Vascular determinants of cancer stem cell dormancy—do age and coagulation system play a role?


  • Invited review.


The inability of tumour-initiating cancer stem cells (CSCs) to bring about a net increase in tumour mass could be described as a source of tumour dormancy. While CSCs may be intrinsically capable of driving malignant growth, to do so they require compatible surroundings of supportive cells, growth factors, adhesion molecules and energy sources (e.g. glucose and oxygen), all of which constitute what may be referred to as a ‘permissive' CSC niche. However, in some circumstances, the configuration of these factors could be incompatible with CSC growth (a ‘non-permissive' niche) and lead to their death or dormancy. CSCs and their niches may also differ between adult and paediatric cancers. In this regard the various facets of the tumour-vascular interface could serve as elements of the CSC niche. Indeed, transformed cells with an increased tumour-initiating capability may preferentially reside in specific zones adjacent to tumour blood vessels, or alternatively originate from poorly perfused and hypoxic areas, to which they have adapted. CSCs themselves may produce increased amounts of angiogenic factors, or rely for this on their progeny or activated host stromal cells. It is likely that ‘vascular' properties of tumour-initiating cells and those of their niches may diversify and evolve with tumour progression. The emerging themes in this area include the role of vascular (and bone marrow) aging, vascular and metabolic comorbidities (e.g. atherosclerosis) and the effects of the coagulation system (both at the local and systemic levels), all of which could impact the functionality of CSCs and their niches and affect tumour growth, dormancy and formation of occult as well as overt metastases. In this article we will discuss some of the vascular properties of CSCs relevant to tumour dormancy and progression, including: (i) the role of CSCs in regulating tumour vascular supply, i.e the onset and maintenance of tumour angiogenesis; (ii) the consequences of changing vascular demand (vascular dependence) of CSC and their progeny; (iii) the interplay between CSCs and the vascular system during the process of metastasis, and especially (iv) the impact of the coagulation system on the properties of CSC and their niches. We will use the oncogene-driven expression of tissue factor (TF) in cancer cells as a paradigm in this regard, as TF represents a common denominator of several vascular processes that commonly occur in cancer, most notably coagulation and angiogenesis. In so doing we will explore the therapeutic implications of targeting TF and the coagulation system to modulate the dynamics of tumour growth and tumour dormancy.


It is increasingly clear that tumour cell heterogeneity (1) is profoundly hierarchical in nature, in that various coexisting cellular subtests may differ (often dramatically) in their relative tumour-initiating capabilities (2). Indeed, by analogy to self renewing tissues it has been proposed that the low mitotic index, indefinite self-renewal capacity, contribution to repopulation events and resistance to various therapeutics of certain subsets of cancer cells resembles that of stem cells in bone marrow, gut or epidermis (3). This led to the notion that, like normal stem cells and their progeny (transit-amplifying, committed and differentiated cells), also cancer cell subpopulations can be separated in a nearly binary fashion into cells capable of growth (tumour) initiation and those devoid of such capacity. The former are often referred to as cancer stem cells (CSCs) (2, 4) or, perhaps more appropriately, as tumour-initiating cells (TICs) and are believed to constitute a minority within the population (3), while their mitogenically active progeny composes the bulk of the tumour mass (3).

This highly consequential concept is derived, at least in part, from the observation that in transplantation experiments very large numbers of cancer cells have to be injected to generate tumours in susceptible animals. This suggests that such cells possess a minimal (or nil) ‘average', stochastic tumour-initiating potential, or that tumours arise from very specialized cells (CSCs/TICs), which, however, occur at a very low frequency (3). When the latter cells are isolated, their much smaller numbers (even individual cells) readily form tumours under similar injection conditions (3). Since the progeny of these cells is generated by asymmetric cell division it may exhibit a more differentiated phenotype and higher constitutive, but ultimately limited mitotic activity, which is thought to be eventually exhausted in the absence of CSCs (3). Tumour growth initiation, recurrence, post-therapy relapse, or metastasis requires that the silent minority of CSCs be activated within the population (3). Due to these properties CSCs are viewed as a particularly valuable (and hitherto ‘overlooked') future therapeutic target, as their elimination is predicted to have long-term curative consequences in contrast to the often futile traditional anticancer therapies, which can achieve a cytotoxic debulking of the majority of dividing cells, but leave the CSC intact, a seed for eventual therapeutic failure (3). Moreover, targeting CSCs seems to be within the reach of current medical technologies, because of the growing library of molecular criteria (markers), on the basis of which CSCs could potentially be identified in various tumours and selectively eliminated. The most common CSC markers include the expression of several surface antigens, such as: CD133, CD34, CD44, CD24, CD166, ABC transporters (ABCB1, ABCG2), absence/downregulation of certain differentiation, or lineage markers (e.g. CD38 in leukemia) and several functional properties (3, 5, 6).

Although the CSC concept is extremely captivating, with accumulating data their accurate description, or even definition, is becoming increasingly complex (6, 7). For example, the prevalence of antigenic markers associated with CSCs in a given tumour cell population frequently differs from the numbers of cells equipped with a detectable colony, or tumour-initiating properties, as defined by the respective functional assays (6). Conversely, the oncogenic transformation may, at least in some cases, dramatically increase tumour-initiating capabilities of a clonal cancer cell population, but this often occurs without a measurable increase in the expression of CSC markers thought to characterize a given tumour (7, 8). Thus, in the indolent glioma cell line U373 the expression of the oncogenic epidermal growth factor receptor variant III (EGFRvIII) results in rapid transition to a full blown tumorigenic phenotype and formation of aggressive tumours in immunodeficient mice, while the transformed cells remain largely negative for markers of glioma stem cells (e.g. CD133; Milsom & Rak, unpublished observation). On the other hand, some of the model cancer cell lines remain very poorly tumorigenic in experimental animals (e.g. Caco-2 colorectal cancer cells) in spite of a robust expression of CD133 (expressed by the putative colorectal CSCs (9). Obviously, this observation does not negate the value of CD133 and other markers in detecting CSCs, especially in primary tumour isolates (10), but it also offers a cautionary note as to equating the marker-defined cellular phenotype with the functional features and “stemness” in a highly heterogeneous and chaotic world of cancer cell populations (6, 8). Moreover, the ontogenetic and phenotypic relationship between CSCs and their normal stem cell counterparts of the given tissue of origin is not entirely clear; neither are the mechanisms of the apparent modulation of tumour cell “stemness” (and expression of stem cell markers) by various influences, such as oncogenic transformation, microenvironmental factors (e.g. hypoxia), cell-cell interactions and impact of the host stroma (6, 7). Therefore, it seems reasonable to consider CSCs in mainly operational terms, namely as cancer cells with a particularly high tumorigenic ability achieved by a combination of endogenous properties and interactions with the wider cellular/tissue context. In some instances (but not always) such CSCs may also express known stem cell markers, e.g. CD133 or CD44.


One aspect of cancer progression where the concept of CSCs may provide valuable insights is in mechanisms that control the onset, or cessation of tumour dormancy (5, 11). The latter term refers to the phenomenon whereby small deposits or single cancer cells remain present in various sites, but their expansion and the related clinical progression of the disease is undetectable, often for extended periods of time (years or decades) (12, 11). The distinction between different types of dormant tumour behaviour is a subject of some controversy, especially whether certain premalignant (premetastatic) states involving ostensibly malignant, but not expanding cells could be classified as dormancy, or whether this term should be reserved for the reservoirs of cancer cells remaining after effective therapy (12). The latter concept is applicable to a condition known as minimal residual disease, where cancer cells or micrometastases can remain silent in effectively treated patients, either for a long time (even permanently), or at least transiently (11). It is unclear whether these different forms of dormant behaviour may involve CSCs, and whether these cells are similar in the case of post-treatment dormancy of the primary tumour, in micrometastasis and prior to the onset of overt malignant growth.

Several intriguing manifestations of the dormant behaviour of tumour cells have been described in the recent literature. In one study, single or multiple cancer cells were detected at autopsy in thyroid glands, prostates or breast tissues of individuals who died of cancer-unrelated causes and had no signs of the corresponding malignancies. Interestingly, the frequency with which these silent cancer cells were detected exceeded the prevalence of the respective overt cancers in the population, often by several orders of magnitude (13, 14). This would suggest that ostensibly transformed cancer cells could lie dormant in a large proportion of healthy individuals and for extended periods of time. Only in some cases would such cells initiate clinical disease (14). Moreover, in a large percentage of patients deemed disease free after 7–22 years post breast cancer therapy, circulating tumour cells (CTCs) can be detected in peripheral blood, a finding that suggests the persistence of their occult (dormant) reservoir (12).

While in some instances dormant cancer cells may remain in a state of mitotic quiescence, elsewhere their small nests may be mitotically active, but this process produces no net increase in tumour mass, or apparent disease due to the simultaneous (and equivalent) cell loss via processes of cell death (apoptosis, autophagy), senescence and/or terminal differentiation (5, 11, 14). This dynamic equilibrium is often described as “tumour mass dormancy” (11).

While the exact mechanisms of tumour dormancy remain unclear (15), they may include one or more of the following components (11): (i) microenvironmental influences, such as certain configurations of extracellular matrix proteins, growth factors, hormones and metabolic conditions that override the intrinsic proliferation program of cancer cells causing their quiescence, differentiation, or increased rate of apoptotic death; (ii) immunosurveillance, whereby proliferating cancer cells are selectively eliminated by effector T- or NK cells; (iii) angiogenic dormancy (16, 17), where the inability of cancer cells to mount an effective vascular growth and remodelling may cause prolonged latency, or long-term dormancy. This can be due to quiescence (18), or increased apoptotic death rate related to either a protracted ischemic stress (16, 17, 19) or resulting from withdrawal of paracrine growth-stimulating factors elaborated by endothelial cells (20). The angiogenic dormancy of cancer cells could be either a consequence of the inherent inability of these cells to elaborate angiogenic factors (or to downregulate angiogenic inhibitors) (21), or be related to the properties of the host vasculature such as genetic background (22, 23), aging (24), or vascular comorbidities that impair angiogenic responses (25). Importantly, angiogenic dormancy may also emerge as the result of a successful antiangiogenic therapy (19). Again, these latter examples point to the tumour-vascular interface as a source of influences that may cause induction or cessation of tumour dormancy.

If tumour initiation is dependent on the activation of CSCs, it is implicit that, conversely, tumour dormancy is a function of their ‘inactivation'. This could be interpreted as numerical depletion, or absence, of CSCs in dormant tumours. In theory, this would preclude the exit from the dormant state. Furthermore, it could be speculated that the state of minimal residual disease accompanied by the presence of viable but quiescent (dormant) cancer cells in the bone marrow of breast, cancer patients, or in other similar circumstances, simply reflects the ‘unsuccessful’ dissemination of cancer cells devoid of the tumour-initiating (CSC) properties. However, both experimental (5) and clinical studies (11), suggest that the presence of dormant cancer cells in the bone marrow creates an increased risk of recurrence, or overt growth at a later time, implicit evidence that at least some of these dormant cells do possess the ability to acquire, the properties of CSCs (5). In addition, CSCs could persist, but undergo a functional change that would diminish their tumour-initiating capacity in a given context. In this regard it is not clear whether, for instance, angiogenic dormancy leads only to changes in numbers or, indeed, in properties of tumour-initiating cells. Since the fraction of cells harbouring stem cell markers tends to increase under hypoxic conditions (6), it could be speculated that a competing mechanism of angiogenic dormancy prevents this increase from translating into a net gain in tumour cell number (mass). This could be realized through the relatively well-described process of hypoxia-related apoptosis, which could affect the progeny of CSCs and/or CSCs themselves. It is possible that hypoxia could also reduce the rate of generation of the CSC progeny, e.g. by influencing the dynamics of the asymmetric cell division. These various scenarios remain largely unproven, but raise the question as to whether the tumour-initiating potential of CSCs is constitutive, or conditional (6), cell autonomous or context-dependent in nature (7), and how it is ultimately regulated in vivo.


It is increasingly clear that the ability of CSC to initiate tumour growth (i.e. their effective “stemness”) is profoundly influenced by the tissue environment in which such cells reside (6, 7, 26). By analogy to properties of normal stem cells this is often referred to as a cancer stem cell or (pre)metastatic (27) niche effect. The composition of the niche to which CSCs could home, and from which they could control growth of primary and metastatic tumours, is presently rather ill defined, but includes a number of growth/survival regulating influences, such as certain types of extracellular matrix, abundance of growth factors, the presence of accessory cells (e.g. endothelial or hematopoietic cells) (3, 26, 27), tissue stroma, or even (possibly) other cancer cells themselves, an obvious source of paracrine interactions (7). The niche may also include vascular elements and, as we have recently proposed, various activities of the coagulation system, which are known to control cancer growth and metastasis (28).

It is implicitly clear that the nature of the CSC niche and changes occurring therein may profoundly alter the behaviour and fate of CSCs. Perhaps the most readily detectable is the ‘permissive' effect of the CSC niche resulting in overt tumour growth and/or metastasis. However, some of the related experiments suggest that these effects are highly context dependent. Thus, injection of the same suspension of cancer cells into different subcutaneous tissue sites leads to a differential rate of tumour growth (29). Similarly, injection of cell suspension obtained from a mouse mammary tumour via subcutaneous, intravenous, intrasplenic and intraperitoneal routes reveals organ-specific thresholds of tumour take (i.e. the heterogeneous nature of CSC niches) (30). Irradiation or exposure to anticancer agents can change the properties of various organ-related CSC niches and often facilitate (or impede) malignant growth, a phenomenon often referred to as a ‘tumour bed effect' (30, 31). One of the most meaningful manifestations of CSC niche effects is the differential growth of experimental tumours within their tissues of origin (orthotopic site) versus in ectopic sites of inoculation (e.g. subcutaneously), or tumour metastasis from these respective sites (32). However, this is also relative as metastasis as such, and organ-specific metastasis in particular, implicitly represents the ability of cancer cells to adapt to (or create) a permissive niche even in the ectopic tissue location (33).

In many instances CSC (tumour-initiating) niche could be defined by the extracellular matrix, either directly (through adhesive mechanisms) (26), or indirectly, e.g. as reservoir of growth, survival or angiogenic factors (5). With regard to the latter, the case in point is the technique of increasing the take of certain poorly transplantable cancer cells by their co-injection with basement membrane extracts known as Matrigel (34). Not only under these conditions can the tumour-initiating potential (stemness) of such cells be considerably increased, but also their highly tumorigenic variants can be re-isolated from Matrigel-supported tumours. In this latter instance, cancer cells appear to sustain a permanent change in several of their biological properties, including the acquisition of a spontaneous (i.e. Matrigel-independent) tumorigenicity, or “stemness” (34–36). Moreover, ostensibly non-tumorigenic cancer cell preparations could be rendered capable of malignant growth (i.e. become equipped with a sufficient CSC potential) by a simple admixture of excess of similar cells pretreated with lethal doses of ionizing radiation (Revesz phenomenon (37). Interestingly, in this case tumour cells themselves are able to exert a niche effect, suggesting that during natural tumorigenesis, the microenvironment of cancer cells themselves could potentially serve as a niche for CSCs (7). Such cell-cell interactions may occur via paracrine, adhesive and exocytic pathways, and their nature may change over time with modification of the tumour microenvironment, cellular composition and oncogenic events. Therefore, CSC niche effects of tumour cells may potentially evolve with progression of the disease. Tumour take can also be enhanced by the admixture of stromal fibroblasts, or other cells, which themselves are devoid of tumour-forming capacity (38), all of which raises the question as to the nature of these niche effects as they unmask, or obscure the CSC properties in cancer cell populations.


Although the concept of cancer stem cells (at least in a qualified sense) may serve as an organizing principle in studies of a variety of human tumours, its exact content is likely dependent on the specifics of each malignant disease. This may have a special meaning in the case of paediatric cancers, for several reasons. First, both the incidence and the spectrum of paediatric cancers is vastly different from those in adults, with paediatric cancers being relatively rare and dominated by leukemias, and to a lesser extent by brain tumours and with a considerable contribution of malignancies that are relatively restricted to the childhood, e.g. neuroblastoma, rhabdomyosarcoma, Wilms' tumour, hemangioma and PNET (39). Second, the natural history of paediatric tumours is much shorter and therefore the formation of the CSC, TA and end cell compartments must occur with different dynamics, with rapid progress and/or very early sometimes prenatal onset. Third, the latter consideration would suggest that certain paediatric malignancies (e.g. neuroblastoma) entail a significant component of a developmental defect with superimposed secondary genetic alterations (40), while the main adult cancers seem to contain more somatic adult-type CSCs (3). Fourth, aging and age-related diseases affect the normal stem cell compartment (e.g. in the bone marrow) (41) and various stem cell niches, including those related to the vasculature (25). As these are the components from which both CSCs and their niches eventually emerge, it is implicit that childhood malignancies would differ in this regard relative to adult cancers. Fifth, post-therapeutic dormancy in paediatric cancers is often shorter than that in adult tumours where this process may continue for five or more years, sometimes decades. Naturally, it is possible that in paediatric malignancies where a cure has been achieved there could be a dormant reservoir of cancer cells, but this question requires further study. Thus, the phenomenon of tumour dormancy and the role of CSCs and their niches in the process are likely unique in paediatric tumours and different from adult malignancies.


In many instances the formation of a ‘permissive' (i.e. growth-facilitating) CSC niche can be linked to the status of the local microvasculature, most notably the onset of tumour angiogenesis. One of the best known examples that may suggest such an influence is the technique of boosting the tumorigenic potential of certain cancer cell lines (and tumour-initiating cells in them) by their co-injection into mice with Matrigel, the proangiogenic extract of basement membrane (34, 42). As mentioned earlier, this often results in increased tumour take, a likely effect of stimulation of the vascular growth by both matrix molecules and growth factors present in the Matrigel preparation (36). Moreover, a direct, enforced expression of a single angiogenic growth factor (e.g. VEGF) in ostensibly non-tumorigenic cancer cells of various origin was shown to render such cells capable of malignant growth (36, 43). On the other hand, suppression of the angiogenic switch is thought to impose a lasting tumour growth inhibition (19, 21, 28). Also, a long latency period observed in the case of non-angiogenic clones isolated from the experimental breast, brain and osteosarcoma tumours stands in contrast to the rapid tumour formation by their angiogenic counterparts, in spite of largely comparable in vitro growth and clonogenic capacities (16). As angiogenic switching in cancer cells is, at least in part, related to oncogenic alterations (44) it is not surprising that some of the aforementioned angiogenic clones exhibited increased levels of c-myc and related downregulation of the angiogenic inhibitor thrombospondin 1 (Tsp-1) (16), a well-known myc target (45).

It should be noted that a vascular niche may contain several components. These may include angiogenic endothelial cells and their bone marrow-derived progenitors (EPCs) (46), bone marrow-derived vascular regulatory cells (VRCs) (47), pericytes, inflammatory cells, elements of the coagulation system (described below (28), as well as angiogenic tumour cells without tumour-initiating capacities and activated/angiogenic host stroma (48, 49). Indeed, it could be argued that the angiogenic phenotype itself is multicellular in nature, as single cancer cells that express reduced levels of Tsp-1, or other features, e.g. as a result of oncogenic mutations (50–52), are unlikely to change the local angiogenic balance amidst their Tsp-1-expressing non-transformed neighbours. In this regard, we have recently uncovered a paracrine mechanism by which even individual cancer cells expressing mutant ras oncogene could orchestrate a widespread Tsp-1 downregulation in a much larger population of adjacent fibroblasts (49). This ‘angiogenic field effect' was found to be mediated by a ras-regulated small molecule secretable factor (49). It is unclear how this complex network may specifically impact and involve CSCs in a given setting. However, it is of note that the vascular interface does control tumour growth and thereby (implicitly) the behaviour of CSCs. Indeed, glioblastoma cells harbouring a stem cell marker, CD133, were recently found to produce higher levels of vascular endothelial growth factor than their CD133- counterparts (53). Moreover, in brain tumours, CSCs sometimes reside at branching points of the tumour-associated capillaries (54). It is not clear what the link is between the CSC compartment (in various contexts) and tumour angiogenesis, but it could be speculated that, like other cells, CSCs may depend on a bloodborne supply of growth stimulating factors, oxygen and metabolites (14) and/or on paracrine effects of endothelial cells and other cells within the vascular system (14, 20). We suggest that also the contact with the coagulation system may affect the behaviour of CSCs, including their dormancy (28).


The interesting paradox in current studies on tumour angiogenesis, antiangiogenesis and dormancy is that they rely on mouse models where tumours occur at a very young age, unlike the majority of human tumours (25). For instance, mice as young as 4–10 weeks old are used as recipients of tumour transplants while transgenic oncomice often succumb to their disease within a few months after birth, having experienced a disease onset at a much earlier age (25). It could be argued this may be roughly reflective of the circumstances in paediatric/adolescent cancer patients, but hardly corresponds to the onset of adult malignancy within the spectrum of the species-specific relative average life expectancy (8%–30% in mouse models and upward of 50% for adult human cancer patients) (25, 55). In addition, humans are susceptible to a range of age-dependent and independent vascular comorbidities, of which atherosclerosis affects virtually the entire adult population (at least to some degree) and to which mice are genetically resistant (25). As the combined effects of aging and vascular comorbidities profoundly modify tumour growth, angiogenesis, the status of circulating endothelial progenitors and responses to antiangiogenic therapies, the emerging question is: how do these processes impact CSCs? In this regard the experience with, at least some of the solid tumours that can enter a prolonged dormant state (e.g. breast cancer (12)), has been that their occurrence in older patients may be associated with a less aggressive behaviour (56). This raises an interesting, and presently unanswered, question as to the possible impact of the organismal aging on the capacity of local (57) and premetastatic (27) CSC niches to support the efficient tumour initiation. Moreover, aging may conceivably change the nature of CSCs themselves, as it does in the context of normal stem cells, e.g. those in the bone marrow (41). Indeed, as the aging process affects a number of stem cell pools (41), putative sources of CSCs (3), it could be speculated that aging should be considered as a factor that may regulate tumour initiation and recurrence and their antithetical onset of tumour dormancy (25, 28).


If the proximity to blood vessels was an irreplaceable constituent of the permissive CSC niche it would be expected that the perivascular space within tumours should be occupied by cells harbouring markers of CSC. Indeed, in certain brain tumours, especially in medulloblastoma, evidence to this effect has recently emerged (54). In this elegant study the nestin-positive tumour (stem) cells have been identified in immediate apposition to tumour blood vessels (54). However, this may not be a generally applicable paradigm, as can be inferred from studies testing tumour-initiating properties of cancer cells from hypoxic and perivascular tumour regions (20, 58). Isolation of cancer cells residing in areas immediately adjacent to perfused blood vessels can be accomplished by a systemic injection of mice harbouring experimental tumours with a viable dye Hoechst 33342, which binds to cellular DNA in a manner defined by the rate of its perivascular diffusion (59). When tumour masses are subsequently extracted and dissociated to single cells, the flow cytometry profile of this mixture reveals brightly fluorescent cells that correspond to a perivascularly located tumour cell subpopulation, while dim cells represent the cell subset residing in tumour areas distant from blood vessels (hypoxic). Using this method we have previously determined that cells with the most aggressive properties (capable of tumour initiation) were located throughout the tumour, while the perivascular zone was occupied by cells with a less aggressive phenotype (20). Moreover, repeated cycles of sorting and re-injection of human transplantable melanoma cells resulted in a derivation of distinct cellular variants, corresponding to either perivascular-type (proximal) or hypoxic-type (distal) tumour cell populations (58). Once again, the former cells were less aggressive when injected into secondary recipients and, surprisingly, had a higher vascular density than tumours derived from purified distal (hypoxic) cells (58). This raises a possibility that while both fractions may contain tumour-initiating cells (CSC), they may differ in either the abundance or the type of such CSCs, especially with respect to their respective ‘vascular' phenotypes. In other words, different subsets of CSCs may differ in their ability to home to perivascular or hypoxic tumour regions. It might be speculated that the ‘hypoxic-type' CSCs could more readily tolerate the ischemic conditions, i.e. exhibit a lower vascular dependence, or vascular demand, as proposed earlier (58, 60). Indeed, embryonic stem cells are capable of forming aggressive teratomas solely on the basis of their aberrant differentiation, which can be triggered by the ectopic inoculation of such cells (i.e. into inappropriate niche). Such cells were found to exhibit a dramatically different ability to home to perivascular, or hypoxic tumour regions, depending on their hypoxia-sensing apparatus. Thus, ES cells in which the expression of the hypoxia-inducible factor 1 alpha (HIF1α) was genetically disrupted were able to remain viable in hypoxic regions of teratoma, sites from which the wild-type ES (teratoma) cells were being progressively eliminated during the expansion of mixed HIF-1α-positive/HIF-1α-negative tumors (58, 61).

The fate of tumour-initiating cells in perivascular and hypoxic tumour regions may be affected by common oncogenic events occurring in human cancer. Thus, p53 tumour suppressor gene is known to regulate tumour cell survival in anoxia (62), as well as control their angiogenic properties (63). We speculated that, by extension, these properties of p53 could also impact the responsiveness of at least some tumours to antiangiogenic therapy (64, 65). Indeed, this was apparent when two isogenic colorectal cancer cell lines with either intact (HCT116), or deleted p53 gene (379.2) were injected into SCID mice and tumours tested for responses to an antiangiogenic cocktail containing metronomic doses of vinblastin and the neutralizing antibody directed at the VEGF receptor 2 (65). In this setting tumours composed of HCT116 (p53+/+) cells responded to treatment far more robustly than those containing their 379.2 (p53−/−) counterparts (65). Interestingly, in the latter type of tumours, the fraction of viable cells in areas of hypoxia was significantly increased. Moreover, in heterogeneous tumours composed of a mixture of p53+/+ and p53-/- populations the latter cells were progressively enriched, especially upon antiangiogenic treatment (65). It should be noted that both of these cell lines were clonal and contained tumour-initiating cells that are operationally equivalent to CSCs. Therefore, it could be postulated that these experiments provide an example of a genetic and phenotypic heterogeneity of isogenic CSCs, the subsets of which may vary with respect to the ability to reside in areas immediately adjacent to, or distant from the vasculature. This example also suggests that tumours may progressively generate subsets of CSCs with different (reduced) niche requirements. These requirements may define the conditions under which the respective CSCs could undergo, or be forced into a state of dormancy.

It is noteworthy that p53 may not be a sole predictor of the susceptibility of human cancers to antiangiogenic agents, or for that matter to the angiogenic, or hypoxic dormancy of CSCs in colorectal and other tumours (66). This is understandable in view of the fact that the reduced vascular demand may be executed by several different pathways and not just p53 (60). However, it should also be noted that the gene expression profiling programs commonly performed on human cancer samples tend to reflect the properties (e.g. p53 status) of the most abundant cell types, whereas more meaningful information in this regard could potentially be obtained from profiling CSCs selectively. In other words it is possible that the impact of antiangiogenic agents could be deduced more accurately from assessing the determinants of the vascular demand of, specifically, CSCs (e.g. their status of mutant p53, ras and other genes) rather than from the total population often dominated by CSCs' progeny.


One of the many complexities associated with tumour neovascularisation is its coexistence with the local and systemic abnormalities in blood coagulation, the latter of which is frequently referred to as cancer coagulopathy, or Trousseau syndrome (67). This syndrome was long thought to be driven by ‘unspecific' anomalies of the tumour vasculature, such as natural discontinuities of the angiogenic capillaries, or defective and leaky vascular walls of the established vessels (68). Other changes that facilitate contacts between platelets and plasma proenzymes of the coagulation system and the procoagulant surfaces of extravascular tumour tissues were also implicated (69). More recently, however, cancer-related coagulation has been linked to molecular events affecting endothelial cells. For instance, chronic exposure of these cells to high concentrations of the inflammatory and angiogenic cytokines (e.g. VEGF) was found to trigger several procoagulant changes, the most significant of which is the upregulation of tissue factor, the entity normally absent from the intravascular space (70, 71).

Tissue factor (TF) is a 47 kDa transmembrane protein with homology to cytokine type II receptors, which serves as the principal inducer of the coagulation cascade (69, 72). Upon ‘exposure' to blood, TF acts as a specific receptor for the circulating coagulation factor VII/VIIa (fVII/VIIa) (73, 74). In turn, formation of the TF/VIIa complex on cellular surfaces triggers proteolytic conversion of the circulating factor X (fX) to an active form (fXa) (along with activation of fIX). FXa activates small amounts of prothrombin (fII) to thrombin (fIIa) (75). FIIa generation is then amplified through the contribution of platelets, and factors Va, VIIIa and IXa. The resulting burst of IIa activity exerts multiple effects that are required for rapid blood clotting (75). For instance, IIa is responsible for conversion of soluble fibrinogen (fI) into insoluble fibrin, the main acellular component of blood clots. Thrombin also acts on cellular protease-activated (G protein-coupled) receptors (PARs, notably PAR-1, 3 and 4), thereby contributing to signalling events within platelets, endothelium and other cells, including cancer cells (76). Indeed, in the latter case thrombin was postulated to cause cessation of a dormant state (77). The TF/thrombin axis could represent a significant vascular regulator of the latter process and therefore it is worth considering how such effects could be executed.

Tumour-associated and TF-dependent coagulation could have a number of biological effects relevant to disease progression, both systemic (metastasis) and localized in nature (regulation of growth, invasion and angiogenesis) (68, 69, 78). These are mainly mediated by growth factors released from platelets, deposition of the fibrin matrix and cellular effects related to the interaction between thrombin and the PAR1 receptor present on cancer and stromal cells (69, 77, 79, 80). However, TF also possesses a more direct capacity to trigger cellular signalling. This is ascribed to the short cytoplasmic domain, which contains at least two phosphorylatable serines (81), as well as the capacity of this domain to interact with the cellular PAR2 receptor, e.g. on endothelial cells (82). Indeed, removal of the TF cytoplasmic C-terminus in mice results in an exaggerated angiogenic activity and faster tumour growth (82). Moreover, TF-/- mouse embryos die in mid gestation amidst anomalies in endothelial-stromal interactions (83, 84) (albeit not all studies are consistent in this regard (85). On the other hand, adult mice with constitutively low levels of TF expression (low-TF mice) exhibit anomalies in tumour blood vessel remodelling (Yu, Milsom, May, Mackman & Rak, unpublished observations also in (28)). These observations suggest that TF could impact tumour neovascularization and formation of the vascular niche that supports or inhibits CSCs.

Cancer cells themselves are likely the major source of TF in malignant tumours (69, 79, 81, 86, 87). Thus, various types of tumour cells express up to 1000-fold greater TF activity than their normal counterparts, and the elevated expression of TF protein and mRNA has been observed in a number of advanced human malignancies, including colorectal, pancreatic, prostate and brain tumours, as well as leukemias and lymphomas (69, 88–92). While the reasons for this seemingly cancer-specific upregulation of TF have long remained unknown and are still incompletely elucidated, recent evidence suggests the role of oncogenes and tumour suppressors (93). In this regard our studies indicated that cancer cells harbouring oncogenic K-ras, HER-2 and epidermal growth factor receptor (EGFR) exhibit much higher expression and activity of TF on their surfaces than that of their isogenic but non-transformed counterparts. Importantly, oncogenic changes also induce cancer cells to emit TF as cargo of procoagulant microvesicles that are released into the culture media and blood of tumour-bearing mice (86, 94, 95). Recent studies revealed that TF is regulated by the loss of p53 (86) or PTEN (92) tumour suppressor genes in solid tumours, and in leukemia by expression of the chimeric PML-RARα (96) oncogene (95). In several instances these influences are modulated by cellular adhesion (97), hypoxia (92) or by residual differentiation programs of cancer cells (28). The observation that quite often TF is also released from cancer cells harbouring active oncogenes (as procoagulant microvesicles) is of particular importance for cancer coagulopathy, but also for tumour progression as such. This is because TF-containing microvesicles can bind to TF-non-expressing cells and platelets (98) and (potentially) render them capable of triggering the coagulation cascade in remote vascular beds, or systemically, possibly aiding in metastatic dissemination, a process which is closely linked to TF expression (99–101). Indeed, while cancer-related procoagulant events may occur for reasons other than TF upregulation (102), it is also true that tumour growth (69, 86, 103, 104), metastasis (99–101) and angiogenesis (86, 104) are often profoundly affected by genetic, or pharmacological manipulations that alter the status/activity of the tumour-associated TF (69).

These findings suggest that: (i) TF is linked to increased disease aggressiveness, including through its overexpression downstream of oncogenic events (95); and (ii) such TF overexpression is coupled with a number of tumour growth-promoting features, possibly as a function of pericellular coagulation, TF signalling, or both (81). This may suggest that the expression of TF is relevant for tumour initiation events, and thereby to the interplay between CSCs and their permissive or non-permissive niches. Indeed, the cascade of interactions triggered by formation of the TF/VIIa complex can potentially influence this intricate equilibrium. For instance, cancer cells expressing TF would be expected to be (and often are (68, 69)) surrounded with aggregates of activated platelets, and thereby exposed to the repertoire of platelet-related growth factors, adhesive fibrin matrix and concentrated enzymatic activities of the coagulation system, some of which (e.g. thrombin) may possess growth regulatory properties in their own right (e.g. via PAR1) (69) (28, 69). The consequences of this apposition may depend on whether such cancer cells also possess properties of CSC. This is because in the case of CSCs the aforementioned components of the coagulation pathway could potentially influence in a particularly consequential manner (positively or negatively) the cellular decision as to the entry into the tumour initiation (repopulation) phase, or retention as a quiescent (dormant) component within the primary tumour or metastatic site. Indeed, coagulation is often activated by processes known to be linked to tumour cell repopulation events, such as lesion recurrence post chemotherapy, injury or inflammation and metastatic dissemination. For this reason we proposed that the effectors of the coagulation (and fibrinolytic) system assembled in the pericellular space of CSC could be viewed as an equivalent of the TF-dependent and coagulation-related (provisional) cancer stem cell niche (8, 28) (Fig. 1). Interestingly, the long history of using anticoagulants as agents capable of exerting anticancer effects (78), and more recently some of the clinical trials with low molecular weight heparin (LMWH) and other agents (105–108) suggest that their effects could be attributed to the interference with the CSC population (28). For instance, in one such clinical study patients benefited from LMWH (beyond thromboprophylaxis) by prolongation of their disease-free survival (dormancy), but only when treated in the premetastatic stage of the disease (105). This would suggest that early growth (tumour initiation) of disseminated metastatic cancer cells could be particularly susceptible to this anticoagulant agent, an effect that could potentially be attributed to perturbations in the coagulation-dependent CSC niche effects.

Figure 1.

The postulated role of tissue factor (TF)-dependent provisional cancer stem cell niche as a mechanism regulating dormant behaviour of premalignant and pre-metastatic cancer cells. Oncogenic events are thought to be particularly consequential when they affect tissue stem cells. The resulting generation of cancer stem cells (CSC) creates a tumorigenic potential within, what initially is a dormant tumour. Oncogene-driven expression of TF by CSCs, or non-genetic upregulation of TF by their adjacent host stromal cells may trigger coagulation. This leads to accumulation of growth-promoting activities such as growth factors (GF), platelets (plts) and fibrin. The combined effect of these influences acts as a provisional cancer stem cell niche, an event that may initiate tumour growth followed by angiogenesis (see text). Similar changes triggered by coagulation-dependent effects, and intercellular signals emanating from TF and PAR receptors may lead to ‘awakening' dormant metastatic cancer cells (adapted from 28).

If the coagulation system contributed to the formation of the CSC niche it could be expected that TF activity (or other procoagulant features) should be associated with either the niche, or with the CSCs themselves. While existing evidence in this regard is either absent or largely circumstantial a few pieces of intriguing data have recently emerged. Thus, highly tumorigenic human squamous cell carcinoma cells, A431, in which the main oncogenic event consists of the amplification and constitutive activation of the EGFR, also express copious amounts of TF protein and activity (109). Interestingly, a small fraction of A431 cells also carries a stem cell marker CD133. When this latter minority subpopulation was isolated by immunomagnetic sorting it was found that CD133-positive tumour cells express up to five-fold greater amount of TF and the related procoagulant activity than their CD133-negative counterparts (8). Moreover, treatment of mice harbouring A431 tumours with a TF-directed neutralizing antibody led to a delayed tumour growth and diminished microvascular density (MVD) (28). Such treatment also induced a prolonged latency period in the case of tumours generated by injections of threshold numbers of A431 cells (Milsom et al. unpublished observation, also in (28)). While there is no definitive evidence that CD133-positive A431 cells represent a functional equivalent of CSC in this setting, it is thought provoking that these cells were markedly more procoagulant and TF-positive, and that TF status clearly impacted the tumour initiation processes (8, 86).

The role of TF in the realization of the tumorigenic potential of CSCs is also apparent in the context of metastasis. Interestingly, while the dependence of primary tumours on the expression of TF by cancer cells is a subject of some controversy (e.g. compare (86, 104, 110) with (101, 111)) there is a much greater congruence of data supporting the role of this receptor in metastasis (99–101), especially in experimental models where cancer cells are injected directly into the circulation and form clonal colonies in lungs, or other organ sites (112). As the number of such colonies is often several orders of magnitude lower than the number of cells initially injected into the circulation, this assay may be amongst the most direct demonstrations of the existence and frequency of CSCs in a given tumour cell population. Indeed, the fate of the injected cells has recently been clarified with considerable precision and, contrary to prior analyses (112), it entails some cell death, but also a frequent entry of cancer cells into solitary dormancy, as well as much less frequent overt metastatic growth (113, 114). Various experimental manipulations lead to the enhancement, or depletion of lung nodules originating from a given number of intravenously injected cells. In some experimental settings a decrease in number of nodules is attributed to the exposure of circulating cancer cells to (and cytotoxic action of) natural killer (NK) cells (101). However, it is not entirely clear whether NK cells actually kill cancer cells in all these cases, or at least in some cases force their entry into an ostensibly more prevalent state of dormancy (113). In this context it is interesting to note that cancer cells in which TF expression was genetically eliminated exhibit a significant deficiency in formation of pulmonary nodules (100, 101), while at the same time they may be capable of forming robust subcutaneous tumours (101). Similarly, anticoagulants (hirudin, heparin, low molecular weight heparin), or genetic depletion of fibrin appear to have a much greater effect on formation of clonal lung nodules than on subcutaneous tumours in mice (69, 80, 115). While it is often argued that this discrepancy is inherent to metastasis per se, versus fundamentally different primary tumour growth, another plausible explanation could be that these two types of experiments (intraveneous injection causing clonal growth and subcutaneous injection causing polyclonal growth) represent completely different scenarios from the perspective of the tumour initiation process, and the related involvement of CSCs. Thus, the i.v. injection of TF-negative and -positive cancer cells (lung metastasis model) essentially interrogates the ability of these individual cells to initiate clonal growth in the mostly ectopic organ site. This could be seen as an equivalent of the typical ‘cancer stem cell experiment’ where subthreshold cell numbers are injected subcutaneously to reveal CSCs (3). On the other hand, standard injection of excessive numbers of cancer of the same type into the subcutaneous connective tissue leads to delivery of multiple tumour-initiating cells and in the presence of their growth factor-rich progeny, all of which coexist in one tissue location. Under these conditions the probability of tumour growth initiation is strongly affected (increased) by the large numbers of CSCs present at this one site. The experiments aimed at testing the role of TF under bona fide tumour initiation conditions (low cell numbers) are currently ongoing. Nonetheless, it is plausible that the expression of TF by CSCs may mediate assembly of the coagulation-dependent provisional niche and affect the decision of CSCs to initiate tumour growth or enter dormancy.

TF-dependent cancer stem cell niche may also be formed with the involvement of host stromal cells. This is evident from studies in which expression of TF was noted in tumour-associated endothelial cells and macrophages (70, 116), but also from experiments with human xenografts, in which tumour-bearing immunodeficient mice were treated with neutralizing antibodies that selectively target either tumour (human) or host (mouse) TF activity (103). While the former treatment is far more potent than the latter, an impact of mouse TF is also measurable, suggesting (103) that host TF may contribute to the CSC niche. In all these instances TF could play a dual role, as an inducer of the coagulation cascade and as a signalling receptor responsive to stimulation with factor VIIa (81).


Tumour dormancy and its cessation at the time of disease recurrence represents a major conceptual and practical challenge. Harnessing the mechanisms that control and maintain the state of dormancy, either in solitary, quiescent metastatic/cancer cells or in a dynamic equilibrium of mitogenically active micrometastases, may enable a better disease control in cancer without disease eradication (14). This may potentially represent a more realistic therapeutic objective than striving to eliminate the entire and largely elusive tumour burden in cancer patients. Of special interest in this regard is the question of dormancy of cells that possess properties of CSC, or conversely the secondary acquisition of CSC phenotype (e.g. through mutations or epigenetic alterations) within the dormant tumour deposits. It should be noted that the tumour vasculature represents a critically important regulatory element that can influence CSCs and tumour dormancy in more than one way, and via several of its many facets (14). Meanwhile, it is generally believed that an angiogenic switch may lead to awakening of dormant cells (14, 21) even if the burst of angiogenic activity is only transient (117). However, in early-stage human melanoma the initial hypervascularity is associated with signs of histological regression (118), and indolent melanoma cells are growth inhibited by products of endothelial cells (e.g. IL-6) (119). Similarly, while certain aspects of the coagulation system may contribute to the permissive CSC niche and promote tumour growth, our observations suggest that this effect may also be biphasic. For instance, high but regulated (endogenous) expression of TF (e.g. under the influence of oncogenes) usually stimulates tumour progression and may be required for tumour growth and angiogenesis (28, 86). On the other hand, high constitutive and non-regulated (exogenous) levels of TF may sometimes impede tumour formation in certain experimental settings (Milsom & Rak, unpublished observation (120)). Similarly, thrombin may possess growth stimulatory and inhibitory effects in various ranges of concentrations and against various cellular targets (77). Thus, the direction in which vascular elements regulate tumour dormancy (stimulation versus inhibition) as well as the related behaviour of CSCs and their niches may vary in different diseases, with host and time-related contexts. For these reasons a better understanding of these linkages is imperative in order to apply the dormancy regulation and vascular-directed therapies in a rational and ultimately more effective manner.


This work was supported by an operating grant to JR from the National Cancer Institute of Canada (NCIC) and Canadian Cancer Society as well as Terry Fox Foundation and to some extent by a grant from the Cancer Research Society (CRS). JR is the recipient of a Scientist Award from the NCIC and the Jack Cole Chair in Pediatric Oncology at McGill University. CM is the recipient of a Studentship from the Bank of Montreal. We are in debted to our families and colleagues for their support and feedback.