Cancer Stem Cells: The Architects of the Tumor Ecosystem

SUMMARY

Cancer stem cells (CSCs) proactively remodel their microenvironment to maintain a supportive niche. Viewed through the lens of an ecosystem, numerous tumor components have multi-directional interactions involving CSCs, supporting the complexity of tumors to maintain growth in a dynamic host. In this Perspective, we discuss how CSCs are active architects of their microenvironment and drive interactions with other tumor components, such as immune cells, cancer-associated fibroblasts and differentiated cells, blood vessels, and other extracellular cues to engineer a sustainable niche. We also highlight considerations for modeling this dynamic tumor ecology and discuss potential therapeutic strategies for targeting these multifaceted interactions.

INTRODUCTION

Recent efforts in oncology have uncovered extensive tumor heterogeneity, or the presence of tumor cells with variable states of differentiation. These cells diverge in their gene expression, immune interactions, tumor propagation potential, and response to therapy (Al-Hajj et al., 2003; Fang et al., 2005; Li et al., 2007; Ricci-Vitiani et al., 2007; Singh et al., 2003; Zhang et al., 2008)). Explanations for these differences are myriad, including genetic and epigenetic clonal variation, as well as metabolic and microenvironmental features. The cancer stem cell (CSC) hypothesis posits that CSCs are responsible for maintaining this tumor heterogeneity, fueling tumor growth and therapy resistance.

The concept of a cancer stem cell (CSC) has proven challenging to define due to limitations in techniques to purify and characterize CSCs. While dynamic states of cellular differentiation have been modeled in many tissue types, most models of cancer cell differentiation remain stuck in a two-state solution of a CSC and all other neoplastic cells. Lineage tracing experiments support differences in some cancer types that may segregate between long-term and short-term self-renewal, but the application to human samples has proven challenging (Chen et al., 2012; Choi et al., 2012; Driessens et al., 2012; Schepers et al., 2012; Youssef et al., 2010; Zhu et al., 2009). Analogous to attempts at creating a grand unified theory in particle physics, stem cell biologists have sought common rules in the form of markers or functional assays to support a unified theory of defining CSCs, but the variance of the cellular hierarchy in normal development suggests that cancers are unlikely to adhere to simple rules. Defining CSCs (and normal stem cells) remains functional. Furthermore, the CSC field is also hampered by a lack of rigor born from convenience.

While much work remains to further clarify the cellular hierarchy, CSCs appear to serve as critical drivers of tumor heterogeneity and malignancy in many types of solid tumors ((Li et al., 2007). They have been identified at varying frequencies in multiple cancer types, including glioblastoma, breast cancer, colon cancer and, with ongoing controversy, in many additional solid cancers such as ovarian cancer, melanoma, pancreatic cancer, liver cancer and others (Al-Hajj et al., 2003; Bapat et al., 2005; Fang et al., 2005; Li et al., 2007; Ma et al., 2007; Ricci-Vitiani et al., 2007; Singh et al., 2003). CSCs are functionally defined by their ability to self-renew and differentiate, recapitulating the heterogeneity found within a tumor (Valent et al., 2012). CSCs also demonstrate an enhanced capacity for therapeutic resistance, immune evasion, invasion, and metastasis (Balic et al., 2006; Bao et al., 2006; Louie et al., 2010; Todaro et al., 2007)). Thus, efficient targeting of CSCs in cancer treatment is critical for developing effective therapeutics.

One likely contributor to the ongoing controversy surrounding CSC characterization is the dearth of prospective markers. While there are many known markers that enrich for stemness in solid tumors, including CD133 in a wide variety of cancers, Sox2 in glioblastoma, ALDH1 in breast and GI cancers, LGR5 in colon cancer and CD44 in breast cancer, among others, highly sensitive and specific markers for CSCs remain elusive (Al-Hajj et al., 2003; Bao et al., 2006; Barker et al., 2007; Li et al., 2007; Marcato et al., 2011). Notably, the controversy regarding CD271 (p75^NGFR) in melanoma has been particularly passionate (Boiko et al., 2010; Boyle et al., 2016; Civenni et al., 2011; Quintana et al., 2010). This is likely due to imperfect functional assays to interrogate tumorigenicity, the heterogeneity of the stem cell population, and the complexity of the perturbed cellular hierarchy in cancer. Binary characterization of population into “stem” vs. “non-stem,” and attempts to describe CSCs through the lens of normal tissue developmental hierarchies is slowly ceding to a more nuanced understanding of stemness as an emergent, contextual property of interactions with the microenvironment and other cell types, within both the cancer and normal hierarchy. The tumor microenvironment is a critical driver of heterogeneity, plasticity and evolution within the CSC population, which in turn modifies and manipulates different tumor niches (Figure 1). This reciprocal crosstalk is a fundamental and crucial component of tumor growth and evolution, maintenance of stemness and, thus, ultimately, therapeutic resistance.

Figure 1. Intratumoral crosstalk drives tumor microenvironmental heterogeneity.

Increasing inter-niche interactions sustain and grow the tumor microenvironment, creating an increasingly complex ecosystem that harbors the features of a high-grade tumor.

Organizational units of the tumor microenvironment

Tumor microenvironments in solid tumors are often divided into discrete compartments based upon histological characterization and the interactions with noncancerous cells that predominate in each niche (Gilbertson and Rich, 2007; Plaks et al., 2015; Zhu et al., 2011). While some of these niches are spatially distinct (e.g. the perivascular and hypoxic regions), others are distinguished by the type of cellular interactions (e.g. the immune niche), and collectively, they generate a diverse and dynamic tumor ecology. While the specific interactions and compositions of niches vary across tumor type, there is also significant overlap of signaling pathways and phenotypic enrichments in niche-CSC interactions, and aggressive solid tumors are generally characterized by increased niche complexity (e.g. cancer-associated fibroblast accumulation during progression in many cancers). CSCs exhibit different transcriptional and epigenetic signatures depending on the niche from which they derive, and they are also enriched for the capacity to maintain these niches (Anderson et al., 2011; Balic et al., 2006; Bao et al., 2006; Gerlinger et al., 2012; Jin et al., 2017; Li et al., 2007; Puram et al., 2017; Zhou et al., 2015). The existence of differential dependencies within the CSCs of each niche and their capacity to interconvert in a mutually supportive system has significant implications for therapeutic targeting of tumor heterogeneity (Jin et al., 2017; Ye et al., 2016).

Within each niche, CSCs engage in a complex array of reciprocal signaling with both tumor and normal cells and encounter a wide variety of extracellular conditions and cues. Signaling within the hypoxic core of solid tumors, characterized by low oxygen, acidic pH and low nutrient availability, has been shown to drive critical mediators of stemness (Diehn et al., 2009; Flavahan et al., 2013; Hjelmeland et al., 2011; Li et al., 2007). CSCs themselves are highly metabolically adapted to survive in this harsh niche, and the increased affinity of CSCs for nutrients such as glucose allows them to survive and even thrive, promoting cellular proliferation and migration, which in turn expands hypoxia and necrosis (Flavahan et al., 2013; Siebzehnrubl et al., 2013). In the perivascular region, leaky tumor vessels provide a scaffold for migration and preferential access to nutrients, while endothelial cells engage in reciprocal signaling to enhance drug resistance (Charles et al., 2010; Lu et al., 2013; Zhu et al., 2011). In turn, CSCs secrete angiogenic factors and promote neovascularization ((Bao et al., 2006). Structural and signaling cues from normal tissue including cancer-associated fibroblasts and the extracellular matrix, as well as the physical, structural properties of normal tissue architecture support invading CSCs as they migrate to form a leading edge and invasive niche (Giannoni et al., 2010; Liu et al., 2011; Su et al., 2018).

Further characterization of the interactions between cancer-associated immune cells and CSCs represents a rapidly expanding field and is of particularly active interest given the broad expansion of immunotherapeutic options. CSCs are enriched in their capacity for both immune evasion and immunosuppression, creating and maintaining populations of tumor-associated macrophage or regulatory T cells to inhibit immune-mediated clearance (Alvarado et al., 2017; Zhou et al., 2015). However, while some mechanisms such as altered MHC class I expression and cytokine production have been described (albeit with ongoing controversy), others remain an open and intriguing area of investigation (Di Tomaso et al., 2010; Korkaya et al., 2011; Schatton et al., 2010; Tallerico et al., 2013). In turn, cytokine production by immune cells promotes CSC maintenance and growth (Ginestier et al., 2010; Iliopoulos et al., 2011).

Emerging technologies are paving the way for a more integrated understanding of CSC-microenvironment interactions. Single-cell genomic and epigenomic technology allow for profiling of stemness gradients within tumors, while 3D culture systems provide new opportunities for studying different niches in vitro (Hubert et al., 2016; Patel et al., 2014). CRISPR-Cas9 and RNA interference screening techniques have yielded an unparalleled array of novel insights into in vivo dependencies and niche-cell interactions (Matano et al., 2015; Miller et al., 2017). Ongoing efforts to elucidate the complexity of the tumor microenvironment have revealed niche-specific dependencies and resiliencies (Brennan and Frame, 2014; Jin et al., 2017). Single-cell sequencing efforts and multiregional tumor studies have identified significant intratumoral diversity (Chung et al., 2017; Li et al., 2007; Puram et al., 2017). Typically, subtyping and CSC isolation have been performed on single biopsy samples. These studies collectively describe a model in which individual tumors are composed of multiple subtypes, implicating microenvironmental diversity in generating cellular heterogeneity (Patel et al., 2014; Puram et al., 2017). Accordingly, therapies targeting a single niche have shown limited efficacy, while cell-autonomous methods neglect the selection pressures of different microenvironments, which promote therapeutic evasion. Further studies should characterize epigenetic and transcriptional CSC diversity across microenvironments to identify potential niche-dependent susceptibilities.

Successfully targeting CSCs will require a deeper understanding of the ways in which CSCs interact with the tumor microenvironment. In the following sections, we describe how each microenvironment maintains stemness and self-renewal properties in CSCs and how CSCs, in turn, promote microenvironmental maintenance and diversity. Finally, we discuss the implications of this framework in therapeutic resistance and the development of effective treatments.

Regions of hypoxia and restricted nutrient availability

Hypoxic, acidic, and necrotic regions are a hallmark of aggressive solid tumors and support CSC maintenance and therapeutic resistance (Carcereri de Prati et al., 2017; Gilchrist et al., 1993; Siebzehnrubl et al., 2013; Zhang et al., 2016, 2015). Hypoxic stress is often accompanied by nutrient restriction and acidic stress. Together, these conditions induce and select for a subpopulation of cells adapted to survive in nutrient-restricted conditions, promoting shifts towards aerobic glycolysis and glutamine-mediated fatty acid production (Dong et al., 2013; Fan et al., 2013; Peng et al., 2018; Warburg, 1927; Zhang et al., 2007). CSCs are enriched for their capacity to survive and thrive in these nutrient-deprived regions, and such conditions promote stemness programs and phenotypes of quiescence and migration (Carcereri de Prati et al., 2017; Kim et al., 2018; Siebzehnrubl et al., 2013). In response to hypoxic signaling, CSCs upregulate pathways that drive the maintenance of the hypoxic niche and promotes immune escape, while inducing paracrine signaling that promotes vascularization and angiogenesis (Conley et al., 2012; Hasmim et al., 2013; Maxwell et al., 1997). Thus, crosstalk between the hypoxic niche and CSCs contributes to the maintenance of stemness, tumor heterogeneity and therapeutic resistance.

The effects of hypoxia are regulated in large part by the hypoxia-inducible factors (HIFs), including HIF-1 and HIF-2. While HIF-1 is globally expressed in many tissues in response to acute hypoxia, HIF-2 remains elevated under chronic hypoxia and upregulates key transcription factors controlling stem cell maintenance, including Klf4, Sox2, and Oct4 (Mathieu et al., 2011). Accordingly, HIF-2 maintains stem-like characteristics in neuroblastoma and glioblastoma cells (Li et al., 2007; Pietras et al., 2007). Acidic stress promotes glioma stemness phenotypes via induction of HIF2a expression (Hjelmeland et al., 2011). HIF1α regulates metabolic adaptation to nutrient deprivation and promotes a mesenchymal shift in hypoxia-treated glioblastoma cells along with expression of pro-survival factors, such as ERK and AKT (Qiang et al., 2012; Siebzehnrubl et al., 2013). Furthermore, HIF1α can also directly activate Notch and WNT signaling (Man et al., 2018; Xu et al., 2017). Glucose restriction increases cancer stemness through upregulation of core stem cell regulators, including Nanog, Sox2 and Oct4 (Flavahan et al., 2013). In breast cancer, hypoxic regions of xenografts are enriched for tumor cells with stem markers and functional stem cell traits (Kim et al., 2018). Moreover, the increase in CSC proportion persisted through secondary xenograft or homogenous tissue culture, suggesting that hypoxia can induce programs that maintain long-term sternness. Interestingly, this result could not be recapitulated using in vitro hypoxic culture as the selection mechanism. The authors identified PI3K/AKT signaling as a positive regulator of stemness in response to hypoxia, but their data regarding long-term preservation of stem-like characteristics suggest an additional, potentially epigenetic, mechanism. Given these findings, further investigation in breast cancers and other tumor types should explore whether CSC heterogeneity manifests not only as a result of current niche residence but is also a shaped by historical localization. Expansion of the CSC population in breast cancer can also be induced in vitro through cyclic hypoxia-reoxygenation (Louie et al., 2010)). Although the endpoints did not include long-term renewal capacity, future work should consider whether hypoxic niche remodeling, which may distinguish dynamic in vivo systems from static in vitro models, may be an important component of CSC maintenance. In breast cancer, hypoxic induction of stemness features has been shown to vary with the mutation status of estrogen receptor, further emphasizing that cell intrinsic traits modulate cell-niche interactions and illustrating the challenge of generalizing findings between model systems (Harrison et al., 2013).

Mechanisms of hypoxia-induced niche maintenance and therapeutic resistance in CSCs

Hypoxia promotes therapeutic resistance in the CSC population across multiple tumor types via induction of EMT signaling and stemness programs (Diehn et al., 2009; Dong et al., 2013; Hubert et al., 2016; Wu et al., 2017; Yan et al., 2018). CSCs exposed to hypoxic stress are enriched for resistance to chemotherapy and radiotherapy via diverse resistance mechanisms such as induction of HIF signaling and consequent upregulation of stem pathways such as CD44 and Notch signaling (Wu et al., 2017; Yan et al., 2018). CSCs are also uniquely positioned to survive therapeutic stressors compared with non-stem tumor cells in the hypoxic region due to their relatively lower production of reactive oxygen species (ROS) (Cojoc et al., 2015). Several studies have proposed that hypoxia promotes a quiescence phenotype in CSCs, facilitating resistance to therapies that induce genotoxic or proliferative stress, such as chemotherapy and radiotherapy (Carcereri de Prati et al., 2017; Chen et al., 2012; Gordan et al., 2007; Hubert et al., 2016). However, this effect is likely more nuanced given the technical challenges of modeling quiescence; thus, the interaction between hypoxia and quiescence remains an open question with many mechanistic aspects yet to be resolved.

CSCs in hypoxic regions maintain the niche as well as generate and reconstitute other microenvironments via migration, angiogenesis, and preferential nutrient uptake (Bao et al., 2006; Flavahan et al., 2013; Maxwell et al., 1997). Under hypoxic conditions, CSCs express high levels of HIF-1, which promotes VEGF expression and angiogenesis in hypoxic region (Maxwell et al., 1997). This likely contributes significantly to the challenges faced by anti-angiogenic drugs in the clinic.

CSCs exposed to hypoxia also induce and maintain other niches. Hypoxic CSCs migrate via upregulation of epithelial-to-mesenchymal (EMT) signaling and changes in adhesion receptor expression (Plaks et al., 2015; Siebzehnrubl et al., 2013; Zhang et al., 2016, 2015). In the context of hypoxia, single cells detach from a collective migratory group, a process known as ameboid migration (Lehmann et al., 2017). Although this study did not specifically segregate CSCs from the general population, nor define the stemness of their study population, previous reports have identified an enrichment of collective to ameboid migratory transition among CSCs in other cancers (Taddei et al., 2014).

CSCs residing in hypoxia demonstrate distinct immunologic features and niche interactions. VEGF suppresses effective dendritic cell function and promote PD-L1 expression, suggesting that dendritic or T cell therapies may be less effective in targeting the hypoxic niche (Meder et al., 2018; Voron et al., 2015). Accordingly, combinatorial treatment with anti-PD-L1 and anti-VEGF treatment has shown promise in a preclinical mouse model of small cell lung cancer (Meder et al., 2018). Several features of the hypoxic niche may inhibit cytolytic T cell activity. CSCs preferentially express the high-affinity glucose transporter, type 3 (GLUT3) to outcompete surrounding normal tissue for glucose uptake under conditions of low glucose stress (Flavahan et al., 2013). As such, glioblastoma CSCs may further restrict nutrient availability for other cells residing in the hypoxic niche. Cytotoxic T cells are particularly susceptible to glucose restriction and thus may function less efficiently in the hypoxic niche (Chang et al., 2015). Moreover, induction of Nanog in hypoxic melanoma cells leads to recruitment of regulatory T cells and immunosuppressive macrophages (Hasmim et al., 2013). Autophagy, which is elevated in hypoxic tumor regions, particularly in proximity to CSCs, inhibits cytolytic T cell activation (Berardi et al., 2016; Noman et al., 2011; Sharif et al., 2017; Zhang et al., 2017a). Hypoxia also exhibits CSC-specific, immunosuppressive and cell autonomous effects, such as downregulation of MHC complexes across solid cancers. Microenvironment-specific findings may explain the variation in reports regarding MHC levels in CSCs by tumor type (Di Tomaso et al., 2010; Schatton et al., 2010; Tallerico et al., 2013). Taken together, these data suggest that CSCs from different niches exhibit complex immunosuppressive and immune evasion properties. Immune-oncologic efforts to effectively target CSCs will require combinatorial immunotherapies due to this microenvironmental variation.

CSC interdependence on perivascular niches

A perivascular niche for CSCs is perhaps best characterized in glioblastoma but has also been described in other solid tumors, including melanoma, skin papilloma, and breast cancer, as well as more broadly in areas of metastatic growth (Beck et al., 2011; Calabrese et al., 2007; Correa et al., 2016; Fazilaty and Behnam, 2014; Lai et al., 2012). The perivascular niche is characterized by interactions with endothelial cells (ECs) and components of the extracellular matrix (ECM). ECs promote a stemness phenotype in cancer cells through Notch, Sonic Hedgehog and nitric oxide signaling pathways (Charles et al., 2010; Varnat et al., 2009; Zhu et al., 2011). In colorectal CSCs, ECs maintain stemness programs through production of Jagged-1, and in skin cancer, via VEGF-mediated signaling through Nrp1 (Beck et al., 2011; Lu et al., 2013). The nutrient-rich environment in conjunction with signaling via VEGF, MYC, and other molecules elevated in the perivascular region promote CSC proliferation, although one study in breast cancer found that this effect is dependent upon the type of vasculature and the signaling molecules expressed (Ghajar et al., 2013; Liu et al., 2017; Zhang et al., 2015). Whereas mature vasculature stimulated dormancy via thromsbospondin-1, neovasculature signaling induced CSC proliferation (Ghajar et al., 2013).

CSCs, in turn, are drivers of vascularization via both stimulation of endogenous ECs as well as vascular mimicry, a process by which tumor cells enriched for CSCs form functional vascular-like structures. Generation of blood vessel-like structures by CSCs has been described in breast cancer, glioblastoma, melanoma, and colorectal cancer, among others (Fan et al., 2013; Maniotis et al., 1999; Shangguan et al., 2017). Furthermore, CSCs regulate vascular function. Glioblastoma CSC-derived pericytes directly regulate the brain-to-tumor barrier via modulation of tight junctions(Zhou et al., 2017). Selective targeting of CSC-derived pericytes by ibrutinib, an inhibitor of the bone marrow and X-linked non-receptor tyrosine kinase (BMX), enhanced drug delivery and improved chemotherapeutic efficacy (Shi et al., 2018). Perivascular CSCs in glioblastoma are marked by CD109, which are positively correlated with progression, providing further evidence for microenvironmental diversity and region-specific targeting strategies for CSCs (Shiraki et al., 2017). In addition to maintaining the perivascular niche, resident CSCs also facilitate the formation of an invasive edge via migration along blood vessel scaffolding, and of a new hypoxic niche through stimulation of proliferation and consequent outgrowth of vascular supply (Baker et al., 2014; Watkins et al., 2014).

In another example of niche interdependence, perivascular macrophages promote metastasis in breast cancer, while perivascular regions are themselves preferential sites of seeding and outgrowth of metastatic breast cancer cells (Charles and Holland, 2010; Wyckoff et al., 2007). Furthermore, L1 cell adhesion molecule (L1CAM) is a CSC marker in glioblastoma and is used by brain metastatic cells to co-opt pericyte-EC interactions and promote perivascular dissemination and metastatic seeding (Bao et al., 2008; Er et al., 2018). In summary, the perivascular region provides a supportive niche and growth cues for CSCs, which in turn augment and drive vascularization and EC maintenance. Targeting the mutually supportive interactions of CSCs with the perivascular niche will be critical to developing effective therapeutic strategies.

CSC as drivers of the invasive front

Metastases are responsible for 90% of patient deaths from cancer. The metastatic process is associated with the acquisition of stem-like characteristics, and CSCs are enriched for invasive potential across a variety of cancers in patients and in vivo model systems. Many pro-invasive pathways overlap with known CSC signaling, and properties of the invasive microenvironment may facilitate migration and invasion by CSCs (Hermann et al., 2007; Liu et al., 2007; Louhichi et al., 2018; Marcato et al., 2011; Reuben et al., 2011; Sheridan et al., 2006; Sliva et al., 2002).

The CXCR4/CXCL12 signaling axis promotes metastasis and invasion across a wide variety of solid tumors. CXCR4 is enriched in CSCs and often defines a pro-metastatic subpopulation while expression of CXCL12 defines higher risk regions for metastasis in breast cancer invasion (Calinescu et al., 2017; Müller et al., 2001). Other stem pathways, such as Hedgehog signaling, increase metastases in pancreatic cancer(Feldmann et al., 2008). In prostate cancer, normal stem cell niches, in particular the HSC niche, appear to enrich for metastatic tumor cells, suggesting potential shared supportive cues among normal and cancer stem cells(Shiozawa et al., 2011). Whereas a defining feature of many aggressive, late-stage solid tumors is metastasis, glioblastoma displays negligible rates of metastasis, but extensive invasive capacity within the brain. Invasion is commonly observed along the brain vasculature in glioblastoma, due in part to expression of adhesion molecules and supportive signals, such as Notch and Sonic hedgehog (Charles et al., 2010; Takebe et al., 2011). This niche also commonly harbors brain metastases from other cancers (Takebe et al., 2011).

Other pro-migratory components of the invasive or metastatic niche include mesenchymal stem cells (MSCs) and stromal features such as hyaluronic acid. MSCs promote CSC proliferation and metastasis in breast cancer via IL6, CXCL7 and CCL5 signaling, and in osteosarcoma via CCL5 (Karnoub et al., 2007; Liu et al., 2011; Xu et al., 2009). Hyaluronic acid in the tumor stroma interacts with CD44 in breast cancer CSCs, inducing Nanog and Sox2 expression, thereby promoting drug resistance and metastasis (Bourguignon et al., 2008).

CSCs themselves are overrepresented at the invasive front. Circulating tumor cells (CTCs) are highly enriched for CSC properties, and CSCs have a higher propensity towards invasion, suggesting that they are enriched in all stages of metastasis, not just the final stages of metastatic growth(Grillet et al., 2017; Toloudi et al., 2011). Only a subset of CTCs effectively colonizes the metastatic site, and several cell intrinsic features promote colonization and outgrowth. Breast cancer cells enriched for EMT-induced integrin signaling interact with surrounding stroma and display increased metastatic potential (Shibue et al., 2013). ALDH1A3-expressing breast CSCs are positively associated with risk of metastasis (Charafe-Jauffret et al., 2009; Marcato et al., 2011)). Markers, such as CD44, CXCR4, and ALDH1A3, may define subsets of CSCs with enhanced migratory potential in different tumor types (Li et al., 2017; Marcato et al., 2011). Moreover, EMT signatures are prevalent at the invasive or metastatic front and are enriched in a subset of CSCs in many tumor types (Polireddy et al., 2016; Puram et al., 2017; Zhang et al., 2017b). CSCs have thus been shown across a wide variety of cancers to form the migratory front, promoting invasion and metastasis. While a further dissection of CSC regulation of the metastatic niche is outside the scope of this review, this topic has been reviewed in depth in (Plaks et al., 2015).

Interactions between cancer stem cells and the immune system

Immune cell infiltration serves a complex and varied role across solid tumors with many immune components co-opted to serve pro-tumorigenic functions (Thorsson et al., 2018). Different profiles of immune cells are predominant in different tumor types, but many interactions are consistent across several solid tumors. For example, M2-polarized tumor-associated macrophages (TAMs) commonly support and maintain CSC populations through chemokine signaling and induction of stem pathways such as Sonic hedgehog (Jinushi et al., 2011). TAMs also promote EMT and CSC characteristics across multiple cancer types (Fan et al., 2014; Li et al., 2018a).

Immune escape plays a critical role in tumor initiation. As CSCs represent key drivers of tumorigenesis, they are generally immunosuppressive. CSCs evade killing by immune cells through a variety of mechanisms. For example, CD133⁺ glioblastoma CSCs and ABCB5⁺ melanoma CSCs down-regulate MHC class I molecules (human HLA-A, HLA-B and HLA-C) to escape from T cell attack (Di Tomaso et al., 2010; Schatton et al., 2010). Glioblastoma CSCs also decrease the expression of low molecular weight protein (LMP) and transporter associated with antigen processing (TAP) to reduce the capacity of antigen processing and presenting pathways (Di Tomaso et al., 2010). Furthermore, low expression of Toll-like receptor 4 (TLR4) in CSCs helps them evade innate immune suppression (Alvarado et al., 2017).

Increasing evidence also supports the role of CSCs in remodeling the immune response within the tumor microenvironment. CSCs secrete many immunosuppressive molecules such as TGF-β and IL4, which attenuate the anti-tumor immune response (Nappo et al., 2017). TGF-β inhibits the proliferation of active T cells while inducing regulatory T cell activation via Foxp3-dependent and independent pathways (Fantini et al., 2004; Oh et al., 2017; Thomas and Massagué, 2005). TGF-β also inactivates natural killer cells and promotes pro-oncogenic M2 macrophage polarization (Ganesh and Massague, 2018). Furthermore, glioblastoma CSCs secrete periostin (POSTN) to recruit M2 macrophages, while breast CSCs physically interact with M2 macrophages via ligand-receptor binding of CD90-CD11b and EphA4-Ephrin (Lu et al., 2014; Zhou et al., 2015). In addition, CSCs in lung cancer promote myeloid cell polarization to an M2 phenotype (Yamashina et al., 2014). Notably, immune checkpoint inhibitors targeting CTLA4, PD-1 and its ligand PD-L1 have achieved great success in clinical trials targeted at a variety of cancers including metastatic melanoma, non-small cell lung cancer, renal cell carcinoma, and others (Li et al., 2018b; Pardoll, 2012). However, studies investigating PDL1 expression in CSCs have yielded mixed results. Some reports showed that PDL1 is preferentially expressed on CSCs in breast, colon, and head and neck cancers, while other studies indicate lower or even undetectable PDL1 in CSCs (Maccalli et al., 2014). Using The Cancer Genome Atlas (TCGA) dataset, a machine-learning algorithm was able to identify cancer stemness features (Malta et al., 2018). In most tumor types, such as lung squamous cell carcinoma (LUSC), glioblastoma (GBM), low grade glioma (LGG), and prostate adenocarcinoma (PRAD), stemness signatures are negatively correlated with PDL1 expression. This finding indicates that immune checkpoint blockage may not effectively target CSCs, but additional immune evasion mechanisms need to be further investigated.

Immune-CSC interactions regulate and are modulated by interactions in other niches. Microglia, which compose the majority of immune infiltrate in glioblastoma, enhance migration and invasion (Markovic et al., 2005; Wallmann et al., 2018; Wu and Watabe, 2017; Zhu et al., 2016). The role of the immune system in metastasis has been the focus of extensive research and serves both a bottleneck and a potentiating factor in homing and proliferation at the metastatic site (reviewed in Kitamura et al., 2015). Immune function also provides a functional link between dormancy and metastasis. Quiescent CSCs are theorized to be critical for metastasis and regulate quiescence via DKK1 and WNT signaling, which also suppress expression of natural killer cell targets (Malladi et al., 2016). Moreover, neutrophils activate dormant tumor cells metastatic to the lung, promoting growth at the metastatic site (Albrengues et al., 2018).

The role of the immune system in experimentally delineating CSCs is likely critical. The proportion of tumor initiating melanoma varies upon implantation into non-obese diabetic/severe combined immunodeficiency mice based on the degree of immunosuppression; the proportion defined as CSCs in the experiment varied from 0.0001% to as high as 25% when the immune system was further compromised by deletion of IL-2 receptor gamma (Quintana et al., 2008). This implies that the functional, experimentally-derived definition of a CSC is inextricable from the cell’s capacity for immune evasion and interaction. Immunotherapeutic strategies must account for the unique immunologic characteristics of CSCs and their capacity for immune escape in order to effectively address tumor recurrence and metastasis.

Interactions between cancer stem cells and other cell types

The CSC microenvironment contains a multitude of cells including cancer-associated fibroblasts (CAFs), differentiated neoplastic cells, and normal cells. Accumulating evidence supports an essential role for reciprocal crosstalk between those cells and CSCs in self-renewal. CSCs secrete a variety of cytokines or ligands to transform normal fibroblasts into CAFs, which, in turn, support CSC proliferation and self-renewal (Gascard and Tlsty, 2016). On one hand, CAFs secrete many growth factors such as HGF and CCL2, which directly upregulate the stemness of stem cells through activation of the key stemness regulator WNT and Notch (Gascard and Tlsty, 2016; Lau et al., 2016; Tsuyada et al., 2012; Vermeulen et al., 2010, 2010). On the other hand, CAFs and CSCs produce the BMP antagonist Gremlinl to block BMP-induced pro-differentiation effects on CSCs (Kalluri, 2016; Yan et al., 2014). A newly identified subpopulation (CD10⁺/GPR77⁺) of CAFs highly correlates with poor patient survival in breast and lung cancers (Su et al., 2018). This subset of CAFs is driven by hyperactive NF-κB signaling and sustains cancer stemness by secreting IL6 and IL8. As cancers are “the wounds that do not heal,” and wound responses are associated with stem/progenitor gene expression profiles, the therapeutic benefit of targeting CAFs likely serves as an indirect, but potentially efficacious route to targeting CSCs.

Differentiated tumor progeny represents the majority of most tumors, relative to CSCs (Al-Hajj et al., 2003; Bapat et al., 2005; Fang et al., 2005; Li et al., 2007; Ma et al., 2007; Ricci-Vitiani et al., 2007; Singh et al., 2003). Although differentiated tumor cells do not initiate tumorigenesis, they are not “waste cells” as designated by some, but rather integral components of the tumor microenvironment and can increase proliferation and self-renewal of CSCs (Tammela et al., 2017; Wang et al., 2018). In lung adenocarcinoma, WNT signaling in CSCs is maintained by elevated expression of WNT ligands in differentiated tumor cells to maintain tumor growth and progression (Tammela et al., 2017). In glioblastoma, co-implantation of CSCs with differentiated tumor cells accelerated tumor growth via a VGF-BDNF loop in which brain-derived neurotrophic factor (BDNF) secreted by differentiated tumor cells activates cognate receptors on CSCs, which then feedback to the differentiated cells through expression of VGF (Wang et al., 2018). Thus, targeting of CSCs and differentiated tumor cells may provide potential targets for cancer treatment.

Modeling niche interactions – challenges and new frontiers

A shared characteristic of the most aggressive solid tumors is the complexity of the tumor microenvironment. CSCs not only reside in, but actively remodel and are in turn regulated by each element of these niches, giving rise to a heterogeneous and dynamic population with differential dependencies and a wide variety of resistance mechanisms (Figure 3). While targeting a single niche may be effective in some cancers, particularly in low-grade tumors, treatment for many aggressive solid tumors remains palliative. Furthermore, the niches themselves are neither isolated nor static. The vascular region becomes a part of the hypoxic core as tumors proliferate and outgrow their blood supply. Immune cells reside throughout the different regions and may exhibit distinct properties and functional interactions across niches.

Figure 3. Effective therapeutic strategies should target CSC-niche interactions.

Robustness is a critical component of a tumor modelled as a self-sustaining ecosystem. Thus, disruption of the intratumoral interactions driving that resilience will be crucial for effective treatment. For example, angiogenesis inhibitors may inhibit VEGF signaling (e.g. bevacizumab) or CSC-derived pericytes (e.g. via BMX inhibition). Checkpoint inhibitors such as PD-L1 or PD-1 antagonists (e.g. nivolumab, atezolizumab) have been used to target the bulk tumor of many solid cancers, but CSC-specific strategies may require the characterization and targeting of additional immunosuppressive or checkpoint mechanisms. Cytokine signaling, for example production of interleukin-6 by endothelial cells, or interleukin-4 by CSCs, promotes CSC maintenance and immunosuppression, respectively. Perivascular CSC-specific molecules, such as CD109 in glioblastoma CSCs, can be utilized in combination with strategies such as HIF inhibitors to target CSCs across multiples niches. CSC, Cancer stem cell.

Modeling the complexity of the dynamic tumor ecology remains a significant challenge. Culturing cells invariably modifies them and cell lines, by their very nature, acquire renewal properties in artificial conditions (grown on plastic in hyperoxia, relatively basic pH, huge excess of glucose and other nutrients, lack of interactions with other microenvironmental features), so it is not surprising that cell lines may fail to replicate the cellular hierarchies present in patient tumors, even if they have a heterogeneity. Serum induces irreversible changes in transcriptional regulation, suggesting that no cell line grown in standard culture conditions is likely to retain the true CSC signature. One substantial limitation of in vitro culture, and even most in vivo systems, is the inability to recapitulate the full scope of tumor heterogeneity.

Considerations for in vitro and in vivo modeling of the complex tumor microenvironment

In vitro attempts to model and study microenvironmental interactions have relied primarily on 2D co-culture systems (e.g. endothelial and tumor cells) or perturbation of growth conditions (e.g. hypoxia or low glucose), each of which models a single niche in isolation. As previously noted, in vitro perturbations such as culturing cells in hypoxia may not recapitulate in vivo effects (Kim et al., 2018). Further illustrating the discrepancy between model systems, another report comparing migration in CSCs vs. NSTCs found that CSCs migrate faster in a 3D culture system, but not in 2D (Thomas et al., 2016). Modeling and manipulation of quiescence has also proven extremely challenging as standard 2D tissue culture selects for rapidly proliferating cells.

Traditional 2-dimensional in vitro cultures constrain cells in a single microenvironment, while in vivo models frequently do not reach the size and complexity of a tumor in a human patient. Furthermore, readouts of success at the bench often focus on magnitude of cell death in the bulk population, neglecting the very thing that kills patients – the residual, resistant cells. Emerging technologies such as single-cell RNA-sequencing and novel strategies for single-cell evaluation of CRISPR-Cas9 screening have begun to address this complexity and may yield new therapeutic approaches.

Culture systems must keep evolving to try to better address intratumoral complexity and inter-niche communication. Use of organoids, which can maintain stemness and nutrient gradients, as in vitro 3D alternatives to traditional 2D culture systems, has rapidly expanded with the goal of more effectively modeling multiple niches and overall tumor heterogeneity in vitro (Abbasi, 2018; Hubert et al., 2016; Matano et al., 2015; Verissimo et al., 2016). Other techniques, such as microfluidic devices, facilitate the study of niche crosstalk, migration, and cell-cell interactions by allowing for modeling of chemical gradients, 3D extracellular matrix structures, vascular interactions, and co-culture of different cell types (Gritsenko et al., 2017; Ray et al., 2017; Wang et al., 2013; Zhang et al., 2012, 2013). Quiescent CSCs, which are generally considered to drive therapeutic resistance, are selected against in vitro, making them a highly relevant, but difficult population to study. Organoids and single-cell microfluidic chambers facilitate high-throughput studies of these cell populations, allowing for better characterization of intratumoral heterogeneity (Hubert et al., 2016; Lecault et al., 2011).

Different limitations and strengths also exist in all in vivo systems - xenograft models frequently utilize immunodeficient mice, which has been demonstrated to affect engraftment and muddle the true definition of a CSC by functional assays. Genetically engineered mouse models have proven extremely useful, but caution must be taken to interpret the results in the context of the original genetic lesion, as exemplified by the previously described study in which ER receptor status modified the interaction between hypoxia and CSCs (Harrison et al., 2013).

Single cell RNA-sequencing studies have begun to map the heterogeneity of CSCs in single tumors, and future studies utilizing multiregional, single cell sequencing and characterization of single cell epigenomes will be important to identify specific signaling interactions that can be further studied in emerging model systems. CRISPR-Cas9 screens utilizing single cell barcoding may facilitate a more in-depth analysis of genes affecting niche interactions in vivo that would otherwise be infeasible in bulk.

Therapeutic strategies targeting resistance and niche interactions

Ultimately, the critical drivers of therapeutic resistance in as-yet untreatable solid tumors are adaptation and heterogeneity, both of which are highly microenvironment-driven. Targeting any individual niche or single marker remains, in many cases, insufficient, and combinatorial approaches will be critical for deriving novel and effective therapeutics (Pei et al., 2016). For example, multiregional studies have demonstrated the existence of differential epigenetic dependencies in CSCs, and thus combinatorial therapeutics based on niche-specific vulnerabilities will likely prove critical in targeting tumors with a complex ecology (Jin et al., 2017; Ye et al., 2016).

Hypoxic conditions drive immunosuppressive signaling, possibly due in part to preferential nutrient uptake by CSCs and increased autophagy. Thus, anti-PD-L1 or cytotoxic T cell therapy alone may not be sufficient for effective treatment in solid tumors harboring a hypoxic niche. Modification of CAR T-cells to promote glucose uptake or combinatorial therapy with autophagic inhibitors may improve immunotherapeutic targeting in hypoxic regions. Furthermore, if HIF signaling mediates long-term changes in CSC potential irrespective of later culture conditions, strategies targeting hypoxia-induced resiliency must move beyond inhibition of immediate HIF signaling and towards epigenetic mechanisms of prolonged CSC maintenance.

The hypoxic niche is known to harbor CSCs that are particularly resistant to traditional chemotherapy and radiotherapy. Hypoxic stress induces a large array of signaling changes, many of which are funneled through pleiotropic HIF factors. Strategies to target this microenvironment have focused on inhibiting HIFs, particularly HIF-1, directly through small molecule inhibition or indirectly through other pathways such as mTOR, PI3K, the unfolded protein response, and others that are known to regulate HIF levels. While several HIF-1 inhibitors have progressed to phase III trials, they have been largely ineffective in prolonging survival (Schito and Semenza, 2016). Other strategies targeting the hypoxic niche may focus on key metabolic adaptations by CSCs that engage in a positive regulatory feedback cycle with HIF, via enzymatic dependencies of aerobic glycolysis (e.g. PDK1), or on mitochondrial remodeling, which is a CSC dependency in the context of increased ROS and low pH.

Strategies to enhance immune-mediated tumor suppression are rapidly evolving. These include a growing number of immune checkpoint inhibitors as well as an expanding array of adoptive immunotherapeutic strategies using both autologous and allogeneic cells. These approaches, which are primarily CAR T-cell or dendritic-cell based, have been designed to specifically target CSC markers including CD133 and ROR1 (Berger et al., 2015; Zhu et al., 2011). Other strategies must account for enhanced immune evasion in CSCs, and thus could focus on augmenting class I MHC expression. CSCs, potentially as a result of their suppressed immunogenicity, may be particularly susceptible to oncolytic viruses that are either engineered against specific CSC markers or that naturally target stem cells.

Infiltrative CSCs are arguably the most critical population for therapeutic targeting in resectable solid tumors. In these cells, integrin signaling molecules may frequently serve as both CSC markers and drivers of migration and infiltration, and several integrin inhibitors are in early phase clinical trials. Chemokine signaling, in particular the CXCR4-CXCL12 axis, CD44, or α6 integrin, is another potentially promising clinical target.

Efforts to target the perivascular microenvironment have included antiangiogenic therapies such as anti-VEGF agents (Jain et al., 2006). However, while these therapies showed efficacy in some cancers, therapeutic success is frequently abrogated by alternate vascularization signaling and by the heightened capacity of CSCs to co-opt normal vasculature and to survive in hypoxic conditions. As anti-VEGF therapies have shown, antiangiogenic strategies must account for the alternative pathways, such as HIF-mediated signaling, by which CSCs can drive vascularization. Other potential therapeutic targets include the EC-CSC signaling axis via inhibition of Notch or WNT signaling. Niche-specific CSC markers, such as CD109 in perivascular glioblastoma CSCs, provide a unique opportunity for microenvironmentally-informed targeting of CSCs. Ultimately, successful strategies in aggressive tumors must target the adaptive potential of a highly heterogeneous population of CSCs, whose diversity is maintained by a complex environmental structure.

CONCLUSIONS

The occurrence and development of tumors is driven not only by cell-autonomous events such as genetic and epigenetic changes, but also through critical interactions with the tumor microenvironment, which serves as a “fertile soil” to support them. CSCs interact in a mutually supportive relationship with their microenvironment, evolving together to promote tumor progression. Tumor fitness, arising from cellular diversity and rapid evolutionary potential, appears to arise from the apex of the hierarchy – the CSC. The currency of therapeutic resistance is heterogeneity and adaptation within this population, which can then pass on its genetically and epigenetically adaptive traits to its progeny. Spatial heterogeneity, e.g. invasion and metastasis, facilitates surgical resistance. Molecular heterogeneity, as with genetic lesions, drives escape from specific molecular targeting strategies. Cell cycle heterogeneity creates, as well as arises from, a population spectrum of quiescence and proliferation, a gradient that has been suggested to span across the hypoxic to perivascular niches and allows therapeutic escape from chemotherapy and radiation.

Thus, the coexistence of diverse microenvironments throughout a solid tumor generates and selects for heterogeneity within the CSC population. This population reciprocally acts to maintain its niche and drive new niche formation, sustaining ecological and therefore cellular heterogeneity. Therapeutic resistance, and tumor resilience and recurrence are all intrinsically linked to tumor heterogeneity and microenvironmental crosstalk. CSCs both are architects of this framework and are molded by it. As critical drivers of tumor heterogeneity, CSCs do not simply survive in harsh conditions, but thrive in and drive them. It is this crosstalk that drives heterogeneity and complexity, which in turn promotes therapeutic resistance and tumor recurrence. Therefore, effective treatment strategies must emphasize not just cell autonomous growth and survival cues, but also reciprocal crosstalk between components of the tumor organizational unit. In order to develop effective and translatable therapeutics, convenient models must give way to accurate ones. It will be critical to continue improving model systems to recapitulate the complexity of the tumor niche in vivo as much as possible and to grasp the full scope of the CSC role in tumor biology.

Figure 2. CSCs engage in reciprocal signaling with each tumor niche that sustains the CSC population and promotes niche development.

Through a wide variety of niche-specific signaling mechanisms, CSCs promote immunosuppression, migration, therapeutic resistance, and EMT, while receiving supportive cues from each microenvironment that maintain stem-like features, such as self-renewal, migration and differentiation. Examples of mutually supportive CSC-niche interactions are detailed for each of the four described niches, including the hypoxic niche (blue), the immune niche (yellow), the perivascular niche (red), and the infiltrating region (green). CSC, Cancer stem cell; EMT, Epithelial-mesenchymal transition; Shh, Sonic hedgehog.