Cancer and microenvironment plasticity: double-edged swords in metastasis

Abstract

Cancer initiates at one site (primary tumor) and, in most cases, spreads to other distant organs (metastasis). During the multi-step process of metastasis, primary tumor cells acquire cellular and phenotypic plasticity to survive and thrive in different environments. Moreover, cancer cells also utilize and educate microenvironmental components by reshaping them into accomplices of metastasis. Recent studies have identified a plethora of new molecular and cellular modulators of metastasis with dynamic or even opposite roles dominating the phenotypic plasticity of both tumoral and microenvironmental components. In this review, we will discuss their bi-potential functions and the possible underpinning mechanisms, as well as the implication in targeted cancer therapy.

Emerging role of bi-potential regulators in tumorigenesis and metastasis

Cancer is a leading cause of death globally, accounting for an estimate of 9.6 million deaths in 2018 [1]. Instead of primary tumors, the distant metastasis is the major cause of suffering and deaths of patients. Although devastating, metastasis is a multi-step process with very low efficiency [2]. In most cases, tumor cells must undergo a four-stage process to form a clinically observable metastasis. These include: (1) breakdown of the basement membrane and invasion of the tumor cells into the nearby stroma (local invasion), (2) accessing (intravasation) and survival in the circulation, (3) exit from the circulation and infiltration into the target organ parenchyma (extravasation), and (4) adaptation to the foreign microenvironment and outgrowth into macrometastases (colonization). In late-stage metastatic patients, the number of circulating tumor cells could be significant [3], but very few of these are able to develop into metastases. One possible reason is that most disseminated cancer cells are not equipped with the phenotypic plasticity to cope with the dynamic challenges of foreign microenvironments for metastatic colonization. Phenotypic advantages for tumor cells at primary sites, such as accelerated proliferation, may not be beneficial, or can be even antagonistic, for their survival and outgrowth in the new metastatic environments. Moreover, disseminated cancer cells at different metastatic sites might need additional adaptation schemes to thrive. This allows the possibility that some cancer cell-derived molecular modulators might elicit different, or even opposite, effects in the dynamic process of metastasis. Indeed, recent studies have revealed the existence of such paradoxical modulators of metastasis [4–7].

Moreover, tumor growth and metastasis also entail the involvement of various microenvironmental components, such as endothelial cells, fibroblasts, macrophages, neutrophils etc., which have been shown to play important roles in cancer progression [8]. This is particularly important in cancer metastasis, as metastatic cells lodge in target organs that have a set of microenvironmental factors that are different than those in the primary site. In the rate-limiting step of metastasis, colonization, cancer cells establish extensive interaction with the local stromal components. These interactions allow cancer cells to educate the microenvironment to tolerate and eventually support the outgrowth of metastasis [9–11]. However, it is also reported that the same stromal cells could help to eradicate the disseminated cancer cells or keep them dormant [12–14]. The specific underlying mechanisms of this apparent paradox of microenvironmental roles of metastasis have been extensively studied and important discoveries have been made recently as will be discussed below.

It is important to note that in metastatic cancer patients, cancer cells from the primary sites keep disseminating into blood vessels and landing at different target organs. Thus, cancer cells at different metastatic stages co-exist and interact with varied microenvironmental components. The heterogeneity of the metastatic cancer cells complicates mechanistic studies of the process, sometimes leading to paradoxical phenotypic outcomes by the same regulators and poses a daunting challenge in targeting them all. In this review, we will discuss the bi-potential molecular and cellular modulators of cancer metastasis, from both tumor cells and microenvironment and the mechanisms by which they exert the dichotomous functions. These modulators have the potential to become targetable vulnerability of metastasis, but their paradoxical roles also call for cautious manipulation to maximize therapeutic efficacy and safety.

Bi-potential signaling molecules that regulate cell-intrinsic phenotypes

Primary tumor growth and metastasis leverage different phenotypic characteristics of cancer cells. For example, high proliferative rate is pro-growth at primary cancer sites while the capability of migration and invasion is more beneficial to metastatic spreading. Indeed, in a variety of cancer types, such as melanoma[15], breast[16, 17], oral [18] and pancreatic cancers [5], the “growth” and “spread” features seem to be separate and sometimes mutually exclusive [15, 19]. Furthermore, a myriad of binary regulators governing this switch have been identified. Among them the most widely studied is transforming growth factor-ß (TGFß). TGFß suppresses cancer initiation by preventing cellular immortalization, inhibiting epithelial cell proliferation and inducing apoptosis (Table 1, Key Table) [20]. Through genetic and epigenetic alterations acquired in the process of tumor progression, cancer cells override or become resistant to the cytostatic function of TGFß, and instead exploit the benefits TGFβ brings to metastasis [21]. One of the major effects of TGFß in metastasis is the induction of epithelial-mesenchymal transition (EMT) [22], a process characterized by loss of epithelial adhesion and gain of mobility. Further, increased TGFß signaling activation has been found at the invasive edge of mammary carcinomas [23]. By bestowing cancer cells with higher migratory and invasive potential, TGFß is widely observed to promote cancer dissemination and metastasis [20]. In line with this, other EMT-regulating factors, such as Y-box binding protein 1 (YB-1) [16, 24] and grainyhead like transcription factor 2 (GRHL2) [17, 18], could also play dichotomous roles in tumor growth and metastasis. In addition, some other molecules, such as homeobox C9 (HOXC9) [25], p38 mitogen-activated protein kinase (MAPK) [26, 27] and ephrin type-B receptor 2 (EphB2) [28, 29] (summarized in Table 1), though not implicated in EMT manipulation, could also regulate cancer cell proliferation and migration oppositely. These opposite functions of the above modulators usually result from their ability to regulate multiple downstream pathways that are relevant to proliferation and metastasis.

Table 1, Key Table.

Examples of bi-potential molecular regulators of metastasis.

Regulator	Anti-tumor/metastasis effects	Pro-tumor/metastasis effects	Cancer types	References
TGFβ	causes cell cycle arrest	induces migration, invasion and metastasis via multiple mechanism, including EMT	multiple	reviewed in 20–22
YB-1	reduces G1/S transition and mitotic cyclins	induces migration, invasion and metastasis	breast, cervical, prostate	16, 24
GRHL2	suppresses migration	promotes cancer stem cell-like property and tumorigenesis	oral, breast	17, 18
HOXC9	reduces proliferation	suppresses invasion	breast	25
p38	causes proliferation arrest of cancer cells and tumor dormancy at metastatic loci	promotes cancer cell dissemination and survival	head and neck squamous cell carcinoma, gastric	26, 27
EphB2	regulates cell organization and positioning transcriptome via PI3K and inhibits invasion in carcinoma	regulates mitogenic transcriptome and promotes cell proliferation in adenomas	colon	28, 29
p53	the wild type is tumor suppressor	R172H and R175H mutants promote invasiveness	pancreatic	30, 31
		R175H and R273H promote constitutive activation of EGFR/integrin signaling pathway and elevated metastasis	lung and intestinal	32
		R175H, R273H and C135Y mutants promote EMT by suppressing miRNA-130 and inducing ZEB1	endometrial	33
RUNX3	causes cell cycle arrest by inducing p21	induces expression of pro-metastatic extracellular matrix	pancreatic	5
WNT signaling	suppresses osteoclastogenesis and osteolytic bone metastasis	promotes primary tumor initiation and growth	various	36–40
	elicits NK-mediated immune surveillance	promotes metastatic growth by regulating myeloid cell recruitment and TGFβ secretion	breast cancer lung metastasis	4, 44
	inhibits bone degradation and growth factor release	supports osteoblastogenesis and osteoblastic growth of tumor cells	prostate cancer bone metastasis	41, 42

In addition, bi-potential effects could also result from mutations of the molecular regulators, which lead to the functional drift of the regulators themselves. One of the most-studied tumor suppressors, p53 (encoded by the TP53 gene) (Table 1), is such an example of face changing from tumor safeguarding to metastasis promotion. In pancreatic ductal adenocarcinoma (PDAC) model, loss of one wild-type p53 copy, coupled with Kras oncogenic mutation, is essential for tumor initiation. However, rather than homozygous deletion of p53, a mutated allele (p53^R172H) along with the loss of the other allele promotes PDAC cell migration and liver metastasis [30]. If the mutation is R175H, the TGFß-driven invasive potential of lung and breast cancer cells is increased [31]. Another study showed that p53 mutants p53^R175H and p53^R273H constitutively activated the epithelial growth factor receptor (EGFR)/integrin signaling pathway and elevate metastasis in lung and intestinal cancer cells [32]. Further, it has also been reported that mutants p53^R175H, p53^R273H and p53^C135Y suppress the expression of microRNA(miRNA)-130, a known inhibitor of the EMT-driving transcription factor zinc finger E-box binding homeobox 1 (ZEB1), in endometrial cancer cells [33]. As somatic mutations of tumor suppressors are commonly observed in cancer cells, the functional discrepancy between the wild-type and mutant allele indicates the necessity of careful examination of various mutations, most of whose functions remain unknown. We believe that with the rapid development of precise genome editing approaches [34, 35], genetically engineered animal models may help discover more bi-potential regulators of metastasis with wild-type/mutant specificities.

Bi-potential signaling molecules that regulate the external environment of cancer cells

In addition to manipulating cancer cell intrinsic properties, some bi-potential modulators can also exert opposite effects by regulating the external environment. An example of such a bi-potential factor is runt related transcription factor 3 (RUNX3), which drives dissemination and metastasis of PDAC by inducing the expression of pro-metastatic extracellular matrix (ECM) 4 proteins and promoting ECM modulation, while suppresses primary PDAC growth by inhibiting cell cycle (Table 1) [5].

The WNT signaling pathways [ß-catenin-dependent (canonical) and independent (noncanonical)] are also bi-potential cues which regulate metastasis in a stage and organ-specific way (Figure 1). WNT signaling is known to induce the differentiation of mesenchymal stem cells to osteoblasts (osteoblastogenesis) and repress the differentiation of osteoclast precursors to mature osteoclasts (osteoclastogenesis) [36, 37]. Abnormal activation of WNT signaling is oncogenic [37–39], but endogenous WNT inhibitors such as dickkopf-1 (DKK1) is linked to enhanced bone degradation by osteoclasts and osteolytic metastasis of cancer cells [40]. It has been reported that targeting of WNT by a WNT inhibitor in breast cancer leads to bone metastasis [4]. Further, WNT signaling was also found to exert opposite effects in bone metastasis (Figure 1A) from prostate cancer, which is characterized by excessive maturation of osteoblasts. It is suggested that prostate cancer cells suppress WNT signaling by producing DKK1 during the early phase of bone colonization (Figure 1A, left panel). DKK1 favors bone matrix degradation by tipping the balance of bone remodeling toward osteolysis, leading to the release of growth factors from bone matrix to fertilize the initial growth of lodged prostate cancer cells. At the later metastasis stage, however, DKK1 expression is lost allowing for WNT activation and maturation of osteoblasts, which in turn facilitate the progression of prostate cancer in bone (Figure 1A, right panel) [41, 42]. Therefore, bone colonization in metastasis from prostate cancer is often a result of the mixed early osteolytic phase and subsequent osteoblastic phase (Figure 1A). How osteoblasts promote cancer cell proliferation is currently not fully understood, but the effect seems to be specific to prostate cancer [43].

Figure 1. WNT as a bi-potential modulator of metastasis.

WNT plays stage-dependent contrary roles in both bone (A) and lung (B) metastasis. In osteolytic bone metastasis (of breast cancer, for example) and the early stage of osteoblastic bone metastasis (of prostate cancer, for example), tumor-derived transient high level of WNT inhibitor DKK1 suppresses osteoblast differentiation to osteocytes but enhances osteoclast differentiation. Bone matrix degradation ensues that releases growth factors to nurse disseminated tumor cells (DTCs) (A, left panel). In the late stage of osteoblastic bone metastasis, DKK1 level is tuned down to unleash osteoblast differentiation, leading to full-blown osteoblastic colonization (A, right panel). Similarly, in the early stage of lung metastasis (B, left panel), the transient high level of DKK1 maintains the quiescence of DTCs to allow evasion from immune surveillance by natural killer (NK) cells. Later, DKK1 level is tuned down and the activation of WNT signaling pathway supports outgrowth of lung metastasis (B, right panel).

Moreover, it has been found that inhibition and activation of WNT signaling are important for the metastasis of breast cancer to the lungs [4, 44]. At the stage of metastatic homing and seeding, WNT suppression by DKK1 is crucial for cancer cells to maintain dormancy and evasion from natural killer (NK) cell-mediated immune clearance (Figure 1B, left panel) [44]. However, inhibiting DKK1 at a later metastatic stage (Figure 1B, right panel) or in a NK cell-deficient microenvironment [4, 44] allows unleashed WNT signaling to promote full-blown lung colonization. Both the findings from the bone and lung metastasis show that bi-potential modulators such as WNT, often act in a stage-specific manner during the metastatic process and targeting them in therapeutics may also require metastasis stage-specific precision.

As metastasis is a systemic process, disseminated cancer cells in different organs respond differently to extrinsic regulation and stimulation. For example, in addition to regulating metastasis in a stage-specific manner, DKK1 is also reported to play opposite roles in the colonization of breast cancer cells to lungs and bones [4]. Through suppression of the canonical WNT signaling in osteoblasts, DKK1 supports osteoclast maturation and bone degradation, leading to enhanced survival and metastatic colonization of breast cancer cells that are driven by TGFß, insulin like growth factor (IGF) and Ca²⁺ released from the degraded bone matrix. However, in lungs, DKK1 suppresses metastatic outgrowth by disrupting the non-canonical WNT signaling pathways of cancer cells, leading to reduced myeloid cell recruitment [4]. Importantly, the serum level of DKK1 is significantly higher in bone-metastatic patients than that in non-metastatic ones, but it is the lowest in lung-metastatic patients, corroborating an organ-specific role of DKK1 in metastasis.

Microenvironment cell components that show opposite effects on metastatic process

The tumor microenvironment (TME) that includes both the non-cancerous cellular and noncellular components interspersed within or adjacent to the tumor mass has been shown to actively contribute to the tumor growth and progression [45]. TME is particularly relevant to metastasis, as the process of metastasis involves extensive interaction with the TME. Disseminated tumor cells (DTCs) survive and proliferate into macrometastases by adapting to the TME and further training it into an accomplice. Thus, the TME can potentially be targeted for metastasis treatments. Indeed, the development of a plethora of TME-targeting approaches, such as anti-angiogenesis and osteoclast-targeting therapies, has shown promising clinical outcomes [46–48]. However, resistance to these therapies and low therapeutic efficacy still occur. This can partially be explained by the heterogeneity of the TME that has been elucidated by recent studies, especially those using single-cell mRNA sequencing technologies [49, 50]. The diverse phenotypes of TME components could be attributed to the different soliciting effects of tumor cells as well as the intrinsic plasticity of TME components. Such heterogeneity in the TME can be both dangerous and therapeutically promising. This would mean that, on one hand, a simple elimination of a certain TME stromal population may be only partially beneficial for tumor treatment; on the other hand, the reversibility of stromal plasticity could open new therapeutic windows to target metastasis. Below we briefly discuss some cellular components of the TME that have shown dual roles in cancer progression and metastasis.

Endothelial cells (ECs)

ECs constitute a natural barrier of metastatic homing by preventing the entry of circulating tumor cells into the parenchyma of the metastasis target organs. A large body of evidence shows that metastatic cancer cells adopt various methods to break down the endothelial walls in target organs [51, 52], especially in the brain where the specialized brain blood barrier is even harder to breach [53]. ECs are also able to suppress metastasis outgrowth by keeping infiltrated cancer cells in prolonged quiescence via secreted factors such as thrombospondin-1(TSP-1) and perlecan [54]. However, when vascular homeostasis is disrupted, the generation of new blood vessels can awake the dormant cancer cells with the secretion of pro-metastasis factors, such as periostin, TGFβ1, fibronectin, versican and tenascin-C (Figure 2) [54]. Furthermore, ECs are also known to secrete epoxyeicosanoids, a group of lipid metabolites, to help cancer cells escape from dormancy in various types of cancer including melanoma, fibrosarcoma and Lewis lung carcinoma [55]. Additionally, tumor-associated ECs respond to chemotherapy and enhance the stem-like properties of cancer cells [56]. Before exposure to chemotherapy, ECs produce IGFBP7 to antagonize the pro-tumor IGF1 signaling pathway. However, chemotherapy suppresses IGFBP7 expression in ECs, leading to upregulated IGF1 signaling and enhanced chemoresistance of cancer cells (Figure 2).

Figure 2. Cellular components of TME that play opposite roles in metastasis.

During metastatic growth, different tumor microenvironment (TME) cells possess seemingly contradictory regulatory effects on tumor cell dissemination and metastatic colonization. These cells include, but are not limited to, endothelial cells (ECs), fibroblasts, neutrophils and macrophages. The pro- and anti- metastasis functions of each are listed below the respective cell type.

While anti-neoangiogenesis drugs are widely used in cancer therapy [57], it is known that genetic disruption of vascular endothelial growth factor (VEGF), a signaling protein that promotes the formation of new blood vessels, or pharmacological inhibition of its receptor promotes local invasiveness and distant metastasis in animal models of breast cancer, pancreatic neuroendocrine carcinoma and glioblastoma [58, 59]. Although the underlying mechanism is yet to be studied, these observations corroborate the complicated effects of ECs on metastasis and indicate that ECs targeting may require more scrutiny to treat metastatic diseases.

Cancer-associated fibroblasts (CAFs)

Normal fibroblasts suppress tumor growth and progression by contact-dependent “neighbor suppression” mechanisms or by secreting soluble factors such as TGFß andtumor necrosis factor-α(TNFα) (Figure 2) [60]. In contrast. cancer-associated fibroblasts (CAFs) have been shown to support tumor growth, metastasis and enhance therapeutic resistance in various cancer types by providing an optimal niche to the tumor cells to survive, proliferate, migrate and by suppressing immunosurveillance [61]. Recent studies keep adding evidence to the pro-tumor effects of CAFs, highlighting the roles of CAF-derived extracellular vesicles in mediating the CAF-tumor interaction for tumor promotion and immune modulation [62, 63]. The difference between the effects of normal fibroblasts and the CAFs on tumors likely reflects a phenotypic transition of fibroblasts from normal tissues to tumor microenvironment. However, it is possible that CAFs might also be able to antagonize tumor progression and metastasis. Clinical correlation has been observed between higher expression of the CAF marker alpha-smooth muscle actin (α-SMA) in tumors and improved patient survival [64]. Further, it was reported that depletion of α-SMA⁺ fibroblasts in PDAC after tumor establishment augmented tumor invasion and worsened animal survival. Mechanistic studies showed that the depletion led to increased infiltration of immunosuppressive CD4⁺ Foxp3⁺ regulatory T cells, which could be reversed by immune checkpoint inhibition [65]. However, it is important to note here that the mouse model used in these studies had all α-SMA⁺ myofibroblasts depleted, regardless of their interaction with PDAC cells. Thus, the data may not differentiate between the functions of normal fibroblasts and CAFs. Nevertheless, these studies indicate the possibility that CAFs possess paradoxically regulatory roles in tumor progression.

Neutrophils

Neutrophils are bone marrow-derived myeloid cells that have been shown to suppress lung metastasis by secreting Tsp-1 [66]. Infiltrated neutrophils are also known to kill metastatic breast cancer cells by generating hydrogen peroxide (H₂O₂) [67]. The cytotoxicity effect of tumor-associated neutrophils can also be mediated by nitric oxide (NO) (Figure 2) [68]. Moreover, neutrophil depletion by anti-Ly6G (protein expressed predominantly on neutrophils) neutralizing antibody has been found to promote lung metastasis of renal cell carcinoma and breast cancer [67, 69], thus highlighting its protective functions against metastasis. However, numerous studies have also shown neutrophils as an accomplice of metastasis, especially for forming the pre-metastatic niche to solicit incoming tumor cells and forming the metastatic niche to promote the survival and proliferation of seeded tumor cells [70]. Further, a recent report has demonstrated that neutrophils support the metastatic initiation of breast cancer in lungs by expanding the stem cell-like subpopulation of cancer cells [71]. As with other TME cells that have shown opposing functions in metastasis, the observations seen with neutrophils may become a huge obstacle to the application of neutrophil-related anti-metastatic therapies. Further studies are still required to determine what reasons contribute to the phenotypic heterogeneity of tumor-associated neutrophils.

Macrophages

Macrophages are known to possess highly diverse phenotypic and functional heterogeneity in both primary tumors and metastases. Macrophages can be polarized to M1 (also known as classically activated macrophages) and M2 (also known as alternatively activated macrophages) subtypes by different stimuli such as interferon-gamma (IFN-γ), interleukin (IL) 4 and IL 13. Typically, M1 macrophages exert anti-tumor functions by secretion of pro-inflammatory cytokines such as interleukins IL6, IL12, IL23 and TNFα. In contrast, tumor-associated macrophages (TAMs), usually resembling the M2 subtype, are shown to promote tumor initiation and progression by various mechanisms such as promotion of angiogenesis, induction of EMT and suppression of anti-tumor immune responses [72]. In line with the tumor-promoting roles of TAMs, potential therapeutic approaches targeting tumor-infiltrated macrophages, such as anti-CCL2 and colony stimulating factor 1 receptor (CSF1R) antibodies, have shown promising anti-tumor effects in animal models [73–76]. However, resistance to these TAM-targeting methods quickly occurs and tumor recurrence ensues [73, 76]. One possible explanation of such unsatisfactory results could be the phenotypic heterogeneity of TAMs. Zhu et al identified that both bone marrow derived inflammatory monocytes and yolk sac-derived tissue-resident macrophages co-exist in PDAC and display different functions toward tumors, with the former enhancing the antigen presentation for immune responses while the latter pro-fibrotic and driving tumor progression [77]. With the application of the single-cell mRNA sequencing, the ontological and phenotypic heterogeneity of macrophages starts to become evident [78], which can be highly beneficial to understand the biology of TAMs and develop therapies for precise elimination of the pro-tumor TAMs or suppression of the tumor-supporting function of TAMs.

More importantly, the dynamic role of TAMs in tumor progression indicates that TAM elimination may not be necessary to treat cancer. Recent studies indicate that manipulating TAM’s reversible phenotypic plasticity, instead of aiming for its depletion, could be a promising strategy that takes advantage of the frequent presence of TAMs in tumor suppression. Nevertheless, discovering potent TAM-converting switches is challenging. Kaneda et al showed that PI3Kγ can serve as such a switch [6]. PI3Kγ suppresses NF-κB but activates CCAAT/ enhancer binding protein beta (C/EBPß) signaling, maintaining an immunosuppressive transcriptional program in macrophages. Selective PI3Kγ inhibition reprograms TAMs to an immunostimulatory phenotype and restores T cell-mediated cytotoxicity against tumor cells. The immunoregulatory function is corroborated in another independent study demonstrating that inactivation of PI3Kγ possesses promising potential to overcome resistance to the immune checkpoint blocking therapy [79]. Additionally, disruption of miRNA biogenesis in macrophages also pushes the M2-to-M1 switch by hyperactivation of IFNγ/STAT1 pathway (Figure 2) [80]. These seminal studies indicate that manipulation of TAM plasticity provides an unprecedented opportunity to thwart tumor-associated immune suppression and unleash the T cell cytotoxicity against tumor cells.

However, before we could apply macrophage “taming” to clinical practice, there are major questions that need to be researched to better understand the biology of TAMs and discover effective approaches to target or utilize TAMs for cancer therapy. These include but are not limited to: (1) what different roles do macrophages derived from various sources (bone marrow, spleen, tissue-resident) play in tumor progression; (2) are polarized TAMs always readily convertible or is there a fixed TAM stage where their cancer-specific functions are specified; and (3) what is the optimal turnover time of TAMs for TAM-targeting therapies and how sustainable will these approaches be? With better answers to these questions, it would be more promising to manipulate TAMs for better therapeutic outcomes.

Concluding remarks and future perspectives

Here, we reviewed the different bi-potential molecular and cellular regulators that show opposing effects in tumor progression, especially metastatic diseases. Though the transition from proliferative growth to metastatic spreading of cancer cells is widely observed in various cancer types, the driving forces and evolutionary benefits of such transition for cancer cells are still unclear and very intriguing. One possible explanation could be the deprivation of resources (space, nutrients, oxygen etc.) at the primary site that restricts continuous growth of primary tumors and renders secondary metastatic growth advantageous. However, in plenty of cases, metastasis begins at the early stage of tumor progression and increasing primary tumor sizes do not strictly link with metastasis [81, 82]. This confounding fact could implicate the importance of the binary switches between primary and metastatic growth, as they could be activated by either environmental cues or genetic regulation. Thus, manipulation of these switch molecules, possibly in combination with other therapeutic approaches, could be of great importance to prevent or control tumor progression. For example, TGFß inhibition has been brought into clinical trials for a wide range of cancers and has shown some promising prospects in cancer therapy [21, 83, 84]. However, it is unknown whether targeting of binary switches will lead to unexpected responses depending on the stage of tumor development. Moreover, with current evidence, it seems that the transition between primary and invasive growth happens spontaneously, but whether additional selective pressure may accelerate the process is unknown. Considering that chemotherapy preferably eliminates highly proliferative cancer cells, the therapy could possibly serve as a potential driving force of invasive growth of cancer cell and metastasis possibly by turning on or off the binary modulators, which could have important implication for cancer treatment.

The seemingly contradictory effects of the bi-potential factors highlight the notion that cancer is a dynamic process. Cancer initiation and metastatic spreading is continuous and systemic, however, cancer cells are remarkably different from each other, based on their development stages (primary tumor vs. metastasis) and stromal surrounding. Previous works have made great strides in analyzing cancer cells of various states one by one. However, it is time to emphasize and tackle the co-existence of heterogeneous cancer cells in the same patient. This requires researchers to look at the disease of cancer in a global and dynamic scope (see Outstanding Questions).

Outstanding Questions.

1. Are there other bi-potential molecular switches affecting the different processes in tumor initiation and progression?

2. How can we target the bi-potential switches of tumor progression and metastasis in cancer therapeutics to maximize the treatment efficacy and minimize side effects?

3. How is the phenotypic plasticity of tumor-associated stromal components related to the different stages of tumor education or to the intrinsic heterogeneity of these stromal components?

4. How to develop experimental approaches to trace and manipulate the plasticity of stromal cells?

5. How to convert the pro-tumor stromal cells into tumor-resistant components for new potential therapies?

6. How to develop standards for systematic and dynamic evaluation of bi-potential switches towards potentially enhancing therapeutic approaches?

The organ and stage specificity in metastasis regulation argue for the necessity of systemic investigation of metastasis. One daunting challenge of such studies, however, is the lack of suitable animal models of metastasis. Studying metastasis by experimental metastasis models, majorly by inoculation of cancer cells via various entrance of circulation, have provided important discoveries regarding the process of tumor colonization in specific organs. However, spontaneous metastatic models, especially autochthonous models (where in situ tumors are induced by carcinogens or genetic engineering) in immune-competent mice, will be of great value for the better understanding of systemic regulation of metastasis. The development of such models, especially for bone and liver metastasis, are highly desirable.

Finally, it is important to note that we also need to be cautious when interpreting all results for bi-potential switches as multiple experimental parameters can potentially lead to contradictory results. These could be from (1) animal models adopted (transplantation of foreign tumors vs. use of autochthonous models); (2) different biological characteristics of various cancer types; (3) approaches to target the molecule or cell population of interest (small molecular inhibitor vs. inhibitory proteins; neutralization antibodies vs. genetic depletion); (4) timing and duration of targeting (prior to vs. post tumor induction or transplantation); and (5) possible genetic drift of tumor cells cultured in different labs. An example is the bone morphogenetic protein 4 (BMP4), a member of TGF super family which has been shown to possess apparently contradictory functions in mammary carcinoma metastasis. BMP4 suppresses breast cancer multi-organ metastasis by decreasing recruitment of myeloid derived suppressive cells (MDSCs) [85], which are known to suppress the anti-tumor immune responses by T cells [86]. Moreover, BMP4 antagonist Coco reactivates the dormant metastatic cancer in lungs by enhancing the cancer stem cell-like properties [87]. However, antagonizing BMP4 by a small molecule DMH1 suppresses primary tumor growth, inhibits tumor-promoting microenvironment and reduces lung metastasis in the MMTV-PyMT spontaneous mammary carcinoma model [88]. Such examples exemplify that it is always advisable to carefully distinguish discrepancy from real dichotomy. Ultimately, the examples of bi-potential regulators reviewed here show the need for cautious stratification of patients by their major clinical symptoms (proliferative tumors or wide-spread metastasis, dominating components in the microenvironment, etc.) and combinatorial therapeutic approaches to avoid causing one problem while solving another.

Highlights.

• Both cancer cell-derived factors and tumor-associated microenvironment components possess phenotypic plasticity and functional dichotomy.

• The plasticity can be controlled by bi-potential molecular switches.

• The bi-potential molecular and cellular switches modulate metastasis in a site- and stage-specific manner and should be studied thoroughly to improve the effectiveness and safety of treatments.

• Targeting bi-potential molecular switches, such as WNT, without thorough knowledge, can potentially lead to incomplete or even deleterious therapeutic outcomes.

• Phenotypic manipulation, instead of simple depletion, of tumor-associated microenvironment components, can probably bring additional benefits for the therapy.