|Home | About | Journals | Submit | Contact Us | Français|
Patients with cancer can develop recurrent metastatic disease with latency periods that range from years even to decades. This pause can be explained by cancer dormancy, a stage in cancer progression in which residual disease is present but remains asymptomatic. Cancer dormancy is poorly understood, resulting in major shortcomings in our understanding of the full complexity of the disease. Here, I review experimental and clinical evidence that supports the existence of various mechanisms of cancer dormancy including angiogenic dormancy, cellular dormancy (G0–G1 arrest) and immunosurveillance. The advances in this field provide an emerging picture of how cancer dormancy can ensue and how it could be therapeutically targeted.
The vast majority of cancer-related deaths are due to metastatic tumour growth that impairs the function of vital organs1. Metastatic lesions invariably originate from disseminated tumour cells, which often undergo a period of dormancy2 (FIG. 1). Cancer recurrence after therapy and long periods of remission is frequent. For example, 20–45% of patients with breast or prostate cancer will relapse years or decades later3-5 (FIGS 1,,2).2). In fact, most cancer types are associated with disseminated disease that after treatment might persist as minimal residual disease (FIG 2; TABLE 1). However, the lack of mechanistic insight into this stage has been a major shortcoming in our understanding of the full complexities of metastatic growth. Functional characterization of disseminated dormant tumour cells is important because these cells most probably contain the information about the future progression of the disease (that is, metastasis development). To fully understand dormancy, cells must be characterized during the dormant state. Therefore, given that these cells are present in a wide variety of cancers, information gathered from studies of cancer dormancy might be applicable to a large number of patients.
Cancer dormancy can be separated into mechanisms that antagonize the expansion of a dividing tumour cell population (tumour mass dormancy) and mechanisms that result in tumour cell growth arrest (tumour cell dormancy, or cellular dormancy) (FIG. 2). In the former, tumour cells usually divide but the lesion does not expand beyond a certain size because of either limitations in blood supply or an active immune system. Cellular dormancy can occur when tumour cells enter a state of quiescence (see BOX 1 for information on the relationship between quiescence, senescence and dormancy). These general mechanisms might explain the dormancy of residual cells that, following treatment, develop loco-regional or distant organ recurrences within different time frames.
A common attempt to explain cellular dormancy is to catalogue it within known mechanisms of growth arrest such as senescence or quiescence. However, whether quiescence or senescence programmes drive tumour cell dormancy is still incompletely understood.
Senescence represents a tumour suppressive mechanism that becomes activated in normal cells in response to replicative or oncogenic stress; the latter activates replicative and oxidative stress104. Specific markers, such as senescence-associated β-galactosidase (SA-β-gal), upregulation of p16, activation of p53, development of senescence-associated heterochromatic foci and a persistent active metabolism despite a G1 arrest, are found in senescent cells105. Senescence might be an irreversible state depending on the machinery that is engaged104. Because dormancy occurs in tumour cells and is reversible, a permanent senescence does not seem to be a plausible mechanism. More importantly, successful tumour development must have escaped senescence106,107. However, perhaps some aspects of the senescence programme can still be invoked in malignant cells, indicated by the positive correlation with SA-β-gal and p16 staining in tumour samples from patients who have been treated with chemotherapy108. However, whether the tumour cells that undergo senescence are the source of the relapse is unknown. It will be important to determine if dormancy and senescence programmes intersect because the abundance of information on the latter might be applicable to understanding tumour cell dormancy.
Alternatively, cellular quiescence might be a better mechanistic fit. Quiescence is thought to be due to a G0–G1 arrest and most adult cells reside in this state89,109. A reversible quiescence might be observed in stem cells or in cells that withdraw into a G0 arrest because of lack of growth-promoting signals (such as serum withdrawal and contact inhibition, for example)89,110. During quiescence, cells pause several cellular activities and this can result in the activation of selective programmes that render cells refractory to differentiation111. Furthermore, quiescence that is induced by growth-factor withdrawal results in attenuation of translational and metabolic activity and the regulation of specific transcriptional programmes109,111,112.
This Review discusses the relevance of cancer dormancy and how this state might manifest. Furthermore, the Review summarizes key discoveries from both experimental and clinical studies that shed light on how cancer dormancy is a part of disease progression and therapeutic response. It is important to note that the mechanisms that have been identified in experimental models have not fully been found to exist in patients, although all are highly plausible. For example, although there is clinical evidence for mechanisms of disseminated tumour cell growth arrest, there is limited documentation of angiogenic dormancy. By discussing the experimental and clinical evidence side by side this Review attempts, rather than to render these mechanisms as homologous, to provide potential mechanistic links between experimental and clinical research. Given the dearth of information on dormancy, it is crucial to take advantage of available mechanistic information, and novel discoveries and hypotheses from the experimental work reviewed here, as they might offer possibilities for translating our knowledge of cancer dormancy into the clinic.
It is thought that genetically fit tumour cells that emerge in the primary tumour (which is proposed to be a ‘late’ event in cancer progression) will be able to metastasize6,7. This was thought to be due to the time that must elapse for tumour cells to mutate and acquire traits that allow them to go through the different steps of the metastatic cascade (FIG. 1). However, recent studies suggest that tumour cell dissemination might occur early (that is, by less genetically progressed cells) and that disseminated tumour cells progress towards more aggressive phenotypes that will lead to metastatic growth in parallel with the primary tumour8. Were the disseminated tumour cells that remain after treatment to proliferate continuously, the relapse time would be expected much earlier than is actually observed9. Therefore, a pause in progression (that is, dormancy) has to be considered the most likely explanation for the discrepancy between the estimated and observed disease-free periods9. Tumour dormancy is observed in local recurrences or metastases. In the case of a primary tumour, the term commonly used is latency — the time that separates the carcinogenic insult from the clinical detection of the tumour. In all cases a portion of the latency period can be attributed to the slow accumulation of genetic alterations that lead to immortalization (that is, loss of TP53, RB1 (retinoblastoma 1), p16 (also known as cyclin-dependent kinase inhibitor 2A) and/or gain of telomerase, and so on) and transformation (that is, gain of Ras-activating mutations, ERBB2 (also known as HER2 or Neu) amplification, BRAF-activating mutations, and so on) after and/or during carcinogenesis6. However, after tumour cells are fully transformed it is not clear how much of their latency is attributable to tumour mass dormancy or cellular dormancy (FIG. 2).
Dormancy of loco-regional lesions, disseminated tumour cells or micrometastases (FIGS 1,,2),2), refers to the pause in cancer progression and the absence of clinical symptoms following treatment of the primary lesion10,11. In the case of disseminated tumour cells or micrometastases, it is in their final destinations (for example, bone marrow, lymph nodes, lungs or liver) that these cells are protractedly unable to resume growth. As mentioned above, a reason for this may be that dissemination might occur very early from ‘non-invasive’ lesions or in patients who are node negative2,8. Studies using comparative genomic hybridization (CGH) have shown that early dissemination of breast tumour cells with limited genetic abnormalities might be followed by progression and microevolution systemically8. These studies propose that, given that genetic abnormalities in diagnosed metastatic disease are quite homogeneous (that is, metastases share similar genetic abnormalities between patients), the stochastic and protracted nature of metastatic disease might depend on the rate at which genetic abnormalities progress in early disseminated tumour cells. Therefore, it is proposed that in the vast majority of patients a dormancy or ‘lead time’ (that is, the time it takes for genetic anomalies to accumulate) of ~5 years precedes relapse12. Alternatively, a true dormancy state, which most probably involves cellular tumour dormancy, might persist beyond 5 years in some patients despite the presence of additional genetic alterations in the disseminated tumour cells12. If early dissemination is applicable to other tumours it might explain a portion of the dormancy time of the disease. However, these cells must still survive during the ‘lead time’ and they seem to be growth arrested13. This suggests that although survival pathways are functional, the mechanisms that propel the growth of the primary lesion are insufficient for metastatic growth. Additionally, the microenvironment and host genetics (see below) might influence genetic progression. These data suggest that oncogenic signalling might not always be dominant and that other programmes (that is, ‘stem’ cell quiescence and stress signalling, or microenvironment restrictions (see below)) might overcome oncogenic signals, allowing tumour cell survival in a dormant state. The fact that disease-free periods in patients can last ~5 years and as much as 20–25 years (for example, in breast and oral cavity cancers)3,14, suggests that a pause in disease progression occurs often and might be explained by different forms of dormancy; this could be a feature of cancer progression (FIG. 2). When dormancy is induced and what mechanisms operate in different cancers is still unclear. However, answers to these questions are most likely to be found in the residual tumour cells during the remission periods after treatment of a diagnosed cancer. understanding the dormancy of remaining disseminated or local tumour cells is crucially important, as metastases or local recurrences, respectively, are invariably derived from these cells.
Disseminated tumour cells usually lodge in a non-orthotopic tissue microenvironment and their ability to interpret it can determine tumour cell fate15. Therefore, a failure to resume proliferation in these sites might result from stress signals that have been activated by a new non-permissive microenvironment or a niche that is conducive for quiescence (FIGS 2,,3).3). By contrast, tumour cells that can remodel and/or reproduce a niche or orthotopic microenvironment that is conducive for expansion might be able to resume growth16-18. Data from different labs reviewed below reveal that deregulation of the reciprocal interaction between the microenvironment and tumour cells might decide between proliferation and growth arrest15,19,20. Some of these studies provide mechanistic evidence for cellular tumour dormancy (FIGS 2,,33).
In some cases, cellular tumour dormancy might result from the ability of tumour cells to attain a ‘differentiated’ state. using reconstituted basement membrane assays it has been shown that breast cancer cells rapidly proliferate in a disorganized fashion with an absolute loss of tissue architecture21. Their growth and aberrant organization is dependent on β1-integrins and epidermal growth factor (eGF) signalling, as inhibition of these signals leads to the re-differentiation of these tumour cells into non-proliferating acinar structures21. Blocking of β1-integrin signalling in these tumour cells in vivo ablates tumour growth but it is not known whether this is due to induction of a protracted state of growth arrest. It is also unclear whether such mechanisms occur in dormant disseminated tumour cells in patients. nevertheless, these results indicate that, upon correcting their ability to interpret their microenvironment appropriately, tumour cells can undergo some degree of differentiation and enter a non-proliferative state21.
A similar switch between proliferation and growth arrest controlled by tumour cell–microenvironment crosstalk was observed in head and neck carcinoma19 (FIG. 3). In this model, the metastasis-associated urokinase receptor (uPAR) drives tumour growth by interacting and activating α5β1 integrins19. This complex in turn recruits focal adhesion kinase (FAK) and the eGF receptor (eGFR), which, in a fibronectin-dependent manner, propagates mitogenic signals through the Ras–extra-cellular signal-regulated kinase (eRK) pathway19,22,23. Blocking of uPAR, β1-integrins, FAK or eGFR, singly or in combination with each other, resulted in tumour suppression in vivo, which is due to induction of dormancy19. In these studies, although dormant tumour cells expressed the appropriate integrins and were surrounded by fibronectin, the downregulation of uPAR and loss of integrin function prevented these cells from transducing proliferative signals from the fibronectin-rich microenvironment24. Imaging and measurements of eRK signalling in vivo revealed that dormancy resulted from an almost complete inhibition of the Raf–MeK–eRK pathway and induction of a G0–G1 arrest as in quiescent cells18. Interestingly, activation of stress pathways is also required for dormancy induction. The disruption of the uPAR complex activates the p38 mitogen-activated protein kinase (MAPK) signalling pathway18,24,25. Furthermore, growing metastatic lesions show sustained eRK activity but greatly reduced p38 signalling, suggesting that proliferation in primary and secondary tumours requires a high eRK:p38 signalling ratio whereas the opposite favours cellular dormancy18.
Moreover, inhibition of p38α and/or β in dormant cells is sufficient to interrupt their growth arrest and resume proliferation in vivo18,24,25. Induction of dormancy by p38 activation required upregulation of p53 (but not p21, p15 or p16) (BOX 1), downregulation of Jun and attenuation of protein synthesis mediated by inhibition of eukaryotic initiation factor-2α (eIF2α) activity (REF. 26 and A. P. Adam, A. C. Ranganathan and J.A.A-G., unpublished observations). The possibility that p38 signalling might exert tumour suppressive functions on metastatic cells is in accordance with data showing that p38α/β activation is associated with tumour suppression by causing growth arrest, senescence or apoptosis27 (BOX 1). Whether these mechanisms are operational in patients is unclear. However, uPAR has been found to be a prognostic indicator and correlate with the recurrence of lesions after minimal residual disease in gastric cancer28 (see below). Therefore, it seems reasonable to suggest that strategies that are aimed at inducing and/or maintaining tumour cell dormancy should include concomitantly blocking receptor tyrosine kinases and uPAR, and/or integrin complexes to inhibit the tumour cell–microenvironment crosstalk (FIG. 3).
Tumour cell dormancy has also been described in a transgenic mouse model for breast cancer, in which polyoma middle-T (PyMT) antigen or eRBB2 signalling was studied in mammary gland tissue that was null for β1 integrin20. These studies show that whereas control littermates develop tumours at high frequency, the loss of β1 integrin results in tumour suppression. Additionally, the PyMT-positive tumour cells that lack β1 integrin expression were in a non-proliferating, dormant state20. Again, a loss of signals from the microenvironment favoured dormancy. loss of FAK-dependent signalling is attributed as a potential mechanism, consistent with previous results in squamous carcinoma cells22,23 (FIG. 3), in which FAK inhibition uncoupled α5β1 integrin from eGFR, leading to tumour cell dormancy.
Systemic signals that influence both stromal and tumour cells might also have a role in regulating dormancy. upon transplantation into virgin female mice, pregnancy-dependent tumours that have been induced by insertional mutations of mouse mammary tumour virus (MMTv) remain dormant for at least 300 days29. Stimulation by hormones that are produced during pregnancy or hormonal treatment caused these tumours to emerge from dormancy29. The ‘dormant’ state here is different from the main mechanisms described above as it is mostly due to a very slow rate of proliferation and it is not clear if cell death compensates for cell division. This slow-cycling but still dynamic tumour cell population seems to acquire new mutations that explain why subsequent transplants become hormone-independent. As in the above examples, growth-factor signalling was important for interruption of tumour dormancy as MMTv insertional mutations in fibroblast growth factor 3 (FGF3) and genes at the integration site 7 (Int7) locus correlated with the emergence of hormone-independent tumours29,30. Other pathways that depend on β1 integrin, including retinoic acid-induced protein 2, Rspo3 (a thrombospondin, type I, domain-2-containing protein) and endonexin, among others, might be required as the loci of these genes have been described as additional insertion sites in recurrent tumours29,30.
The above studies suggest that reduced tumour cell–microenvironment crosstalk might induce dormancy through quiescence or ‘differentiation’ of malignant cells. This might explain the behaviour of disseminated tumour cells in mouse models of experimental breast tumour metastasis10,11,31. In these studies, although most disseminated cancer cells die, there is a fraction that remain viable and do not proliferate in lungs and liver, but form tumours in the orthotopic mammary fat pad10,32. Similar results were obtained when detecting experimental brain metastases through the use of nano-ferromagnetic particles and magnetic resonance imaging (MRI)11. It is important to emphasize that in these early stages, solitary dormant tumour cells are surrounded by normal functioning tissue vasculature. Therefore, rather than invoking a mechanism that depends on neovascularization (see below), it is more likely that deficient crosstalk with the new microenvironment (such as those described above) and/or stress signalling (for example, JUN N-terminal kinase (JNK) and p38) explain the impaired proliferation of these cells. Interestingly, the identification of genes that selectively inhibit the growth of metastases, but not the primary tumour (that is, metastasis suppressor genes), revealed that they can inhibit mitogenic signals and activate stress signalling pathways33,34.
Metastatic growth is crucially dependent on the ability of tumour cells to survive the stress of dissemination and the new microenvironment, as the majority (99.9% (REFS 10,35)) of disseminated tumour cells die (FIG. 1). The loss of metastasis suppressor gene expression or function36-38 is thought to explain the development of metastases39. Their function seems to be dominant in solitary cells that reach their target organs37. Genes of interest are those that curtail metastasis by inducing apoptosis or dormancy once the cells have lodged (intra- or extra-vascularly) at the secondary site39. KISS1 (kisspeptin) blocks metastases through the induction of prolonged dormancy of solitary cells40. This response is not mediated by its receptor, G-protein-coupled receptor 54, suggesting that either a paracrine signalling mechanism or an alternative receptor is important. The tetraspanin protein CD82 (also known as kangai-1), which has a role in adhesion signalling, inhibits mouse melanoma cell metastases when bound to the endothelial cell-expressed ligand DARC (Duffy antigen chemokine receptor, which is also known as CD234)36. In vitro analysis implicated tumour cell senescence that was induced by the CD82–DARC interaction as the inhibitory mechanism36. However, whether senescence is operational in vivo and whether it is reversible (dormancy implies reversibility) is unknown. This is important to determine, because it suggests that dormancy of metastases might be driven, at least in part, by mechanisms of reversible senescence (BOX 1).
Other metastasis suppressors such as nM23 (which is encoded by non-metastatic cells 1 (NME1) and NME2) seem to inhibit Ras signalling or, as in the case of MKK4 (MAPK kinase 4), can activate the JnK and p38 stress pathways34. In prostate cancer, MKK4 is lost and its suppressive function was found to operate through JnK activation41. In ovarian cancer, MKK4 is also lost42 and it exerts its metastasis suppressor function through the activation of p38 (REF. 38). The fate of these cells and the mechanism of dormancy are still unknown (that is, senescence, cellular dormancy, apoptosis, angiogenic switch or immunosurveillance). If JnK or p38 activation is exceedingly strong, cells might undergo apoptosis43. However, achieving p38 signalling levels that induce growth arrest without apoptosis25 might represent an adaptation or survival mechanism that favours subsequent re-growth. In many ways dormancy is a survival mechanism26,44 (see BOX 2 for information suggesting that cellular stress adaptation might recapitulate programmes of organismal stress adaptation). Therefore, mapping the pathways that metastasis suppressor genes regulate might offer targets for therapeutically preventing the development of metastases.
Dormancy is a phenomenon that is observed in cancer but also in whole organisms. For example, in Caenorhabditis elegans development can be paused at the dauer stage (that is, diapause) until the conditions are again propitious for successful development113. Similarly, a plant seed can be dormant until the appropriate signals (that is, light, humidity or salinity) favour germination114. The entry into a stage of diapause was shown to be a selective advantage that helps organisms adapt to stress with the ultimate goal of producing offspring. In C. elegans, the pathways that control the dauer stage (dauer formation (DAF) genes and the insulin-like growth factor-1 signalling pathway) are involved in lifespan extension115. Similarly, pathways that sense oxidative stress (such as the p38 mitogen-activated protein kinase pathway) were shown to induce a dauer stage and these larvae are more resistant to various types of stress such as nutritional deprivation116-119. Therefore, the theme of growth arrest or retardation coupled to stress resistance might be a basic programme of survival that is recapitulated by tumour cells and applicable to some of the models described here. If tumours progress following rules that apply to evolutionary and ecological processes120 then the acquisition of traits that allow for a pause in growth and an adaptation phase might fit this paradigm. It is possible that tumour cell dormancy recapitulates evolutionary conserved mechanisms of adaptation, robustness and resistance to stress that are imposed by dissemination and the ‘hostile’ microenvironment of the target or transit organs (the bone marrow stroma, for example)120,121. Testing whether pathways that control the dauer stage are activated in disseminated tumour cells might be informative about dormancy. For example, do disseminated tumour cells show nuclear localization of Forkhead transcription factors, which when inactivated induce growth arrest in mammalian cells and dauer in C. elegans, or enhanced expression or activation of pathways that are involved in lifespan extension, stress resistance and/or growth arrest pathways (such as the unfolded protein response and p38 pathway)116,119,122? This might be interesting to explore, as genes that are involved in lifespan extension seem to suppress tumour formation in C. elegans and humans115,123.
An exciting possibility is that metastatic efficiency might be modulated by polymorphisms that are present in the heterogeneous genome of the human population45. Crossing PyMT-expressing mice into different syngeneic strains and assessing both primary tumours and metastatic efficiency showed that, although primary tumour formation is not affected, metastatic efficiency is either down- or upregulated depending on the F1 genetic makeup. Therefore, the genetic background of the host might determine metastatic efficiency46. This led to the identification of a metastasis modifier locus (Mtes1), which encodes, among other genes, Sipa1 (signal-induced proliferation-associated gene 1), a RAP1-GTPase activating protein47. These studies reveal that Sipa1 is a metastasis-promoting gene and that polymorphisms in the PDZ protein–protein interaction domain might cause loss of function, reducing metastatic efficiency, which could be because of the induction of dormancy47. It will be exciting to determine whether polymorphisms in humans cause loss of function in metastasis modifier genes that, for example, suppress metastases through the induction of cellular dormancy (that is, growth arrest) or through the suppression of angiogenesis.
once tumour cells emerge from quiescence and the tumour mass reaches a certain size that cannot be supported any longer by the normal tissue vasculature, it will become highly dependent on the availability of oxygen and nutrients. This leads to the development of a vascular bed through sensing low oxygen levels, and recruiting blood vessels (angiogenesis) — by inducing the expression of hypoxia-inducible factor 1 (HIF1), for example48 — and lymphatic vessels (lymphangiogenesis). Interestingly, tumour cells, although proliferation competent, might be unable to recruit new blood vessels or remodel the pre-existing vasculature as the tumour reaches a certain size49. This leads to a state of anoxia or hypoxia in the tumour mass that triggers cell death in the tumour cell population48. If cell death exceeds division, then these lesions would most probably regress. However, in tumours that are unable to recruit blood vessels the fraction of dying cells equals the dividing ones and there is no net increase in tumour mass over time50. This results in angiogenic dormancy (FIG. 2).
The ability of tumour cells to respond to hypoxia, promote neovascularization and interrupt angiogenic dormancy is known as the angiogenic switch (for excellent reviews see REFS 50,51). A balance between pro-angiogenic factors (such as vascular endothelial growth factor (veGF) and platelet-derived growth factor (PDGF)) and anti-angiogenic factors (thrombospondin, endostatin, vasculostatin and angiostatin) is thought to control this switch, which involves the regulation of transcriptional programmes48,52. Therefore, a tumour mass might enter this state if a tumour cell population with an active programme of proliferation lacks angiogenesis or hypoxia-response sub-programmes52. The balance between negative and positive angiogenesis factors can be regulated by the balance between genes and pathways that are involved in tumour promotion or suppression53-55 (FIG. 2). For example, active Ras signalling seems to oppose the induction of anti-angiogenic factors through a Rho and MYC-dependent pathway53. It is intuitively logical that pathways that regulate cell-cycle progression also regulate the expression of neovascularization factors. Why then are proliferating tumour cell populations unable to induce angiogenesis? It is possible that the distinct genetic alterations or dosage of a mutated gene (such as oncogenic Ras53) might be sufficient to favour cell-cycle progression, but might be below a threshold level for the angiogenic programme. A possible scenario can be envisioned in which growth arrest or slow cycling of early transformed cells might explain an initial pre-angiogenic phase of primary tumour dormancy18,56. once these cells progress and acquire new mutations that self-perpetuate cell-cycle progression, a second programme might be required to turn on the angiogenic switch51 (FIGS 2,,44).
Interruption of dormancy through the re-induction of angiogenesis is poorly understood (FIG. 2). For example, in a non-orthotopic mouse model of subcutaneous lymphoma growth caused by inducible MYC expression, sustained cancer regression can be observed upon oncogene de-induction57. However, in lymphomas that are p53 null, a disease-free period is invariably followed by relapse, which is inhibited by the expression of the p53 target gene thrombospondin 1, an anti-angiogenic factor57. Additional studies show that a brief and strong angiogenic stimulus might reverse angiogenic dormancy58. Human leukaemia cells (MOlT-3 cells) that are unable to form growing tumours in non-obese diabetic severe combined immunodeficient (NOD/SCID) mice but form dormant lesions can resume growth after a brief stimulation with VEGF or co-inoculation with angiogenic-competent Kaposi sarcoma cells. This stimulation was suppressed by exogenous angiogenesis inhibitors58. Growing and dormant tumour cells express similar levels of pro-angiogenic factors, suggesting that re-initiation of angiogenesis might need higher levels of angiogenic inducers than vascular maintenance. Alternatively, although not addressed by the authors58, downregulation of endogenous angiogenic inhibitors after the pro-angiogenic ‘spike’ might explain this result.
The above studies were carried out on primary tumour masses, but it is less clear whether angiogenic dormancy occurs in patient metastases or whether current anti-angiogenic therapies59 induce angiogenic dormancy. evidence from experimental models using lewis lung carcinoma and T241 fibrosarcoma cells in C57Bl6/J mice supports dormant metastases being in a balance between mitosis and apoptosis60. However, this must occur after a period of expansion because disseminated tumour cells can grow intravascularly, bypassing the need for new blood vessels61. In addition, clinical evidence for this mechanism is weak owing in part to the difficulty of detecting and molecularly characterizing micrometastatic lesions or accessing these samples after autopsy. Furthermore, there is evidence that solitary disseminated tumour cells in animal models and in patients remain growth arrested and do not express proliferation markers13,62 (see below). An unanswered question about angiogenic dormancy is how the angiogenic switch can remain inactive for years in a dynamically dividing tumour cell population that is prone to accumulating additional genetic changes that could activate this switch. Given that solitary tumour cells survive on the pre-existing vasculature, a cessation of proliferation triggered by signals that are determined by the new tissue microenvironment, as discussed previously, is the most probable explanation for their cellular dormancy32. Alternatively, a tumour mass, although angiogenesis competent, might be prevented from expanding and kept dormant by an active immune response63.
The role of the immune system in controlling tumour growth has been recognized for decades63-65. Studies using syngeneic animals that are immunized with a subcutaneous implantation of tumour cells and then challenged by intraperitoneal injection of the same cells have shown that the immune system can target and kill most of the tumour cells in the challenge injection63,66. These responses are mostly mediated by cytotoxic CD8+ T lymphocytes, which induce cytolysis of the tumour cells63 (FIG. 2). However, it seems that some residual cells persist and this population might be kept clinically dormant by the immune system64,67. Therefore, the immune system prevents the expansion of proliferating tumour cells66. In other studies, interruption of tumour dormancy of mouse lymphoma and leukaemia was proposed to result from reduced expression of tumour-associated antigens and evasion of the immune system66,67. Studies of BCl1 mouse lymphoma that developed spontaneously in a BAlB/c mouse revealed that a humoral response that controls a regulatory network of anti-idiotypic antibodies activates B-cell receptor signalling, which inhibits proliferation and is followed by apoptosis or dormancy (for a comprehensive review see REF. 68). In this model dormancy was also shown to be mediated by CD8+ T cells69. By contrast, residual and dormant DA1-3b mouse acute myeloid leukaemia cells overexpressed B7H1 (also known as CD274 antigen and programmed cell death 1 ligand 1) or B7.1 (B-lymphocyte activation antigen B7) co-stimulatory molecules70. These were proposed to abrogate cytotoxic CD8+ T-cell-mediated responses allowing dormant leukaemia cells (that is, those in cellular dormancy) to evade recognition by the immune system70 (FIG. 2). However, it is not clear whether other mechanisms might be required because B7H1 does not mediate anti-tumour immunity in other models70.
Additional studies have shown that proliferating mouse lymphoma cells are kept at a low number in the bone marrow owing to persistent antigen and memory T cells that are able to coordinate a CD4+ and CD8+ T-cell-mediated response71-73. These results correlate with clinical studies showing that in the bone marrow of patients with breast cancer who carry cytokeratin-positive cells (breast cancer cells), a higher proportion of memory T cells among the CD4+ and CD8+ cells correlated with larger tumours72. These results suggest that in some situations the immune system might still be operating to suppress residual tumour cell expansion, whereas other mechanisms of dormancy favour immune-system evasion (FIG. 2). Additional studies are needed because it is still unclear whether immuno-surveillance is an operational mechanism controlling dormancy of non-virally induced tumours74. How the immunological response intersects with angiogenic dormancy or cellular dormancy is unknown. It is possible, although unknown, that the genes that control escape from dormancy (for example, quiescence or angiogenic-switch dependent) might also coordinate the ability of tumour cells to evade an immune response.
Characterization of minimal residual disease revealed that these cells can remain dormant and perhaps some of the mechanisms described above (for example, enhanced uPAR and/or eRBB2 signalling) might explain the transition between dormancy and recurrence. Furthermore, the potential existence of tumour stem cells and drug resistance mechanisms, although less characterized in residual tumour cells, might help explain the biology behind dormancy of minimal residual disease (FIG. 3). of note is the fact that although the evidence from epithelial and haematopoietic cancer dormancy is discussed jointly, ultimately some of the mechanisms might diverge because of intrinsic differences between these tumour types. For example, mechanisms that depend on the regulation of uPAR and eRBB2 will most probably not be applicable to lymphoma dormancy. However, although the dormancy-initiating signals might be different in these cancers, it is possible that they might converge on similar general mechanisms (that is, transcriptional repression or methylation) to induce, for example, tumour cell quiescence.
Minimal residual disease is a significant problem as it is observed in many cancers and, in breast cancer for example, the relapse rate after >5 years is about 20% (REF. 75; TABLE 1). These residual tumour cells are usually found in circulation or in the bone marrow (FIGS 2,,3),3), which are the most commonly inspected sites, but they most probably lodge in several organs that are not usually scrutinized. Minimal residual disease can also be detected in the lymph nodes or target organs for metastases, such as liver or lung1. The population of disseminated tumour cells in the bone marrow is of significance as increased counts of cells that express certain biomarkers, such as cytokeratin (see below), indicate poor prognosis even for tumours that do not usually metastasize to the bone (for example, colon cancer)1,76. Therefore, detecting these cells in bone marrow might be useful for predicting bone or other secondary organ metastases (for reviews see REFS 1,13).
Characterization of disseminated tumour cells has proven to be extremely difficult because of their low abundance (1 disseminated tumour cell per 106 bone marrow cells and 1–2 disseminated tumour cells per 20 ml of blood)13,77. Recent intriguing studies have shown that patients who are free of clinically detectable disease >20 years after treatment still have circulating disseminated tumour cells77. It was proposed that because tumour cells could not persist arrested in circulation for so long, a tumour stem cell population that sheds tumour cells into circulation might be responsible for this phenomenon77 (see below) (FIGS 2,,33).
Disseminated tumour cells in the bone marrow can be examined for expression or overexpression of specific biomarkers, such as Ki67, p120 (also known as CTNND1), EGFR, ERBB2, extracellular matrix metallo-proteinase inducer (eMMPRIn) or uPAR; for genomic rearrangements (by CGH); or for mutations in specific genes (for example, KRAS and TP53)1,8. However, only a few studies have looked at stem cell markers and it is not clear if these serve as prognostic indicators78 (see below). In other studies, the absence or low proportion of disseminated tumour cells staining for markers such as Ki67 and proliferating cell nuclear antigen (PCnA)1,2,8,77,79 indicates that these cells are in G0–G1 arrest, which is indicative of quiescence as a form of cellular dormancy (FIG. 3). However, the mechanisms that are operational in these cells are poorly understood because simultaneous detection of multiple markers is usually difficult. In gastric cancer, the detection of uPAR in cytokeratin-positive disseminated tumour cells in the bone marrow predicted relapse28. By contrast, disseminated tumour cells that lacked uPAR staining predicted longer disease-free periods and better overall survival28. However, whether lack of uPAR staining correlated with the absence of proliferation markers is unknown.
On the basis of the data discussed above it is tempting to speculate that in disseminated tumour cells downregulation of uPAR might be responsible for their dormancy19,22,23,28,76,80,81 (FIG. 3). Similar findings were reported for prognosis and expression of eRBB2 in disseminated breast cancer cells13. It is possible that using the inducible systems for eRBB2 expression82, as well as other oncogenes (for example, MYC and Wnt57,83,84) some of these questions could be modelled in mice. However, it is unclear whether there is reduced eRBB2 expression, despite the presence of gene amplification in some cases, in disseminated tumour cells, so the relevance of the inducible systems is uncertain. Similarly, it is important to determine whether disseminated tumour cells that express eRBB2 also express proliferation markers. If not, then the mechanisms by which eRBB2-positive cells become growth arrested should be explored. If uPAR and eRBB2 upregulation serve as a switch to interrupt dormancy, these might be important targets for combination therapy of residual disease. Interestingly, the PLAUR (encoding uPAR) and ERBB2 genes were found to be co-amplified in disseminated breast cancer cells81. Recently, identification of the interaction site for α5β1 integrin with domain III of uPAR85 and the development of monoclonal antibodies that target this region19,86 has provided an excellent opportunity to target this mitogenic pathway85 in combination with anti-ERBB2 drugs (for example, trastuzumab, lapatinib and EKB-569). Together, the available data suggest that disseminated tumour cells can be found in a non-proliferating state and that some of the mechanisms that have been discovered in experimental models might aid in the understanding of disseminated tumour cell behaviour as well as providing new markers or targets for therapeutic applications.
It is still unclear whether stem cells are indeed the cells that harbour the genetic alterations that cause cancer. However, properties of normal stem cells, such as their quiescent state within a niche that is both nurturing and protective87, might potentially explain tumour cell dormancy. Specific recruitment signals, such as tissue injury, activate stem cells to exit quiescence and proliferate88; tumour stem cells might also retain this property89,90. Therefore, current hypotheses suggest that metastases are the consequence of disseminated tumour stem cells that, after undergoing a state of quiescence, subsequently resume growth (FIG. 3). Recent studies report that disseminated breast cancer cells in the bone marrow are enriched for markers of breast tumour stem cells: ~65% of cells in the bone marrow were CD44+/CD24-/cytokeratin (CK)+ compared with <10% in the primary tumour78. However, the fate of these cells is unknown and whether only these cells are the source of relapse is also a mystery. It is also important to determine whether CD44+/CD24-/CK+ cells in the primary lesion, lymph nodes or bone marrow are in different proliferating states. The existence of tumour stem cell quiescence also implies that in those cells with defined genetic abnormalities that drive tumour progression (that is, oncogene addiction) the stem cell quiescence (cellular dormancy) programme might be dominant over oncogenic signalling. For example, interferon therapy can induce durable remission in a minority of patients who display residual Philadelphia chromosome (BCR–ABl oncogene)-positive chronic myeloid leukaemia (CMl) progenitors90. Furthermore, although genomic BCR–ABl-positive leukaemia cells can persist in CMl patients who exhibit complete cytogenetic remission, the BCR–ABl transcript levels are absent or extremely low90,91. These studies suggest that despite the presence of the oncogenic aberration(s), other signals can overcome oncogene expression in dormant cells. Although these are exciting hypotheses, much work needs to be done to understand whether tumour stem cells are indeed the origin of relapse.
It is assumed that dormant disease is chemotherapy resistant because dormant cells are not dividing1,9,92. However, it is still unclear whether this mechanism explains cancer cell drug resistance in patients or whether the active survival mechanisms that have been discovered in experimental models protect disseminated tumour cells. Induction of p21 or p27 in colon cancer cells causes G1 arrest and doxorubicin resistance in vitro 93. In vivo studies of green fluorescent protein (GFP)-tagged disseminated breast cancer cells showed that they are growth arrested and resistant to doxorubicin92. In addition, active mechanisms might protect dormant cells from chemotherapy. Breast cancer cells that re-differentiate after inhibition of β1 integrin activity are more resistant to chemotherapy94. Surprisingly, this is independent of proliferation but dependent on the level of polarity (appropriate polarity activates an anti-apoptotic β4-integrin–nuclear factor-κB (NFκB) pathway94) and differentiation (FIG. 3). In squamous carcinoma cells in which a low eRK:p38 activity ratio induces dormancy, activation and induction of the eIF2α kinase RnA-dependent protein kinase-like endoplasmic reticulum kinase (eIF2AK3, also known as PeRK) and the chaperone protein HSPA5, respectively, protected dormant cells from chemotherapy independently of proliferation. HSPA5 did so by inhibiting BAX (BCl2-associated protein X) activation26. Accordingly, HSPA5 has been shown to protect breast cancer cells from oestrogen starvation through inhibition of the pro- apoptotic proteins BIK (BCl2-interacting killer) and BAX95. Interestingly, in breast and prostate primary tumours, HSPA5 detection is prognostic for a shorter time to recurrence and a poor response to either adriamycin in patients with breast cancer or hormonal ablation in patients with prostate cancer96,97. Therefore, disseminated tumour cells expressing HSPA5 might be more refractory to treatment (FIG. 3). Finally, normal and tumour stem cells express ABC transporters such as ABCG2 (ATP-binding cassette G2)98,99. In neuroblastoma, cancer stem cells expressing ABCA3 are mitoxantrone resistant100. However, the link between ABC transporter expression in disseminated tumour cells, relapse and poor prognosis is still unknown. More detailed analysis of these mechanisms and detection of these markers in disseminated tumour cells will be important for the choice of therapy when patients are positively identified as bearing minimal residual disease.
The future offers two exciting possibilities: first, to induce and/or maintain dormancy of tumour cells, and second, to induce cell death in residual dormant cells by targeting their survival and drug resistance mechanisms. It will also be important to determine whether a patient has dormant disease and what type of mechanism is active. The models described here will be instrumental in identifying the mechanisms and markers of dormant cancer. Cellular or serum biomarkers might help detect dormant disease, which will require detection of low abundance cells or circulating biomarkers. In addition, transcriptional profiles from dormant disseminated tumour cells or experimental models of dormancy might help determine whether primary tumours carry a cancer dormancy ‘signature’, which might have prognostic value101.
Many questions remain. For example, do the mechanisms of dormancy described in this article occur independently or are they all part of progression? (See FIG. 4 for an integration hypothesis.) Is dormancy a phenomenon of less progressed cancers? For example, are patients who relapse within months devoid of any dormant disease because of highly progressed tumour cells that can adapt and grow in the new microenvironment, while less progressed disseminated tumour cells can only adapt and survive in a dormant state? Do different types of tumour undergo dormancy through particular mechanisms? Can tumour-initiating (stem) cells undergo angiogenic dormancy and/or can they evade immuno-surveillance? What happens to survival of tumour cells that are highly dependent on an oncogene102 when, despite carrying oncogenic abnormalities, they become quiescent? Finally, it will be important to use the available mouse models of cancer to model more accurately aspects of human cancer progression including, but not limited to, early dissemination, surgery of a tumour mass, chemotherapy, residual disease and spontaneous, rather than experimental (that is, intravenously injected cells), metastases. The importance and urgency in understanding cancer dormancy is highlighted by a recent report on a u.S. national Cancer Institute-organized workshop on tumour cell dormancy103. The future understanding of cancer dormancy will most likely lead to a shift in how we think about cancer progression and in consequence how we diagnose and treat the disease.
I apologize to those authors whose work I could not cite directly because of space constraints. I would like to thank the members of my laboratory and L. Ossowski (Mount Sinai School of Medicine) for stimulating discussions and critical reading of the manuscript. I also thank D. Schewe for discussions and help with the figures. This work is supported by a grant from the Samuel Waxman Cancer Research Foundation Tumour Dormancy Program and the NIH/National Cancer Institute grant CA109182 to J.A.A.G.