Cellular Plasticity in Bone Metastasis

Abstract

Metastasis is responsible for a large majority of death from malignant solid tumors. Bone is one of the most frequently affected organs in cancer metastasis, especially in breast and prostate cancer. Development of bone metastasis requires cancer cells to successfully complete a number of challenging steps, including local invasion and intravasation, survival in circulation, extravasation and initial seeding, and finally, formation of metastatic colonies after a period of dormancy or indolent growth. During this process, cancer cells often undergo a series of cellular and molecular changes to gain cellular plasticity that helps them adapt to various environments they encounter along the journey of metastasis. Understanding the mechanisms behind cellular plasticity and adaptation during the formation of bone metastasis is crucial for the development of novel therapies.

Introduction

With the aging and urbanization of the human population, cancer has become a leading cause of mortality in the modern world. While advances in early diagnosis, prevention and therapies have significantly improved the survival odds of cancer patients, metastasis remains the overwhelming cause of mortality in cancer patients. Bone is one of the major organs that is frequently affected by metastasis, occurring in more than 70–80% of late stage breast, prostate, and other cancers [1]. Similar to metastasis to other organs, development of bone metastasis is a rather complicated process where tumor cells first go through single cell or collective invasion from the primary tumor to surrounding tissues, followed by intravasation into vascular or lymphatic circulation, and dissemination into the bone [2]. Upon arriving at the bone, interactions between tumor cells and bone stromal cells are needed to facilitate the initial seeding in different bone niches. Although disseminated tumor cells (DTCs) are commonly seen in the bone marrow of cancer patients, most of them remain dormant and only some of the patients will eventually develop bone metastases, indicating that only a small proportion of cancer cells with specific features can successfully colonize and thrive in the bone. Considering this cascade of events that the tumor cells need to go through to finally form bone metastases, the various environments cancer cells encounter during this process appear to be rather different from each other. Conceivably, successful colonization in the bone requires a considerable degree of cellular plasticity so that cancer cells can survive in and adapt to these different environments.

Cellular plasticity refers to the ability to switch cell fates or phenotypes in response to varied environments [3]. In development, cellular plasticity gives rise to strong self-renewal ability and differentiation potentials in stem cells; while in cancer, plasticity promotes cell flexibility, stemness, cancer cell survival, and metastasis [3, 4]. Moreover, the high adaptability of cancer cells can render them resistant to chemotherapies [5]. Thus, understanding the mechanisms of cancer cellular plasticity in metastasis will help the development of novel targets to control and even eliminate cancer. In this review, we summarize recent research regarding the cellular plasticity in each step during bone metastasis, its underlying molecular mechanisms as well as its contributions to bone metastasis.

Epithelial-mesenchymal transition promotes cellular plasticity

Epithelial-mesenchymal transition, or EMT, is an essential cellular program that allows polarized epithelial cells to switch to a mesenchymal phenotype. Besides facilitating key processes during embryonic development such as gastrulation and neural crest delamination, EMT is also largely involved under pathological conditions, including cancer [6]. EMT was first believed to promote cancer progression and metastasis through enabling tumor cells to detach from the primary tumor and attain migratory and invasive phenotypes. During EMT, epithelial markers such as E-cadherin are downregulated while mesenchymal markers, including N-cadherin and vimentin, are upregulated, along with systemic changes in a large number of genes related to different aspects of cellular features [7–9]. This conversion of epithelial to mesenchymal state promotes cell migration and invasion, which is essential for the first step of metastasis. With the application of better disease models and technologies in cancer research, EMT has been recognized as a more complicated process beyond merely supporting a mesenchymal phenotype. Rather, EMT represents a cellular program giving rise to cancer cellular plasticity, which promotes cell stemness, tumor cell survival in the circulation, chemoresistance, metabolic adaptation and immune evasion [10] (Figure1).

Figure 1. EMT promotes cancer cell plasticity.

EMT is characterized by the loss of epithelial features such as E-cadherin downregulation and loss of cell-cell junctions. Instead of being a binary process, EMT represents a spectrum of cell states and cancer cells can fall between fully epithelial and fully mesenchymal states, which is referred to as partial EMT. Cells with complete EMT tend to disseminate into the blood as single cells, while cells with partial EMT are more likely to form CTC clusters, where cell-cell junctions, including those mediated by E-cadherin, Keratin 14 and Plakoglobins, are maintained. CTC clusters confer survival advantages to cancer cells, as they can associate with platelets to form thrombi, which in turn protect them physically from shear stress and immune attack. Meanwhile, platelet-derived TGF-β can enhance cancer cell EMT. Furthermore, cancer cells can form microtentacles to resist shear stress in the circulation. To overcome immune surveillance, cancer cells can upregulate PD-L1 expression to inhibit CTL-mediated killing. Additionally, downregulation of ligands for NK cell-stimulating receptors help cancer cells to evade NK-mediated killing. Abbreviations: EMT, epithelial-to-mesenchymal transition; CTC, circulatory tumor cells; TGF-β, transforming growth factor-β; PD-L1, programmed death-ligand-1; PD-1, programmed death-1; MHC-I, major histocompatibility complex-I; TCR, T cell receptor; CTL, cytotoxic T lymphocytes; NK, natural killer.

It has been increasingly evident that EMT is not a binary process; instead, it appears as a spectrum, and cancer cells often fall between fully epithelial and fully mesenchymal states [6]. Cells exhibiting partial EMT are featured with co-expression of both epithelial and mesenchymal markers and their presence has been seen in both animal models [11, 12] and cancer patients [13–15]. For example, circulatory tumor cells (CTC) that are positive for both vimentin and keratin are frequently found in non-small cell lung cancer (NSCLC) patients [15]. Additionally, partial EMT is also observed in head and neck squamous cell cancer (HNSCC) [14], and in skin and mammary tumors [13]. Moreover, epithelial cells with mesenchymal traits are also found during organogenesis of mouse embryos by single-cell RNA-seq [16], and a gene signature incorporating embryonic plasticity genes can predict metastatic competence in breast cancer cells [17], indicating that cancer cells may adopt similar mechanisms in normal physiology to acquire cellular plasticity. Indeed, it has been reported that partial EMT can enhance cancer cellular plasticity [12] and promote cancer progression and metastasis [11, 18]. In a pancreatic ductal adenocarcinoma cancer (PDAC) mouse model, Aiello et al demonstrated that cancer cells with partial EMT induced by E-cadherin internalization have enhanced plasticity and they are more likely to give rise to E-cadherin⁺ tumors compared to those with full EMT transformation [12]. Consistently, Padmanaban et al also illustrated the importance of E-cadherin expression in distal metastasis [18]. They found that E-cad⁻ breast cancer cells exhibited greater ability in migration and invasion than E-cad⁺ cells but less potential to metastasize, which is due to accumulated reactive oxygen species (ROS) upon detachment and enhanced apoptosis in the case of E-cadherin loss [18]. Interestingly, they also found that the few metastases formed by E-cad⁻ cells are cytokeratin⁺, suggesting a necessity in acquiring an epithelial phenotype with epithelial markers re-expression upon colonization in distal sites [18]. Importantly, while partial EMT seems to be a hybrid state with co-expression of epithelial and mesenchymal functional molecules, its features cannot be phenocopied by mixing fully epithelial and fully mesenchymal cells [11], indicating that partial EMT represents an integral cellular program that allows flexible cell fate transition.

Besides promoting cellular plasticity, partial EMT also contributes to metastasis by enhancing stemness of cancer cells, and supporting their survival in circulation. Single-cell analysis on patient-derived xenograft (PDX) models of breast cancer reveals a correlation between EMT and cancer cell stemness [19]. Specifically, elevated expression levels of stem-cell and EMT markers including LGR5, TWIST1, SNAI2 etc. are found in metastatic cells isolated from low-burden metastases, which represents the early phase of metastasis formation in distal organs [19], supporting a role of EMT and stemness in promoting cancer dissemination and metastasis. Forced EMT by ectopic expression of Twist and Snail in human mammary epithelial cells (HMLEs) gives rise to mainly CD44^high/CD24^low expression pattern on those cells [20], which represents a typical breast cancer stem cell phenotype [21–23]. Furthermore, this expression pattern is accompanied by increased ability to form mammospheres in vitro, suggesting enhanced stemness [20]. A similar phenomenon was also observed in MCF10A cells [24], breast cancer cells [25], gastric epithelial cells (GIF-14) [26], pancreatic and colorectal cancer cells [27]. Moreover, EMT and stem cell characteristics are seen in CTCs derived from metastatic breast cancer patients [28]. Importantly, fully mesenchymal state induced by Zeb1 expression led to decreased tumorigenicity and a shift from canonical Wnt signaling, which is induced as a stem cell program in cells with partial EMT, to non-canonical Wnt signaling [11]. This suggests that the cellular plasticity, rather than mesenchymal state generated by EMT, drives cancer cell stemness.

Intravasation and survival in the circulation

With enhanced motility and invasiveness induced by EMT, cancer cells can invade the vasculature and travel to distant organs. The presence and viability of disseminated cancer cells (CTCs) in the blood stream are correlated with advanced stages [29, 30] and poor prognosis in cancer patients [31]. Notably, CTCs can exist as single cells or oligoclonal clusters [12, 29, 32–35]. CTC cluster formation is reported to be largely induced by partial EMT, while complete EMT leads to single cell dissemination [12, 34]. Instead of being generated by intravascular aggregation of single CTCs, the CTC clusters are likely to derive from existing clusters of cells in the primary tumors and they are identified to be positive for plakoglobin [34] or Keratin14 (K14) [35]. Interestingly, plakoglobin is mainly involved in desmosome and adhesion complex formation, which are enriched in epithelial cells, indicating a requirement of an epithelial feature in CTC clusters formation. Consistently, RNA-seq analysis also showed an enrichment for desmosome and hemidesmosome adhesion complex genes in K14⁺ breast cancer cells [35]. This may support intercellular connection in CTC clusters and thus mediate collective migration [35]. Indeed, tumor spheres formed by cancer cells with partial EMT maintained epithelial phenotypes and exhibited collective invasion [12]. Moreover, Aiello et al showed single CTCs from complete-EMT tumors lacked E-cadherin expression, while tumor cell clusters from partial-EMT tumors retained E-cadherin only at the cell-cell junctions rather than cell surface [12], supporting the notion that epithelial phenotype maintained during partial EMT is involved in promoting cell cluster formation and collective migration. Although CTC clusters are rare compared to single CTCs, they are highly metastatic, as tail vein injection of artificial clustered tumor cell leads to enhanced lung metastasis [35] and reduced overall survival [34] in breast cancer models. Clinical analysis also shows a correlation between the presence of CTC clusters and poor prognosis in breast or prostate cancer [34]. CTC clusters are more metastatic than single CTCs due to their enhanced entrapment within capillaries of distal organs and increased resistance to apoptosis in the blood stream [34].

In the blood stream, tumor cells encounter harsh conditions and an overwhelming majority of them could not survive such stress, as revealed by apoptotic CTCs found in patients [30, 36]. To survive and finally colonize distal organs, CTCs have to overcome a series of obstacles including anoikis, shear stress and immune attack [37]. E-cadherin loss, which is one of the hallmarks of EMT, has been elucidated as a mechanism to develop resistance to anoikis in several studies [38–42]. Importantly, Cieply et al showed that EMT upregulated a stem cell marker CD44S which confers resistance to anoikis, suggesting that EMT can synergize with cancer stemness to maintain cell viability [42]. Additionally, CTCs can also associate with platelets and activate the coagulation cascade to form thrombi [43] by producing tissue factors [44]. Those thrombi not only exist as physical barriers to protect the tumor cells from sheer stress in the blood flow [45] and immune attack [46, 47], but also enhance EMT by platelet-derived TGF-β [43]. Interestingly, EMT has been shown to promote coagulation through upregulating tissue factors, as silencing Zeb1 not only inhibited EMT-associated transcription factor expression but also reduced clot formation [44]. Thus, the bi-directional regulation of EMT and coagulation may comprise a positive feedback loop to enhance cancer cell survival in the circulation. Microtentacles formation is indicated as another means to resist shear stress and anoikis. These are microtubule-derived membrane protrusions induced by detachment [48] and they are supported by vimentin, α-tubulin and detyrosinated tubulin [49, 50]. Interestingly, CTCs from breast cancer patients can aggregate with each other or with blood cells through microtentacles, with more microtentacles being observed in CTC clusters [50], indicating that microtentacles may be a way to promote and maintain CTC cluster formation.

Another difficult challenge CTCs have to cope with is immune surveillance. Multiple pathways are implicated in immune evasion of cancer cells. For example, cancer stem cells are found to be defective in antigen processing pathways caused by decreased expression of transporter associated with antigen processing (TAP) genes [51]. Programmed cell death-ligand 1 (PD-L1) expression is another way to compromise immune killing mediated by cytotoxic T cells (CTLs). Importantly, the reciprocal regulation between EMT and PD-L1 expression was shown to play important roles in immune escape [52]. It has been revealed that mammary tumor cells derived from more epithelial carcinoma cell lines express high levels of major histocompatibility complex class 1 (MHC-I), which supports antigen presentation, and low levels of PD-L1, while cells from more mesenchymal breast carcinoma cell lines exhibit low levels of MHC-I and high levels PD-L1 [53]. In another study by Hsu et al, EMT can transcriptionally induce N-glycosyltransferase STT3 through β-catenin, and subsequent glycosylation of PD-L1 by STT3 can stabilize PD-L1, leading to immune suppression [54]. Conversely, PD-L1 can also induce EMT in breast cancer [55], renal cancer [56], glioblastoma [57] etc. Furthermore, cancer cells can also escape natural killer (NK) cell dependent killing, which is another pivotal way to eliminate tumors by the immune system. The major mechanism involves downregulation of ligands that can activate NK cells through binding to their stimulatory receptors [58–60]. Interestingly, Lo et al found that CTC clusters exhibit higher resistance to NK cell killing, and NK cell depletion selectively increases lung metastasis derived from single CTCs, but not metastasis from CTC clusters [60]. Thus, it is possible that partial EMT, which promotes CTC cluster formation, enables immune evasion of CTCs from NK cells. Taken together, EMT plasticity seems to promote cancer metastasis not only by enhancing a migratory phenotype, but also by offering cancer cells more advantages to survive.

Bone metastasis predilection and osteomimicry

Although DTCs in the bone are mainly reported in breast cancer and prostate cancer patients, with their presence associated with poor prognosis [61–63], DTCs are also frequently detected in the bone marrow of patients with gastric cancer [64, 65], colorectal cancer, pancreatic cancer [65] and other types of cancers [66]. This is probably due to the discontinuous endothelium in the bone sinusoids [67], which enables passive entry of cancer cells to the bone marrow without the need of active extravasation. However, bone metastasis is mostly seen in breast cancer and prostate cancer patients, indicating that specific traits of cancer cells are required for bone metastasis. The ability or preference of homing to the bone, which is usually referred to as bone tropism, represents specific cellular programs of cancer cells (Figure 2A). For example, hypoxia in the estrogen-receptor negative breast tumors can induce lysyl oxidase (LOX) expression in the tumor mass [68]. LOX can then promote NFATc1-driven osteoclastogenesis in the bone, thus disrupting normal bone homeostasis and promoting the formation of pre-metastatic lesions to facilitate bone colonization [68]. Meanwhile, high throughput profiling revealed a series of bone metastasis-associated gene signatures [69–74], and many of them have been proven to be mediating the process of bone metastasis. For instance, IL-11 is upregulated in highly bone metastatic cells [69], and it activates JAK1/STAT3 pathway to induce c-Myc expression, which allows osteoclastogenesis and leads to osteolytic bone metastasis [75]. C-X-C motif chemokine receptor 4 (CXCR4), the receptor for stromal cell-derived factor 1/C-X-C motif chemokine 12 (SDF-1/CXCL12), is another protein involved in bone tropism. First, CXCR4 is found to be upregulated in highly bone metastatic cells lines [69] as well as in human bone metastases of breast cancer patients [73]. In a metastatic prostate cancer model, Sun et al found a positive correlation between the levels of SDF-1 and tissues in which metastatic lesions were observed [76], which implicate SDF-1/CXCR4 signaling axis as a predilection factor for bone metastasis. Consistently, in vitro study reveals that CXCR4 mediates tumor cell migration in response to SDF-1 [77]. Furthermore, application of CXCR4 antibodies significantly reduces the total metastatic load compared to IgG control, which validates the importance of SDF-1/CXCR4 axis in directing cancer cells homing to the bone [76]. Notably, both IL-10 and SDF-1/CXCR4 axis are also important mechanisms regulating normal physiology of bone. Briefly, IL-11 produced by bone marrow stromal cells can induce osteoclasts formation and bone resorption [78]; the SDF-1/CXCR4 axis controls the homing of blood stem cells to and their exit from the bone [79, 80]. Additionally, there are numerous other molecules that were found to be involved in both maintaining normal bone functions and promoting bone metastasis, such as Cadherin-11 [81–84], osteopontin (OPN) [69, 85, 86], connective tissue growth factor (CTGF) [69, 73, 87] and Runt-related transcription factor 2 (RUNX2) [88–90]. This suggests that cancer cells may exploit some existing mechanisms in normal physiology to facilitate bone metastasis. Consistent with this notion, metastatic prostate cancer cells in the bone are reported to compete with hematopoietic stem cells (HSCs) for occupancy in the HSC niche, and they can be mobilized to the periphery blood by methods routinely used for HSC mobilization such as G-CSF treatment [91].

Figure 2. The course of bone metastasis.

To successfully colonize the bone, cancer cells need to invade the local tissues and leave the primary tumor, intravesate into blood vessels, followed by traveling to and seeding in the bone. (A) A schematic graph depicting a series of factors implicated in bone metastasis predilection. The ability or preference of cancer cells to metastasize to the bone is called bone tropism. Cancer cells take advantage of some existing mechanisms used by normal physiology to facilitate bone metastasis, and most of the predilection factors are also found to be involved in normal bone homeostasis, such as the SDF-1/CXCR4 axis, cadherin-11, OPN, CTGF and RUNX2. Moreover, cancer cells often gain some features of normal bone cells in depositing ECM. For example, prostate/breast cancer cells or tumors can deposit osteocalcin, BSP, and HA to the ECM, whereas normal prostate/breast epithelial cells cannot. (B) After arriving at the bone, multiple pathways as well as cell-cell interactions are involved in mediating successful seeding. The interactions between cancer cell-derived E-cadherin and osteogenic N-cadherin is reported to activate mTOR signaling and confer growth advantages. Cadherin-11 can also promote bone metastasis through homophilic binding. Interactions between the endothelial cells and cancer cells are also important. Fucosylated Glg1 on cancer cells can associate with E-selectin on endothelial cells of bone capillary. This binding promotes MET in cancer cells and also activates Wnt signaling to promote cell stemness, leading to bone metastasis. Abbreviations: SDF-1, stromal cell-derived factor-1; CXCR4, C-X-C chemokine receptor 4; OPN, osteopontin; CTGF, connective tissue growth factor; RUNX2, Runt-related transcription factor-2; BSP, bone sialoprotein; HA, Hydroxyapatite; ECM, extracellular matrix; mTOR, mammalian target of rapamycin; Glg1: Golgi glycoprotein 1; MET, mesenchymal-to-epithelial transition.

Cancer cells can also gain features similar to bone cells in depositing ECM. In breast and prostate cancer, tumor cells are found to give rise to local mineralization similar to the bone [92–95], which is termed osteomimicry [96]. For instance, in prostate cancer, osteocalcin (OC) and bone sialoprotein (BSP), which are bone matrix proteins abundantly expressed by osteoblasts, are found to be expressed by malignant but not normal prostate epithelial cells [93]. This phenomenon is found to be promoted by treatment of conditioned medium from prostate cancer cells or bone stromal cells, which subsequently activates protein kinase A (PKA)-cyclic AMP (cAMP)-response elements (CRE) signaling axis and enhance the OC/BSP promoter [93]. Interestingly, the extent of OC/BSP promoter activated is positively correlated with the aggressiveness of prostate cancer cell lines that produce the conditioned medium, indicating a positive correlation between osteomimicry and prostate cancer malignancy [93]. A similar phenomenon is also observed in breast cancer. Research on mammary microcalcification shows that calcium oxalate deposition is mostly found in benign lesions, while hydroxyapatite (HA) deposition, which is found in normal physiological mineralization by osteoblasts, is associated with both benign and malignant tumors [92]. Furthermore, HA can enhance tumor cell migration in vitro, whereas calcium oxalate cannot [94], suggesting a tumor promoting role of HA deposition. Moreover, tumor cells with mesenchymal characteristics are enriched in tumors with microcalcification, suggesting a role of EMT plasticity in eliciting osteomimicry programs [95]. Another in vivo study also revealed that breast cancer osteomimicry induced by forkhead box F2 (FOXF2) promotes bone specific metastasis [97], indicating that the intrinsic osteomimetic properties acquired by cancer cells may contribute to their predilection to bone metastasis.

Mesenchymal-epithelial transition and colonization in the bone

The hypothesized involvement of mesenchymal-epithelial transition (MET), the reversed process of EMT, in metastatic seeding in distant organs was first proposed based on pathological observations that metastatic tumors often display an epithelial phenotype similar to their primary counterparts, indicating the requirement of epithelial state for successful colonization. Consistently, DTCs in distal organs are typically detected using epithelial markers, such as cytokeratin [63, 64, 66, 98–101]. Supporting this hypothesis, opposite trends in expression of certain EMT markers or inducers were observed upon metastasis. For example, E-cadherin loss is a hallmark of EMT, while its expression is required to form metastases [18]; Prrx1 confers migratory and invasive properties, but its loss is needed to form distal metastasis [102]; miR-200s inhibits Zeb1/2 to maintain E-cadherin expression [103] and its downregulation promotes EMT in the early phase, but regaining miR-200s expression in the late stages promotes epithelial phenotypes and enables metastatic colonization [104]; TWIST1 also promotes EMT, whereas its shutoff at distal organs is essential for DTCs to proliferate and form metastases [105]. Furthermore, in a study involving single-cell profiling and clustering analysis, Lawson et al discovers that cells from high-burden metastases, which represents late phase of metastasis, cluster with primary tumor cells; in contrast, cells from low-burden metastases, which are at early phase of metastasis, appear as a distinct population from the primary tumor cells, indicating that cancer cells manage to switch to a state similar to the primary tumor before forming metastases at distal organs [19]. Notably, MET is not a complete reverse process of EMT, and those genes being oppositely regulated during MET are not merely promoting an epithelial phenotype. Rather, they can also elicit novel programs to help cancer cells adapt to the new environment. For instance, loss of Prrx1 not only promotes MET conversion but also confers stem cell properties to BT-459 breast cancer cells [102]. Furthermore, re-expression of Prrx1 in Prrx1-negative MDA-MB-231 cells leads to reduced stemness, as revealed by less potency in forming secondary and tertiary mammospheres [102]. Similarly, miR-200s promote E-cadherin expression; meanwhile, they directly target Sec23a and suppresses Sec23a-mediated secretion of extracellular proteins, which results in reduced secretion of some metastasis suppressors, including Igfbp4 and Tinagl1 [104].

Besides MET, interactions between cancer cells and resident cells in the bone are also required for successful bone dissemination (Figure 2B). Wang et al found that heterotypic adherens junctions involving cancer-derived E-cadherin and osteogenic N-cadherin can activate mTOR pathway in cancer cells, thus driving tumor growth [106]. Similarly, cadherin-11, which is also known as osteoblast-cadherin, is found to promote bone metastasis through homophilic binding with bone derived cadherin-11 [81]. A recent study by Esposito et al has illustrated an E-selectin dependent program coupling MET with cell stemness [107]. They found that fucosylated Glg1 binds to E-selectin expressed on bone marrow endothelial cells and induces a non-canonical MET program, where cancer cells exhibit an epithelial phenotype, as revealed by EpCam and Keratin-14 staining, whereas master transcriptional regulators of traditional EMT, including Snail1/2, Twist1/2 and Zeb1/2, remain unchanged [107]. Furthermore, Wnt pathway is activated after this binding, and Sox2/9 are induced to promote cancer stemness as well as bone metastasis formation [107]. Similarly, bone marrow endothelium-derived E-selectin is also reported to provide pro-survival signaling in acute myeloid leukaemia (AML) blasts through activating PI3K/Akt/NF-κB pathway [108].

Survival and outgrowth in the bone

Arriving at the bone is just the beginning to form bone metastasis, and disseminated cancer cells must manage to maintain and expand to successfully colonize the bone. In clinical practice, the time frame of forming detectable distal metastasis from disseminated tumor cells range from years to decades [109, 110], indicating the presence of a dormant state before complete tumor outgrowth (Figure 3A). Tumor dormancy refers to the state in which patients appear asymptomatic and have no clinically detectable metastatic lesions, but harbor tumor cells in their bodies [109]. This can occur in the tumor mass level, where tumor growth is counteracted by other mechanisms such as apoptosis, ineffective vascularization, immunosurveillance, etc. [109]. At the cellular level, dormancy usually refers to the growth arrest of tumor cells, including cell quiescence and senescence [109]. Quiescence describes the state of cell cycle arrest, which is often referred to as the G0 phase [111]. This state is reversible, and cells are able to exit quiescence and resume proliferation in more permissive conditions. Whereas senescence is irreversible, and represents an alternative cell fate to apoptosis when cells encounter severe conditions, such as extensive DNA damage [111]. For research cited in this review, the term ‘cell quiescence’ is used to describe a lack of cell proliferation markers such as Ki67, while the term ‘cell senescence’ refers to the detection of cell senescence markers such as positive β-galactosidase staining.

Figure 3. Surviving and thriving in the bone.

Cancer cells must manage to survive and maintain in the bone before bone metastasis outgrowth. Here we show two consequences of bone DTCs, namely, dormancy (A) and outgrowth (B). (A) Endothelial cells contribute significantly to dormancy induction. TSP-1 derived from bone vasculatures promotes dormancy. Additionally, DARC on endothelial cells can interact with KAI on cancer cells and lead to upregulation of p21 and downregulation of TBX2, which consequently lead to cell dormancy. TGF-β2 in the microenvironment can also trigger dormancy through activating p38α/β and thus inhibiting CDK4 functions. Autocrine Wnt inhibition by DKK1 not only confers a dormant cell state but also enables evasion from NK cell-mediated killing. Meanwhile, SDF-1/CXCR4 is also reported to promote cell survival and this effect requires Src activity. Furthermore, Src activation can inhibit TRAIL-mediated apoptosis to enhance cell survival. (B) The role of Jagged1/Notch signaling in eliciting bone metastasis outgrowth has been clearly demonstrated. Cancer cell-derived Jagged1 can bind to Notch on osteoblasts, which stimulates IL-6 secretion by osteoblasts. IL-6 can then promote tumor growth. Cancer cells can also associate with osteoclast precursors through Jagged1/Notch interactions, resulting in osteoclast maturation and osteolysis initiation. TGF-β released upon osteolysis will in turn promote Jagged1 expression in cancer cells and form a vicious cycle of osteolytic bone metastasis. In addition, VCAM-1 secreted by cancer cells also promotes the recruitment and activation of pre-osteoblasts through binding to integrin α4β1. Notably, Jagged1/Notch also contributes to chemoresistance, as chemotherapy induces Jagged1 expression on osteoblast, which confers survival advantage to cancer cells through interacting with cancer cell-derived Notch. Abbreviations: DKK1, Dickkopf-1; TSP-1, Thrombospondin-1; DARC, Duffy antigen receptor for chemokines; TBX2, T-Box transcription factor 2; TRAIL, TNF-related apoptosis-inducing ligand; DR4/5, death receptor 4/5; IL-6, Interleukin-6; VCAM-1, vascular cell adhesion protein 1.

Ghajar et al reveals that in breast cancer, dormant DTCs reside on the microvasculatures of lung, bone marrow and brain, and endothelial derived Thrombospondin-1 (TSP-1) promotes breast cancer cell quiescence [112]. Interactions between cancer cells and endothelial cells can also induce dormancy. For example, KAI (CD82) on tumor cells can directly interact with Duffy antigen receptor for chemokines (DARC) on vascular endothelial cells, leading to tumor cell senescence through downregulation of T-Box transcription factor 2 (TBX2), a senescence repressor and upregulation of p21, a cell cycle inhibitor [113]. In addition to regulations by endothelial cell derived signals, multiple pathways, including p38, ERK, BMP, TGF-β and hypoxia signaling, are also reported to be involved in dormancy induction [114–121]. For instance, in a head and neck squamous cell carcinoma (HNSCC) model, strong TGF-β2 signaling in the bone marrow activates p38α/β, which leads to downregulation of cyclin-dependent kinase 4 (CDK4) and prolonged dormancy, while the low levels of TGF-β2 in the lung allows DTCs to go through short-lived dormancy followed by metastatic growth [115]. Furthermore, the dormancy promoting role of TGF-β2 is found to require the interactions between Axl and growth arrest specific 6 (Gas6) in prostate cancer cells [122]. Hypoxia is another important source of signal in inducing dormancy, and it has been reported that hypoxia in the primary tumors can present dormant features in a subset of tumor cells, which can be carried over by DTCs in distal organs [123].

Before successful metastases formation, DTCs must receive a series of survival signals to help maintain themselves in the bone marrow. SDF-1/CXCR4 axis is one of the best studied. Onone hand, as mentioned earlier in this review, this signaling axis helps maintain the residency of cancer cells in the bone [76, 77, 80, 91, 124]. On the other hand, it has been reported that bone stromal cell derived SDF-1 can stimulate cell proliferation and migration through interacting with CXCR4 on cancer cells, and this pro-survival privilege can protect cancer cells from cytotoxic chemotherapy [125]. Furthermore, Zhang et al revealed that Src activity is required for cell survival mediated by Akt activation in response to SDF-1 [126]. In addition, Src activation also confers resistances to apoptosis mediated by TNF-related apoptosis-inducing ligand (TRAIL) [126]. Meanwhile, DTCs need to overcome the threat from immunosurveillance. First of all, the bone marrow represents a favorable immune environment where regulatory T cells (T_reg) provides immune privilege by mediating immune suppression [127]. Secondly, there are numerous regulators mediating cancer cell immune evasion. It has been reported that loss of tumor -intrinsic interferon signaling in prostate or breast cancer cells leads to reduced immunogenecity, making them more resistant to immune attack [128, 129]. In a study by Malladi et al, latency competent cells (LCC), which are cancer cells that remain dormant in distal organs but still retain their tumor initiating abilities, were isolated and characterized [130]. They found that those cancer cells highly express Dikkopf-1 (DKK1), which inhibits Wnt pathway to promote cell quiescence and to enhance downregulation of natural killer (NK) cell activators, thus enabling evasion from NK cell killing [130]. Consistently, DKK1 is proven to promote bone metastasis in several other studies [131–133].

The final step is to exit from dormancy and form bone metastasis (Figure 3B). This process involves a general switch from quiescence to proliferation, including downregulation of dormancy-associated genes such as TGFB2 and upregulation of cell-cycle promoting genes such as CDK2 [19]. Besides, some specific pathways have been investigated in depth. One example is the Notch pathway. Tumor-derived Jagged1 can promote bone metastasis through activating the Notch pathway in bone cells [134]. Jagged1 stimulates IL-6 secretion by osteoblasts and provides growth advantages for cancer cells; meanwhile, it also supports osteoclast maturation and promotes osteolysis [134]. Interestingly, Jagged1 can be activated by TGF-β [134], which is abundantly deposited in the mineralized bone matrix and is released upon osteolysis [2], suggesting this process as a positive feedback loop, which forms a vicious cycle of exacerbated bone metastasis and osteolysis. Moreover, Jagged1 in osteoblasts is found to be elevated by chemotherapy and subsequently promotes cancer cell survival as well as develops chemoresistance [135]. This indicates that Jagged1 expression in both the tumor and stromal compartments can confer growth advantages. Additionally, some adhesion proteins are also able to elicit dormancy exit. For instance, vascular cell adhesion molecule-1 (VCAM-1) can promote the transition from indolent bone micrometastasis to overt metastasis by recruiting and activating osteoclasts precursors [136, 137]. Importantly, this effect is dependent on the interactions between VCAM-1 and integrin α4β1, as antibodies against VCAM-1 and integrin α4 inhibit bone metastasis [136].

Concluding remarks

Despite unprecedented progress in cancer research, cancer metastasis remains mostly incurable. Most of the major obstacles in the treatment of metastatic cancer, including uncontrollable growth at the metastatic site, treatment resistance, and immune evasion are more or less related to phenotypic plasticity of cancer cells. As the first step in metastasis, EMT confers cellular plasticity and stemness, which enables a global transcriptomic reprogramming to provide transformed cancer cells pro-survival advantages. However, this cellular plasticity also offers new opportunities for targeted therapy. Taking advantage of EMT-induced plasticity, Ishay-Ronen et al revealed a potential strategy to trans-differentiate EMT-derived breast cancer cells to pre-mitotic and even functional adipocytes [138]. Thus, it is possible that, with more in-depth research on cancer cellular plasticity in the future, we will be able to make cellular plasticity program controllable. Cancer cellular plasticity functions to enhance the maintenance and survival of cancer cells when they arrive at the bone. It also helps the transition between dormant and proliferating states that allows cancer cells to thrive under favorable conditions. For cancer patients, dormancy represents the optimal window for treatment or prevention of metastasis, yet no clinically practical therapy is available [139]. This may be due to the lack of complete understanding of dormancy. With current available models, it is hard to perfectly mimic human cancer dormancy which can last for years to decades. Moreover, it is difficult to distinguish genes eliciting dormancy or dormancy exit from genes that generally inhibit or promote tumor growth. Optimized animal models and experimental tools of higher resolution such as single cell sequencing and imaging may be applied in the future to characterize this process more accurately. Furthermore, since cellular plasticity programs also play important roles in promoting cell survival upon treatment of chemotherapies [5, 108, 135], which are still one of the first-line therapies for treating bone metastasis, blockage of such signaling pathways mediating cellular plasticity may reduce the probability to develop chemoresistance or even render chemoresistant tumors sensitive to chemotherapy again. One example is the anti-Jagged1 antibody, which sensitizes bone metastasis to chemotherapy by blocking the protective effect of osteoblastic Jagged1 induced by chemotherapy [135]. Undoubtedly, more thorough investigations of cellular plasticity will give rise to more candidate drug targets and eventually offer more options for patients with metastatic cancers. In addition to cancer cell plasticity, stromal cell plasticity represents another non-negligible aspect in bone metastasis. It has been reported that cancer-associated endothelial cells can be converted to osteoblasts under the stimulation of BMP4 secreted by prostate cancer cells, thus driving osteoblastic bone metastasis [140]. More detailed delineation regarding the compartments in which cellular plasticity program is functional will help to develop target therapies more precisely.

Lastly, there are several rising fields in cancer biology that are worth further investigation. Besides cancer immunology which is under extensive investigation nowadays, epigenetic modifications on RNA including adenosine-6-methylation (m6A) is indicated in influencing various aspects of cancer progression [141]. Neurogenesis in tumors is also correlated with cancer progression and metastasis [142], including bone metastasis [143]. Future studies on emerging fields like these will supplement our current knowledge and give a broader and more complete picture for cancer metastasis.

Highlights:

• Cellular plasticity promotes bone metastasis by conferring pro-survival advantages

• EMT and MET are two important programs promoting cell plasticity in bone metastasis

• Osteomimicry of cancer cells contributes to the predilection for bone metastasis

• Cancer cells adopt mechanisms used by normal cells to adapt to various environments

• Mechanistic insights of cellular plasticity may lead to new targeted therapies