The emerging molecular machinery and therapeutic targets of metastasis

Hurdles in eliminating metastasis-associated mortality

Metastasis is a multi-step process which begins when primary tumor cells break away from their neighboring cells, such as nearby stromal cells, and invade through the basement membrane. Subsequently, metastasizing cells enter the circulation (intravasate), either directly or via lymphatics, and then home to distant organs where they exit the vasculature (extravasate). Eventually, tumor cells that successfully adapt to the new microenvironment proliferate from micrometastases into clinical detectable metastatic tumors (Figure 1) [1]. Although great advances have been made in combating cancer, particularly in its early stages, metastasis remains a formidable and frequently fatal challenge [2–4]. It is becoming increasingly clear that the seeds of metastasis are present in many cases of early disease [3, 5], leading to deaths that might be prevented. A number of processes, factors, and signaling pathways have been implicated in regulating metastasis, including epithelial-mesenchymal plasticity, cancer stem cells, non-coding RNAs, cytokines, hormones, and receptor tyrosine kinase (RTK) pathways, with the list of determinants of metastasis still expanding. However, few molecules have been translated into effective metastasis prevention or treatment in the clinic.

Over 90% of cancer-related deaths are caused by metastasis. For instance, breast cancer, the most common malignant disease in women, begins as a local disease and later metastasizes to lymph nodes and other organs. The most common sites of breast cancer metastasis are vital organs, such as the lung, liver, bone, and to a lesser extent, the brain [6, 7]. National Cancer Institute (NCI) Surveillance, Epidemiology, and End Results (SEER) data indicate that the percentages of patients with localized, regional, metastatic, or unstaged breast cancer at diagnosis are 61%, 32%, 5%, or 2%. Their corresponding five-year survival rates are 98.5%, 84.6%, 25%, and 49.8%, respectively. However, many patients with localized or regional cancers show evidence of local invasion or disseminated, micrometastatic tumor cells at diagnosis, meaning that it is too late to stop the early steps of metastasis [8]. Therefore, for those 93% of patients, preventing metastatic colonization – the growth from disseminated, micrometastatic tumor cells to macroscopic metastases – holds the most therapeutic promise. For those 5% of patients with metastatic disease at diagnosis, shrinking established metastases must be the goal.

Surgery, radiation therapy, and chemotherapy can eliminate many primary tumors, and thus approaches to preventing metastatic colonization should be most effective as adjuvant therapy [8]. The major roadblock to devising adjuvant metastasis prevention treatments is that the current clinical trial system is not designed to test metastasis preventive drugs [9]. In the current setting, running metastasis prevention trials on patients with early-stage cancer would be prohibitively lengthy and costly and would require many thousands of patients. Therefore, drugs today have to induce regression of established metastatic tumors in late-stage cancer patients who failed standard treatment, if those drugs are to receive regulatory approval and to be advanced to adjuvant metastasis prevention trials [2, 9]. This is in contrast to the preclinical setting, where most anti-metastatic agents that have been tested prevent the formation of metastases but do not shrink established metastatic tumors [2, 9]. It has been suggested that the format of clinical trials be changed to accommodate metastasis preventive drugs [9], but to do so would require a new approach to defining patient eligibility and to predicting drug response.

Ideally, trials of metastasis preventive drugs would enroll patients with early-stage disease who are at high risk of developing metastases as well as those who already have metastases and are at risk of developing more [9]. One major obstacle to this ideal, however, is that we have no good means of identifying these high-risk patients. We also do not know how to select patients who might benefit from specific metastasis preventive agents. To surmount these barriers and facilitate metastasis prevention trials, we need to find better prognostic markers for metastasis, effective anti-metastatic drugs, and predictive markers for drug response.

In addition to the lack of appropriate clinical trials for metastasis prevention, the heterogeneity of metastatic tumor cells may account for the failure in therapeutic targeting of a specific pathway, since different subpopulations of metastatic tumor cells could employ distinct molecular machinery. For instance, treatment of patients with triple-negative breast cancer (TNBC), which metastasizes more frequently than other breast cancer subtypes and is associated with poor clinical outcomes, has been challenging due to the heterogeneity of this disease and the lack of well-defined therapeutic targets [10]. Thus, there is a pressing need to understand tumor heterogeneity and elucidate the mechanisms by which different metastases originate from different subpopulations of cancer cells co-existing within a tumor.

In this review, we dissect the processes of metastatic progression. These processes depend on genetic and epigenetic aberrations in tumor cells and alterations in the associated microenvironment. In addition to reviewing the emerging molecular determinants and therapeutic targets at each step of the invasion-metastasis cascade, we discuss the molecular basis of the cellular plasticity of tumor cells. Such plasticity is likely to underlie therapy resistance and metastatic relapse, which suggests the importance of understanding tumor heterogeneity and the need to develop new combination therapies to target all types of cancer cell subpopulations, including cancer stem cells (CSCs), circulating tumor cells (CTCs), disseminated tumor cells (DTCs), and differentiated cancer cells.

Role of epithelial-mesenchymal plasticity and cancer stem cells in metastasis

The ability of cancer cells to metastasize depends on their genetic and epigenetic alterations as well as the microenvironmental cues they receive. Recent studies suggest that many of the properties associated with invasion and metastasis do not arise as purely cell-autonomous processes; instead, the surrounding tumor stroma becomes ‘activated’ during primary tumor progression and begins to release signals, such as transforming growth factor (TGF)-β, hepatocyte growth factor (HGF), tumor necrosis factor (TNF)-α, Wnt, and platelet-derived growth factor (PDGF). Subsequent adaptation of carcinoma cells to these heterotypic signals can lead to acquisition of highly-malignant, cell-biological traits through processes such as the epithelial-mesenchymal transition, EMT (Figure 2) [11].

EMT is characterized by repression of epithelial marker expression, acquisition of mesenchymal markers, loss of cell adhesion, and increased cell motility and invasiveness (Figure 2) [12, 13]. During development, the EMT program, together with its reverse process, mesenchymal-epithelial transition (MET), enables cells to move from one part of the embryo to another and then differentiate, which contributes to the formation of various organs [13, 14]. In adult cells, the EMT program is usually silent but can be reactivated in processes such as wound healing [15]. Recent studies suggest that cancer cells can resurrect this developmental program: while inducing EMT in epithelial tumor cells facilitates migration, invasion, and dissemination, the MET process enables metastatic colonization [16, 17]. Microenvironmental stimuli emanating from the tumor stroma, such as TGF-β, can activate the expression of several master regulators of embryogenesis, including transcription factors Twist [18], Snail [19, 20], Slug [21], ZEB1 [22], and ZEB2 [23], which have been identified as inducers of EMT and tumor metastasis (Figure 2). Despite initial skepticism, in vivo models and studies investigating EMT features in clinical tumor samples have provided strong evidence for the involvement of EMT and MET in metastasis [24, 25]. In an elegant study, Yang and colleagues generated mice with skin-specific, doxycycline-inducible Twist transgene and induced skin tumors using chemical carcinogens. Then either oral (to induce Twist in both primary and disseminated skin tumor cells) or topical (to induce Twist in only primary skin tumor cells) administration of doxycycline promoted EMT, tumor invasion, and dissemination. Strikingly, mice receiving topical doxycycline had much more lung metastases than mice receiving oral doxycycline, and the metastatic tumors from mice treated with oral or topical doxycycline lost Twist expression and had epithelial features, indicating reversion of EMT [16]. These findings suggest that both EMT and MET are essential for tumor cells to accomplish the invasion-metastasis cascade in certain cancers. However, it should be noted that EMT and MET may not be the prerequisite for metastasis in all tumor types; alternative mechanisms, such as “collective invasion” [26] and “amoeboid movement” [27], have been proposed.

Another model proposes that cancer stem cells (CSCs), which are defined operationally as tumor-initiating cells, are responsible for generating secondary tumors [28]. Interestingly, induction of the EMT program in carcinoma cells can generate cells with properties of CSCs (Figure 2) [29, 30]. Hence, the invasion and intravasation steps of metastasis may involve EMT, which confers both motility and ‘stemness’ on carcinoma cells, while the metastatic colonization step may require the MET program, which facilitates the differentiation of CSCs into non-CSCs. Moreover, the epithelial-mesenchymal plasticity may underlie the non-CSC-to-CSC plasticity. For instance, a recent study demonstrated that TGF-β-induced expression of ZEB1 can drive basal breast cancer cells to undergo EMT and convert from non-CSC state to CSC state [31], while ZEB1-targeting microRNAs (miRNAs), such as miR-205 and the miR-200 family, have been found to promote MET and suppress CSC properties [32–34]. Interestingly, ZEB1 binds to the promoter region of miR-200 genes and represses their transcription, forming a doublenegative feedback loop [35]. Consistent with its MET-inducing effect, the miR-200 family has been found to suppress cancer cell migration and invasion [33, 35] but enhance metastatic colonization after tumor cells have already disseminated [36, 37].

The implication of EMT and CSCs in metastasis has offered potential opportunities for therapeutic intervention [24, 25]. Small-molecule inhibitors of ALK5, MEK, and Src were found to block EMT induction by HGF, epidermal growth factor (EGF), or insulin-like growth factor (IGF)-1 [38], while rapamycin (mTOR inhibitor) and 17-allylamino-17-demethoxygeldanamycin (17-AAG; HSP90 inhibitor) were identified as inhibitors of TGF-β-induced EMT, migration, and invasion [39]. These approaches designed to inhibit EMT induction will likely block tumor cell invasion in early-stage carcinomas; however, in patients with disseminated, micrometastatic tumor cells, killing mesenchymal cancer cells or preventing MET should be the goal. For instance, salinomycin was identified as a compound that induced selective killing of mesenchymal-type breast cancer cells and reduced the proportion of breast CSCs [40]. To date, the signals that trigger MET at the metastatic site remain unclear. Identifying such signals may reveal new therapeutic targets to prevent metastatic colonization.

Molecular determinants of the metastatic process

Determinants of metastatic colonization: key regulators of the bottleneck of metastasis

The organ distribution of metastases not only depends on the vascular pattern, but also reflects the adaptability of tumor cells to specific organ microenvironment. In a pioneering study, Kang, Massagué, and colleagues compared gene expression profiles of MDA-MB-231 human breast cancer cells and the bone metastatic subline derived from intra-cardiac injection of the parental cells, and identified a set of four genes (IL11, CTGF, CXCR4, and MMP1) that act collectively to facilitate metastatic colonization in the bone [95]. Similar approaches have been used to identify genes that regulate breast cancer colonization in the lung [96] and brain [97], which revealed the molecular basis of organ tropism.

Certain physiological processes can be hijacked by cancer cells during metastatic colonization. The bone undergoes remodeling reflecting the balance between osteoclasts which degrade mineralized bone and osteoblasts which reconstruct the bone. Osteoblasts secrete: (1) receptor activator of NF-κB ligand (RANKL), which binds to its receptor (RANK) displayed by the osteoclast precursor to induce its maturation into the osteoclast, and (2) osteoprotegerin (OPG), a soluble decoy receptor that binds secreted RANKL, preventing it from interacting with the RANK receptor. Osteolytic cancer cells often overexpress osteoclast-inducing factors, such as parathyroid hormone-related protein (PTHrP), IL-1, IL-6, and IL-11, which act on osteoblasts to stimulate production of RANKL, leading to osteoclast maturation and bone degradation; on the other hand, matrix-embedded cytokines and growth factors released from the dissolved bone matrix, such as TGF-β and IGF, act on DTCs to stimulate production of osteoclast-promoting factors. This positive feedback loop is often referred to as “the vicious cycle of osteolytic bone metastasis” [98]. Approaches to breaking this vicious cycle and treating bone metastasis include bisphosphonates, OPG, and PTHrP-neutralizing antibodies; among them bisphosphonates are being used clinically to prevent or treat diseases of bone loss, including osteoporosis, Paget’s disease, and cancers that cause osteolytic metastasis. Bisphosphonates inhibit osteoclastic bone resorption by promoting osteoclasts to undergo apoptosis [98].

DTCs at the distant organ site can either grow into clinically significant metastases, or remain dormant due to the lack of proliferative signals and/or the presence of anti-proliferative signals in the new environment, obstacles that they need to overcome in order to proliferate from occult micrometastases into macroscopic secondary tumors. Dormant DTCs, the seed of distant relapse, are extremely difficult to eradicate because they are clinically asymptomatic and resistant to conventional or targeted therapies. Although evidence indicates that CSCs, cell cycle regulators, epigenetic factors, and microenvironment play important roles in regulating tumor cell dormancy, our understanding of this field is still limited. Due to its importance in metastatic recurrence, the mechanism of dormancy and reactivation of DTCs has become an awakening field of cancer research [99, 100]. Fibrosis can induce tumor progression and metastasis not only through tumor-stroma interaction at the primary site, but also by reactivating dormant DTCs at the metastatic site. For instance, the transition of breast DTCs from quiescence to proliferation is mediated by binding to fibronectin or type I collagen (Col-I) often found in the fibrotic metastatic lesions, which induces otherwise dormant breast cancer cells to proliferate through β1-integrin activation of Src and focal adhesion kinase (FAK); genetic or pharmacologic inhibition of this signaling cascade blocked cytoskeletal reorganization and cell proliferation in vitro and reduced metastatic outgrowth in vivo [101–103]. Fibrosis is associated with metastasis and poor prognosis in breast cancer, pancreatic cancer, and other cancers [2]. Anti-fibrotic drugs that have been developed for fibrotic diseases, such as the CTGF-neutralizing antibody (FG-3019) mentioned above, may prove useful as anti-metastatic agents. On the other hand, stromaderived growth-inhibitory signals, such as bone morphogenic protein (BMP) produced by the lung parenchyma, represent a barrier to metastatic colonization. Gain-of-expression of Coco, a secreted BMP antagonist, induces reactivation of otherwise dormant breast DTCs to proliferate and form metastases in the lung [104], suggesting that counteracting the anti-metastatic signals in the distant organ leads to metastatic outgrowth.

The link between metastasis and therapy resistance

Tumor cells with therapy resistance, including radioresistance and drug resistance, give rise to tumor recurrence and metastatic relapse [2]. Emerging evidence has suggested that some of the molecules that endow tumor cells with metastatic ability also confer treatment resistance. Therefore, targeting these molecules has the potential to overcome therapy resistance and to eliminate local and distant recurrence.

Recently, CSCs have been found to promote tumor radioresistance though activation of the DNA damage response. This was first reported in glioblastoma, in which glioma cells expressing the brain CSC marker CD133 are resistant to ionizing radiation because they are more efficient at repairing damaged DNA than the bulk of the tumor cells [105]. Later, similar findings were reported for other tumor types including breast cancer [106, 107]. CSCs are also believed to be resistant to chemotherapy due to high expression or activation of ATP-binding cassette (ABC) transporter proteins, aldehyde dehydrogenase (ALDH), anti-apoptotic proteins, pro-survival signaling components, and DNA repair molecules [108]. The association between EMT and CSC properties, including chemoresistance, radioresistance, and resistance to targeted therapies, has been reported by a number of studies (Figure 6) [109–117]. Does EMT itself or specific EMT regulators play a causal role in therapy resistance? Moreover, are all EMT inducers equal?

A recent study demonstrated that it is not the epithelial or mesenchymal state itself that dictates tumor radioresistance; instead, it is a specific EMT inducer, ZEB1, that regulates the response to radiation, whereas Twist and Snail do not affect radiosensitivity [117]. Mechanistically, radiation-induced activation of ATM kinase phosphorylates and stabilizes ZEB1, which in turn recruits USP7 and enhances its ability to deubiquitinate and stabilize CHK1, leading to increased DNA repair and radioresistance independent of EMT [117]. In parallel, ZEB1 represses its own negative regulator, miR-205, resulting in further increased levels of ZEB1 [118]. Radiation-induced upregulation of ZEB1 has been observed in breast cancer cells [117], lung cancer cells [119], and nasopharyngeal cancer cells [120]. These studies suggest that radiation treatment may cause therapy-induced radioresistance through ZEB1, leading to local and distant recurrence eventually. In support of this notion, among patients who received radiotherapy, those with high ZEB1 expression or low miR-205 expression in their breast tumors had much worse metastatic relapse-free survival outcome than those with low ZEB1 expression or high miR-205 expression [117, 118]. Moreover, therapeutic delivery of ZEB1-targeting miRNAs, including miR-205 [118] and miR-200c [121], sensitized tumors to radiation treatment in preclinical models. As a driver of EMT, ZEB1 can promote tumor metastasis and stemness by repressing E-cadherin and stemness-inhibiting miRNAs [22, 122], or by repressing other target genes including HUGL2 (also named LGL2, lethal giant larvae homolog 2), PATJ (Pals1-associated tight junction protein), and Crumbs3 [123, 124]. In addition, depending on the specific tumor type and treatment type, ZEB1 can employ EMTdependent and EMT-independent mechanisms to regulate resistance to chemotherapeutic agents (such as temozolomide, gemcitabine, 5-fluorouracil, cisplatin, and docetaxel) [110, 114, 115] and targeted therapies (such as the PI3K inhibitor and the EGFR inhibitor) [116, 125]. Taken together, ZEB1 represents a pleiotropically acting transcription factor that links EMT, metastasis, and therapy resistance. ZEB1-targeting agents, such as miR-200c and miR-205 mimics, may provide new therapeutic opportunities [126].

Concluding remarks

Metastasis is the leading cause of cancer-related death. Although significant progress has been made in understanding the mechanisms of tumor progression and metastasis during the past century, the knowledge of the molecular machinery governing metastasis is still incomplete. Heterogeneity that exists within both the primary tumor and the metastatic tumor may underlie the failure of cancer treatment, and thus it is critical to improve molecular characterization of heterogeneous metastatic cells. In addition, a number of approaches have been established for metastasis research, but until recently, there was a paucity of technologies for studying CTCs, DTCs, and metastatic dormancy and reactivation. CTC enrichment methods, single-cell sequencing techniques, patient-derived xenograft models, and other new tools will facilitate metastasis research and clinical development. Furthermore, despite the emerging new regulators of metastasis, the knowledge is rarely translated into clinical advances. There is a pressing need to develop novel biomarker-driven clinical trials for metastasis prevention and treatment.