTERT cells as assessed by reduced expression of stem cell markers and viability of CSCs [203].

3.3.2. Notch pathway

Notch signaling is believed to regulate both normal stem cell and CSC function [204]. Inhibition of the Notch pathway is expected to significantly diminish self-renewal, clonogenicity, and tumorigenicity in CSCs, leading to a reduction in CSC-like subpopulations and increase in the sensitivity of CSCs to radiation-induced apoptosis [205]. In primary and metastatic tumors, Notch participates in tumor stroma and tumor endothelium interactions [206, 207,208]. Notch also interacts with multiple transcription and growth factors that participate in EMT, including snail, slug, and TGF-β [206]. Due to a significant interest in selectively targeting CSCs with novel therapeutics, Notch is one of the most intensively researched therapeutic, with several Notch inhibitors and delivery systems currently under investigation [208, 209].

Clinical use of Notch inhibitors is limited by severe side effects, thus, targeted delivery approaches have been evaluated for these molecules. Mamaeva et al. used imageable mesoporous silica nanoparticles (MSNPs) as vehicles for targeted delivery of γ-secretase inhibitors (GSIs), potent interceptors of Notch signaling [210]. Since CSC self-renewal is promoted by Notch signaling in breast cancer, they display high glycolytic activity and aggressive hormone-independent tumor growth. The same research group further utilized CSCs’ glycolytic phenotype and reliance on Notch activity, and engineered glucose functionalized MSNPs loaded with a γ-secretase inhibitor to selectively target CSCs. The data showed that these nanoparticles can be efficiently internalized by CSCs in vitro and in vivo [211].

Notch inhibition can also be used to sensitize tumors to other chemotherapeutic agents. Shen and his group developed a micellar nanoparticle (MNP) formulation to deliver platinum (IV) prodrug and siRNA targeting Notch1. They observed that siRNA-mediated suppression of Notch1 improved the sensitivity of hepatocellular carcinoma (HCC) cells to platinum drugs and reduced the proportion of HCC CSCs, resulting in improved inhibition of proliferation and induction of apoptosis in HCC cells in vitro [212].

3.3.3. Wnt pathway

The Wnt pathway controls a set of genes that are important for different biological functions. Canonical Wnt/β-catenin deregulation has been observed in various kinds of cancer, including colon, liver, lung, breast, ovarian and leukemia, making it an appealing target [213]. Wnt/β-catenin signaling has been speculated to play a key role in CSC development [198, 213, 214, 215, 216, 217]. Jang’s group found that Wnt/β-catenin regulates self-renewal and migration of CSCs to promote breast cancer development as well as metastasis/systemic dissemination. The authors suggest that Wnt/β-catenin signaling could serve as a novel target for breast CSCs (BCSCs) [218]. In order to inhibit BCSCs self-renewal, Huang et al. developed self-assembled PLGA/HA block copolymer-based nanocarriers and delivered a β-catenin inhibitor, sulforaphane (SFN) to silence the Wnt pathway in BCSCs. The results showed that SFN-loaded nanoparticles were more effective than free SFN in inhibiting BCSCs. They also found that combination therapy with docetaxel (DTX)- and SFN-loaded nanoparticles could strongly inhibit BCSC self-renewal and demonstrated targeting differentiated breast cancer cells and BCSCs simultaneously [219]. Zhi et al. found that Wnt signaling is activated in HCC, with decreased βII-Spectrin (SPTBN1, the most common non-erythrocytic member of the β-spectrin gene family). Their data indicated that loss of SPTBN1 activates Wnt signaling, promotes acquisition of stem cell-like features, and ultimately contributes to malignant tumor progression [220]. Zhao et al. also conjugated an immune tolerant elastin-like polypeptide (iTEP) to salinomycin, a Wnt/β-catenin signaling pathway inhibitor, and constructed a self-assembled nanoparticle system, which displayed significant tumor growth reduction in a 4T1 orthotopic breast cancer model [221, 222]. To efficiently treat resistant tumors, Miller-Kleinhenz et al. developed a small magnetic iron oxide nanoparticle (IONP) drug carrier combined with peptides that dually targeted Wnt/LRP5/6 and urokinase plasminogen activator receptor (uPAR). Wnt/β-catenin and cancer stem phenotype of tumor cells were inhibited by the dual receptor directed IONPs, leading to greater tumor growth inhibition [223].

3.3.4. Epithelial-mesenchymal transition

Recent studies suggest that EMT plays a significant role in disease progression and generation of distant metastasis in many solid tumors [108, 109]. EMT is a regulatory development program essential for the creation and maintenance of CSCs [110]. Acquisition of CSC characteristics is strongly linked to the progress of EMT transdifferentiation [111]. EMT is also crosslinked with multiple signaling pathways. Wnt and Hh pathways can operate together to control cell behavior across epithelialmesenchymal boundaries and are significant EMT mediators in tumor progression [185]. Inhibition of EMT as a means to target CSC has been established as a promising therapeutic strategy [224]. Liu et al. developed gelatinase-responsive nanoparticles to co-deliver CSC inhibitor miR-200c and docetaxel, to inhibit both CSCs and non-CSCs. This system showed significant inhibition of EMT characterized by greater E-cadherin expression and reduced CD44 expression. The synergistic delivery of miR-200c and docetaxel inhibited tumor growth in a synergistic manner [225].

Novel materials have recently been discovered to target CSCs via inhibition of multiple signal pathways. For instance, according to a study published by Fiorillo and coworkers, graphene oxide can selectively inhibit the proliferative expansion of CSCs through inhibition of different signal transduction pathways, including Wnt, Notch, STAT1/3 and NRF-2 [226]. Petpiroon et al. found that Ti_0.8O₂ nanosheets exhibit similar properties and have promising lung CSC suppression potential with minimal cytotoxicity to normal stem cells. This is due to Ti_0.8O₂ nanosheet treatment attenuating CSC properties through the suppression of Akt/GSK3 pathway, which in turn suppresses Akt/GSK3β/β-catenin and Notch1, as well as EMT [227].

3.4. Targeting tumor microenvironment or tumor niche

Because CSCs rely on unique niche factors, targeting the CSC niche in tumors or interfering with the CSC-niche interactions may improve treatment efficacy [98]. For example, the niche transforms into a supportive environment for leukogenic leukemia stem/progenitor cell (LSPC) survival and expansion protecting them from chemotherapy. Gul-Uludag et al. developed lipid-substituted polyethyleneimine/siRNA complexes to deliver siRNA into difficult-to-transfect myeloid leukemia cell lines. The polymeric nanoparticle-mediated silencing was able to inhibit LSPCs’ interactions with their microenvironment, resulting in the inhibition of leukemia progression [228].

The rarity of CSC population, tumor heterogeneity and dense ECM network are major challenges to intra-tumoral penetration of drugs and drug delivery systems [229, 230]. Rather than being trapped in the tumor matrix just outside tumor vessels, the drug carriers must demonstrate good tumor accumulation and tumor penetration to access CSCs [104, 230, 231, 232]. For instance, co-treatment with a TGF-β inhibitor resulted in enhanced vascular leakage and improved tumor penetration of siRNA loaded nanoparticles, followed by internalization by CSCs [104]. Deep tumor penetration can also be achieved by bioinspired strategies. Ferritin is an endogenous iron storage protein, which presents as a cage-like structure that can load a wide range of therapeutic agents [233, 234]. It was found that CSCs in brain tumor may preferably require ferritin to promote in vivo propagation and tumorigenesis [235]. By using this attribute, Tan et al. developed tumor-penetrating, biomimetic nanocages as a promising drug nanocarrier to treat metastasis and improve access to CSC fractions in the tumor [236].

Sun et al. proposed that the extra domain B of fibronectin (EDB-FN), linked to tumor angiogenesis, could be a new biomarker for BCSC detection due to a high expression of EDB-FN in breast cancer stem cell line NDY-1 cells compared to other cells [237]. They also investigated a new liposomal system (APTEDB-LS-siRNAEDB) for EDB-FN targeting and knockdown. According to the report, this system presented potent therapeutic efficacy in the BCSC-derived tumors in vivo. The siRNA-mediated EDB-FN knockdown led to lower expression of VEGF in endothelial cells [238].

It is known that CSCs are often located within hypoxic niches and HIFs play an important maintenance role. HIF inhibitors targeting CSCs can enhance the efficacy of angiogenesis inhibitors and cytotoxic chemotherapy. By using decoy oligodeoxynucleotide (ODN) and lentivirus-siRNA delivery system, Zhu et al. showed that downregulation of HIF-1α reversed chemoresistance and induced apoptosis in MDA-MB-231 cells, majority of which were CD44⁺CD24^−/low BCSCs [239]. A metallofullerenol nanomaterial, Gd@C₈₂(OH)₂₂, was recently reported to have inhibitory activity against TNBC cells by blocking EMT, with resultant efficient elimination of BCSCs. Under hypoxic microenvironment, cellular uptake of Gd@C₈₂(OH)₂₂ increased and the nanomaterial was found to be a potent HIF-1α and TGF-β inhibitor [240].

3.5. Targeting metabolism

In recent years, metabolism has been shown to play a vital role in supporting CSC phenotype. Unlike bulk cancer cells, CSCs are thought to have a metabolic phenotype that is comparable to ordinary stem cells [241]. The metabolic singularities and the impact of diverse metabolites on CSC physiology have not been fully explored. However, some studies have showed that compared with the bulk tumor, CSCs have a distinctive metabolic phenotype. According to some studies, CSCs appear to bear a differentiated metabolic phenotype in different types of cancers and are dependent on oxidative phosphorylation (OXPHOS) and/or mitochondrial function [242, 243, 244, 245, 246]. Reports of metabolic flexibility of CSCs, including glycometabolic pattern and mitochondrial function, have also been published [247, 248]. By using the metabolic characteristics, smart and effective drug delivery systems could be designed to target therapeutic agents to CSCs. Based on the overexpression of glucose transporter 1 (GLUT1), glucose-installed sub-50-nm gold nanoparticles (Glu-NPs) were fabricated by Yi et al. to deliver siRNA to achieve sequence-specific gene silencing in BCSCs [249].

Effective and preferably selective distribution of drug to mitochondria in cancer cells is a significant consideration in designing a new CSC targeting strategies. CSCs that rely on OXPHOS demonstrate higher mitochondrial mass as well as higher mitochondrial membrane potential compared to non-CSC [243, 250]. Complex drug delivery systems or nanocarriers can be designed to establish drug accumulation in mitochondria. Studies have shown that mitochondria-targeting liposomes can enhance ROS generation, induce morphological and functional changes, enhance apoptosis or necrosis, or prevent the development of resistance [251, 252, 253, 254]. This strategy can also lead to mitochondrial drug accumulation, pro-apoptotic protein recruitment and activation, mitochondrial permeability transition pore (mPTPs) opening and mitochondrial inner membrane depolarization, eventually resulting in the activation of apoptotic machinery [255].

Previous studies have reported that following radiation exposure, reduced levels of ROS in CSCs are related to lesser DNA damage and a greater rate of cell survival [256–258]. This not only maintains CSC stemness but also confers treatment resistance [259, 260, 261, 262]. Increased ROS production and decreased oxidative defense can cause CSCs to pharmacologically lose their stemness. Reduction in ROS levels has the potential to be used as a combination therapy with standard chemotherapy [263, 264, 265, 266, 267, 268, 269, 270, 271, 272]. Notably, a combined therapy of demethoxycurcumin (DMC) and temozolomide (TMZ), which resulted in modifications in various cell signaling pathways including ROS generation and caspase-3 signaling of mitochondria-related apoptosis activation as well as inactivation of the JAK / STAT3 signaling pathway, was regarded an efficient anti-glioma stem cell treatment [273].

Recently, metal complexes containing endogenous metals and metallopeptides as well as metal-containing drug carriers have been found to have the ability to kill CSCs by targeting mitochondria, and elevating ROS levels [274]. Silver nanoparticles (AgNPs) have been reported to have a significant effect on cell viability, ROS generation and LDH leakage in A2780 ovarian cancer cells (and four different subpopulations of CSCs derived from A2780 cells). Further, ALDH⁺/CD133⁺ cells were more sensitive to the treatment than bulk cells. AgNPs have the capacity to increase ROS generation and oxidative stress. Thus, AgNPs were found to have the potential as CSCs-targeted therapeutic agents [275, 276]. Laws et al. developed a metallopeptide that can trigger mitochondrial damage in BCSCs [277]. Purushothaman et al. reported a Ru(ii) complex that can target mitochondrial and endoplasmic reticulations to effectively eliminate CSCs in MCF-7 breast cancer cells [278]. Stilgenbauer et al. developed a spermine-conjugated lipophilic Pt(IV) prodrug, which can accumulate Pt(IV) in mitochondria, reduce CSC population, and overcome drug resistance in ovarian cancer [279].

Another evolving idea in the field of CSC metabolism is that compared to the bulk cancer cells, CSCs rely highly on lipid metabolism for maintaining stemness and fulfilling energy demands [280]. The dysregulated lipid metabolism in CSCs is a result of upregulation of FA synthesis, extracellular lipid uptake, intracellular lipid droplet accumulation and β-oxidation. The increased lipogenesis in CSCs is linked to modifications in the levels of key lipogenic regulators, which could be exploited for therapeutic CSCs targeting [280, 281]. For example, CSCs are highly dependent on the operation of lipid metabolic enzymes such as 3-hydroxy-3-methylglutharyl-coenzyme A reductase (HMG-CoAR) and stearoyl-CoA desaturase 1 (SCD1). SCD1 can enhance the lipid-modified Wnt pathway and lead to both β-catenin and YAP/TAZ release. Following this, β-catenin and YAP/TAZ translocate and interact with their transcriptional partners, triggering the reprogramming of cancer cells to CSCs. Similarly, HMG-CoAR is a key point in the mevalonate pathway, a cascade culminating in the production of farnesyl pyrophosphate (FPP) and geranylgeranyl pyrophosphate (GGPP). This maintains correct membrane anchoring of Ras homologous family small guanosine triphosphatases (RHO GTPases), an essential enzyme for CSC stemness. Further, SCD1 also activates the NF-kB pathway, increases the expression of lipid desaturases and feeds a positive feedback loop. Thus, targeting of SCD1 and HMG-CoAR could be a promising therapeutic approach [282]. Finally, lipids play an important role in regulating membrane P-gp function, membrane trafficking, drug transport and endocytic function; changes in lipid biosynthesis in cancer can influence drug transport and endocytic functions [283]. Softening of the cell membrane increases drug resistance, migration and metastatic potential. Engineering the biophysical properties of the cancer cell membrane such as altering stiffness by phosphatidylcholine liposome treatment promoted caveolae recruitment and increased dynamin-dependent uptake of nanoparticles [284]. Hence, lipid-related drug delivery strategies are another approach to overcome drug resistance. In addition to these, many studies have focused on other metabolic characteristics in CSCs and revealed peculiar metabolic patterns. For example, CSCs are characterized by high cellular iron content, and enhanced iron levels correlate with sphere forming ability as well as with significant modifications in stem cell marker expression that can be reversed by iron removal [285].

3.6. Autophagy activation

Autophagy is an evolutionarily conserved mechanism that CSCs utilize to protect themselves against environmental stresses arising from hypoxia, nutrient deficiency, radiation, and chemotherapy [286, 287, 288, 289, 290]. As an adaptive mechanism, autophagy plays a crucial role in the survival, self-renewal, differentiation, and tumorgenicity of CSCs [30].

Accumulating evidence confirms the dual function of autophagy, as either a cancer inhibitor or promoter, in diverse settings and various tumor types [272, 291]. For instance, some research suggests that autophagy could induce apoptosis in breast and pancreatic CSCs through different molecular mechanisms [292, 293]. On the other hand, autophagy may enable cancer cells to survive multiple environmental stresses or promote carcinogenesis by influencing cell signaling or inducing intracellular toxicity [30]. In addition, autophagy signaling is linked to EMT, which has strong correlation with invasion and metastasis [294, 295, 296]. In some tumor types, autophagy suppression could inhibit the migration of CSCs, whereas its upregulation might restore the mesenchymal phenotype [297, 298]. Furthermore, it has been shown that the role of autophagy in regulating CSC resistance to various therapeutics is context- and cancer type-dependent [299, 300]. Some studies have shown that the combination of autophagy inhibitors and anticancer drugs could enhance the susceptibility of CSCs to chemotherapeutics [301].

Clarifying the relationship between autophagy and stem cell-like properties in various types of CSCs requires more thorough investigation. Nevertheless, application of nanoparticulate delivery systems to target autophagy in CSCs is an exciting new avenue for cancer treatment. For example, Sun et al. showed that nanocarrier-mediated autophagy inhibition markedly reduced the ‘stemness’ of human BCSCs and enhanced their susceptibility to chemotherapeutics [301]. In addition, combining AgNPs with salinomycin was found to exert a synergistic effect in inducing apoptosis and autophagy in ovarian cancer cells [302]. Furthermore, Lu et al. found that an albumin nanoparticle delivery system loaded with paclitaxel and autophagy inhibitor chloroquine, could effectively cross blood brain barrier to target glioma cells, induce apoptosis in glioma cells, and increase the sensitivity of glioma stem cells to paclitaxel [303]. Similarly, Chang et al. developed a formulation, termed 188Re-liposome, to effectively suppress the function of CSCs by targeting autophagy and mitophagy in ovarian cancer [304].

3.7. CSCs-targeted immunotherapy

Based on greater immunological characterization of CSCs and better understanding of interactions between cancer tissue and the immune system, immunotherapy, especially immunological cytotherapy, shows great promise in the treatment of cancer. In vitro generated dendritic cell (DC) vaccines have long been known for their ability to induce effective anticancer immune response. Selection of the tumor antigen and the transfection method to introduce the antigen in the DCs are critical determinants in DC-based vaccines [305]. In one previous study, DCs were pulsed with colon CSCs lysates, which acted as a potent antigen to target CSCs in colon cancer [306]. Interestingly, Szaryńska et al. found that the maturation and activation of DCs was influenced by the detailed configuration of CSC-like markers after cancer cell lysate or culture supernatant stimulation [307].

NK cells can selectively identify and lyse CSCs. Steven and coworkers found that CSCs expressing high amounts of the CSC-associated proteins like ALDH, CD133, CD24, or CD44 can be killed by activated allogeneic NK cells [308]. However, CSCs have been observed to evade NK cell mediated cytotoxicity by shedding NK cell ligands such as major histocompatibility complex (MHC) class I chain-related protein A (MICA) and MHC class I chain-related protein B (MICB). Hence, one approach to improve NK cell-mediated CSC elimination is by using CSC-epitope-specific monoclonal antibodies [309].

T cells play a central role in many immunotherapy strategies including inhibition of immunosuppressive cells (T_regs), adoptive T-cell therapy (ACT), and T cell activation through immunogenic cell death induced by oncolytic virotherapy (OVT), among others. [310]. Over the past decade, engineered CAR T-cells have showed enormous potential in tumor treatment, including targeting CSCs. CAR T-cells have been shown to eradicate CSCs selectively in several tumor models [311, 312, 313]. The blockade of immune checkpoint inhibitors, such as PD-1/PD-L1 and CTLA-4 has also shown to decrease CSC escape from immune system, thus enhancing the eradication of CSCs [314, 315, 316, 317].

4. Problems and prospects

4.1. CSCs isolation and identification

Because CSCs, normal stem cells and cancer cells share many attributes, there have been several attempts to develop methods to specifically isolate and identify CSCs [100, 318]. Serial colony-forming unit assays (replanting assay), label-retention assays and sphere-formation assays are commonly used in vitro assays to identify CSCs [319, 320, 321], while side population method, cellular markers method and establishment culture methods are utilized for CSCs isolation [322]. While each of these methods have advantages, there are also potential pitfalls that complicate interpretation of the results [321, 322]. Thus, failure to adequately recognize and isolate CSCs is still a significant obstacle to specifically target CSCs. For example, in some cases, cancers believed to have only rare tumorigenic cells may actually have common tumorigenic cells [323]. As a result, it has been suggested that multiple techniques, applied in combination or in sequence, should be used for CSC identification and isolation [321, 322, 323, 324]. In addition, efficacy of anti-CSCs drugs should also be evaluated by more relevant methods. The efficacy of CSC therapy is related more to the tumor initiating ability than the tumor size. Molecular imaging technologies could be valuable tools to observe the CSC populations [325].

4.2. In vitro CSC models

An important future direction for oncotherapeutic research is developing correlative in vitro models to improve our understanding of tumor progression. Due to the difficulties in studying CSCs through in vivo models due to high costs and ethical concerns, there is a clear need for efficient and predictive in vitro model systems and in vitro screening methodologies to study CSC targeting. One of the most widely used methods for CSC isolation is detection of CSC specific markers such as CD133, CD44, EpCAM, Nanog, Oct4, Sox2, ALDH-1, CXCL12, ABCG and bmi-1. CSCs could be sorted and separated from existing tumor lines or created cell lines by genetic manipulation or defined serum-free conditions [326, 327, 328]. CSCs proliferating in 3D culture systems display physiologically similar tumor characteristics as seen in vivo. These include similar signaling pathways, gene expression, cell–cell interactions, and drug resistance mechanisms. Using tumor spheroids (floating spheres) in vitro has been shown to improve drug efficacy and toxicity predictions. Several other approaches have been established to examine CSC populations. As EMT plays a vital role in tumor progression and drug resistance, it is essential to establish 3D culture models mimicking in vivo settings where reversibility of EMT and its role in tumorigenesis is accounted. Wang et al created a 3D bioprinted glioma cell-laden scaffold to enrich glioma stem cells and found that the stemness and EMT-related gene expression of glioma cells were increased in the 3D platform [329]. Rashidi et al. reported an in vitro model of chemoresistance and stemness in ovarian cancer using the 3D hanging drop spheroid platform [330]. Tumor spheroids can also be arranged by distinctive sources of cells, allowing in vitro studies to closely mimic the in vivo conditions and TME. Multicellular tumor spheroids (MCTS) of MCF-7 cells and the ones enriched with CSCs (eMCTS) were evaluated by Herheliuk et al [327]. The authors found that eMCTS model could improve the anti-CSC drug screening efficiency. A unique synthetic tumor microenvironment mimic system (STEMs) composed of human lung epithelial cells, pulmonary vascular endothelial cells and human marrow-derived mesenchymal stems cells (MSCs) was created for in vitro tumor biology and drug test studies [331]. A key drawback with tumor spheroids is that they do not develop the complex (yet underdeveloped) tumor vasculature. Thus, all transport (drugs, nutrients, gas exchange) occurs only from outside to inside and relies primarily on diffusion. The transport processes in a tumor tissue in vivo, on the other hand, is much more complex with diffusion, convection, and convective diffusion all playing a role. This difference in transport mechanisms is especially important while evaluating delivery systems [332, 333]. The use of microfluidic models may address some of the transport issues, but these systems lack the complexity of tumor spheroids [334, 335]. Since tumor vasculature plays a key role in management of TME, its importance in in vitro tumor models cannot be overlooked. The proximity of tumor cells to perfused blood vessels is essential to receive nutrients and oxygen. Solid tumors, hence, become angiogenic to survive and proliferate. Chemotherapy relies largely on tumor vasculature to deliver drugs to tumor cells. A quantitative understanding of tumor vasculature with in vitro models based on average vessel diameter, vascular density, and the relative vascular volume will allow for improved vascular targeting and selecting appropriate radiation therapies. A hybrid model that incorporates transport complexity of microfluidics, cellular/TME complexity of spheroids, and quantitative tumor morphology could further address the limitations of in vitro systems.

4.3. Role of drug delivery systems in CSC-targeted therapy

Improved understanding of molecular mechanisms related to CSC generation and development, CSC signaling pathways, context-dependent interactions and regulation, as well as other aspects of CSC biology will increase the possibility of designing effective therapeutic regimens [206]. Although more successful targeting and elimination of CSCs could be achieved through identification and elucidation of specific characteristics of CSCs, developing effective drug delivery platforms is still an important objective for opening up new possibilities for therapeutic intervention and improving the success of current treatments [211, 318, 335]. As discussed previously, drug delivery systems can overcome the limitations of CSC-specific agents, such as poor water solubility, short circulation time and instability and also hold great potential for overcoming the difficulties in CSC elimination (such as poor niche penetration, low drug accumulation in CSCs, radio-/chemo-therapy resistance and detrimental effects on normal somatic stem cells). Smart drug delivery platforms that respond to specific environmental cues can further enhance the performance of anticancer therapy [101, 183, 336, 337]. In addition, formulation characteristics such as particle size, shape and surface modifications could be fine-tuned to design patient specific nanomedicines [336, 338]. Incorporation of more than one agent in the same delivery system could help address the resistance mechanisms utilized by CSCs. For example, incorporation of a drug efflux inhibitor along with an anticancer drug that is a substrate for a drug efflux pump could overcome drug resistance [339]. However, it should be noted that cancer cells in general employ more than one mechanism to evade drug-induced cytotoxicity. Thus, an inhibitor with a wide-ranging mechanism of action is more likely to be successful than those that are focused on any one mechanism. In addition, physical-chemical modalities such as photodynamic and photothermal therapies that induce rapid cell kill via necrosis are less likely to develop resistance than pharmacological approaches [340]. Targeting such modalities specifically to CSCs is challenging but warrants additional investigation.

In summary, CSCs represent a formidable barrier to the success of anticancer therapies in several different cancer types. A combination of strategies that include inhibitors that can either directly kill CSCs or improve the effectiveness of co-delivered therapies as well as formulations that can target such therapies specifically to CSCs can significantly improve treatment outcomes for patients diagnosed with cancer.

Fig. 3.

Link between CSCs, tumor metastasis and epithelial-mesenchymal transition (EMT)