A Perspective on Cancer Cell Metastasis Christine L. Chaffer, et al. Science 331, 1559 (2011); DOI: 10.1126/science.

1203543

This copy is for your personal, non-commercial use only.

Permission to republish or repurpose articles or portions of articles can be obtained by following the guidelines here. The following resources related to this article are available online at www.sciencemag.org (this infomation is current as of March 11, 2012 ): Updated information and services, including high-resolution figures, can be found in the online version of this article at: http://www.sciencemag.org/content/331/6024/1559.full.html Supporting Online Material can be found at: http://www.sciencemag.org/content/suppl/2011/03/24/331.6024.1559.DC1.html A list of selected additional articles on the Science Web sites related to this article can be found at: http://www.sciencemag.org/content/331/6024/1559.full.html#related This article cites 61 articles, 16 of which can be accessed free: http://www.sciencemag.org/content/331/6024/1559.full.html#ref-list-1 This article has been cited by 17 articles hosted by HighWire Press; see: http://www.sciencemag.org/content/331/6024/1559.full.html#related-urls This article appears in the following subject collections: Medicine, Diseases http://www.sciencemag.org/cgi/collection/medicine

Science (print ISSN 0036-8075; online ISSN 1095-9203) is published weekly, except the last week in December, by the American Association for the Advancement of Science, 1200 New York Avenue NW, Washington, DC 20005. Copyright 2011 by the American Association for the Advancement of Science; all rights reserved. The title Science is a registered trademark of AAAS.

Downloaded from www.sciencemag.org on March 11, 2012

If you wish to distribute this article to others, you can order high-quality copies for your colleagues, clients, or customers by clicking here.

SPECIALSECTION
REVIEW

A Perspective on Cancer Cell Metastasis

Christine L. Chaffer1,3* and Robert A. Weinberg1,2,3* Metastasis causes most cancer deaths, yet this process remains one of the most enigmatic aspects of the disease. Building on new mechanistic insights emerging from recent research, we offer our perspective on the metastatic process and reflect on possible paths of future exploration. We suggest that metastasis can be portrayed as a two-phase process: The first phase involves the physical translocation of a cancer cell to a distant organ, whereas the second encompasses the ability of the cancer cell to develop into a metastatic lesion at that distant site. Although much remains to be learned about the second phase, we feel that an understanding of the first phase is now within sight, due in part to a better understanding of how cancer cell behavior can be modified by a cell-biological program called the epithelial-to-mesenchymal transition.

etastasis is responsible for as much as 90% of cancer-associated mortality, yet it remains the most poorly understood component of cancer pathogenesis. During metastatic dissemination, a cancer cell from a primary tumor executes the following sequence of steps: It locally invades the surrounding tissue, enters the microvasculature of the lymph and blood systems (intravasation), survives and translocates largely through the bloodstream to microvessels of distant tissues, exits from the bloodstream (extravasation), survives in the microenvironment of distant tissues, and finally adapts to the foreign microenvironment of these tissues in ways that facilitate cell proliferation and the formation of a macroscopic secondary tumor (colonization) (1). Here we suggest that this complex metastatic cascade can be conceptually organized and simplified into two major phases: (i) physical translocation of a cancer cell from the primary tumor to the microenvironment of a distant tissue and then (ii) colonization (Fig. 1). We propose that an understanding of physical dissemination is in sight, whereas the second phase, colonization, involves complex interactions that may still require several years of research before they come into clear view. From a therapeutic standpoint, understanding the mechanisms of physical translocation is likely to be important for preventing metastasis in patients who are diagnosed with early cancer lesions, whereas understanding the mechanisms leading to successful colonization may lead to effective therapies for patients with already-established metastases.

In the discussion that follows, we provide our perspective on a selection of issues relevant to contemporary cancer metastasis research. Physical Translocation from the Primary Tumor to the Site of Dissemination In order for individual or small groups of cancer cells to break away from the primary tumor and initiate the metastatic process, these cells must acquire the ability to migrate and invade. These traits enable cells to degrade and move through the extracellular matrix of the surrounding tissue toward blood and lymphatic vessels, which in turn provide highways for their passage to distant secondary sites. The first clinical indication of metastatic dissemination may come from the presence of cancer cells in the draining lymph nodesthose connected directly with the site of primary tumor formation through lymphatic vessels; more often than not, these draining lymph nodes seem to represent dead ends rather than temporary stopping points from which more distant metastases are launched (2). In fact, spread to the anatomically distant sites seems to occur almost entirely through the blood via the process of hematogenous dissemination (3, 4). Carcinomas, the tumors on which we focus in this review, arise in epithelial tissues. Normally, the cells forming the epithelial sheets in these tissues are tightly bound to neighboring cells and to underlying basement membranes by adherens junctions, tight junctions, desmosomes and hemi-desmosomes, effectively immobilizing them in these sheets. These tight physical constraints encumber not only normal epithelial cells, but also those within many benign carcinomas. However, as a tumor progresses, carcinoma cells liberate themselves from these associations and begin to strike out on their own, first by dissolving underlying basement membranes and then invading adjacent stromal compartments. This invasiveness seems to empower carcinoSCIENCE VOL 331

1 Whitehead Institute for Biomedical Research, 9 Cambridge Center, Cambridge, MA 02142, USA. 2Department of Biology, Massachusetts Institute of Technology (MIT), Cambridge, MA 02139, USA. 3Ludwig MIT Center for Molecular Oncology, Cambridge, MA 02139, USA.

*To whom correspondence should be addressed. E-mail: weinberg@wi.mit.edu (R.A.W.); chaffer@wi.mit.edu (C.L.C.)

ma cells to both intravasate and subsequently extravasate (5). A central question, as yet unaddressed, is whether this acquisition of malignant traits occurs as an almost-inevitable consequence of primary tumor progression or as an accidental product thereof. A widely accepted but still unproven model of primary tumor formation posits that cancer cells acquire a sequence of genetic and epigenetic alterations, each of which confers one or another form of increased fitness (6). Each of these alterations can trigger a clonal expansion of the cells that have acquired it, leading to a succession of clonal expansions that resembles, at least formally, a scheme of Darwinian evolution. This working model raises the question of whether the development by cancer cells of aggressive traits, such as invasiveness and metastatic dissemination, reflects the fact that these traits are locally advantageous for carcinoma cells within the confines of a primary tumor. The alternative is more subtle: that multistep clonal evolution selects for cells that have greatly enhanced powers of survival and proliferation within primary tumors. Once formed, such cellsas an almost-accidental, unselected consequence of their phenotypic statecan now respond to contextual signals that induce them to express highly malignant traits. We do not resolve between these alternatives here, although observations described below may help to illuminate this question. Cancer stem cells and metastasis. One critical input into this discussion comes from recent observations that the neoplastic cells within individual tumors are not homogeneous. An important source of intratumoral heterogeneity has been revealed by the discovery that populations of cells within a tumor, like those in the corresponding normal tissues, are organized hierarchically (7, 8). Thus, the scheme of self-renewing stem cells (SCs), partially differentiated transitamplifying (i.e., progenitor) cells, and fully differentiated end-stage cells seems to be recapitulated in many carcinomas and other tumor types (9). The discovery of these cancer stem cells (CSCs) has forced a major rethinking of tumor biology, because a variety of cancer-associated traits that were at one time ascribed to tumor cell populations as a whole must now be associated with one or another of these subpopulations: the nonCSCs and CSCs within each tumor. (For clarification, we do not equate the biological properties of CSCs with normal tissue SCs; instead, we use the term to define a subpopulation of cancer cells with greatly enhanced tumor-initiating potential relative to other cancer cells within a tumor. CSCs should also display self-renewal potential and the ability to spawn non-CSC progeny.) These considerations become especially critical in the present discussion, because many of the biological traits of high-grade malignancy have now been traced specifically to the subpopulations

www.sciencemag.org

25 MARCH 2011

1559

Downloaded from www.sciencemag.org on March 11, 2012

Cancer Crusade at 40
would empower them to disseminate from primary tumors and seed metastases (15); hence, it is an attractive solution to understanding the mechanics of dissemination. Moreover, the heightened resistance to apoptosis that is integral to cells generated by an EMT is surely critical to the ability of carcinoma cells to survive the rigors of the voyage from primary tumors to sites of dissemination (18). In addition, the CSC-like state approached by carcinoma cells that have passed through an EMT may be critical in their sites of dissemination for launching new colonies of cancer cells. Merely because the EMT is an attractive solution does not make it a unique one, however, and it remains possible that other, still-undiscovered cell-biological programs operate in certain carcinoma cells as drivers of malignancy. Activation of an EMT program during tumorigenesis often requires signaling between cancer cells and neighboring stromal cells (19). Islands of cancer cells in advanced primary carcinomas are thought to recruit a variety of cell types into the surrounding stroma, such as fibroblasts, myofibroblasts, granulocytes, macrophages, mesenchymal stem cells, and lymphocytes; these recruited cells create a reactive stromain effect, an inflammatory microenvironment that appears to result in the release of EMT-inducing signals. The carcinoma cells respond to these contextual signals by activating expression of certain transcription factors (EMT-TFs) that proceed to orchestrate EMT programs within these cells. This scenario implies that activation of an EMT program flows from two major sources. First, the Physical translocation Colonization biology of the cancer cell-of-origin from primary tumor to distant organ before its transformation conspires E Survival at with the genetic and epigenetic A Acquisition secondary site changes sustained during primary of invasive C phenotype tumor formation to generate carcinoCTCs transit to ma cells that are responsive to EMTdistant organ inducing signals. Second, inductive signals released by the reactive stroma impinge on these responsive B D carcinoma cells, causing them to exLocal invasion CTCs extravasate press various EMT-TFs and thereby F cells invade into surrounding and invade into the activate previously latent EMT prostroma, then intravasate to enter Adaptation and parenchyma of grams. When portrayed in this way, hematogenous circulation proliferation to foreign tissue form metastases the EMT-associated traits are not direct products of genetic and epiDifferentiated Transitioning Cancer Inflammatory cell Stromal cell genetic evolution in primary tumors; cancer cell cancer cell stem cell instead, they represent adaptations to Fig. 1. The metastatic cascade. Metastasis can be envisioned as a process that occurs in two major phases: (i) physical contextual signals experienced once translocation of cancer cells from the primary tumor to a distant organ and (ii) colonization of the translocated cells primary tumors have formed. Though it is appealing in conwithin that organ. (A) To begin the metastatic cascade, cancer cells within the primary tumor acquire an invasive cept, the role of the EMT in enabling phenotype. (B) Cancer cells can then invade into the surrounding matrix and toward blood vessels, where they intravasate to enter the circulation, which serves as their primary means of passage to distant organs. (C) Cancer cells metastatic dissemination remains traveling through the circulation are CTCs. They display properties of anchorage-independent survival. (D) At the largely unproven, in part because of distant organ, CTCs exit the circulation and invade into the microenvironment of the foreign tissue. (E) At that foreign the technical difficulties of capturing site, cancer cells must be able to evade the innate immune response and also survive as a single cell (or as a small this transitory process in human cancluster of cells). (F) To develop into an active macrometastatic deposit, the cancer cell must be able to adapt to the cer patients. Moreover, the EMT may microenvironment and initiate proliferation. only operate in a tiny fraction of of CSCs within carcinomas (1012). Thus, traits such as motility, invasiveness, and self-renewal, which are central to malignancy, may in fact be the ref lection of the actions of the elusive CSC subpopulations within larger populations of neoplastic cells. In many tumors, such cells may represent a tiny fraction of the total cellularity of individual tumors, yet these CSCs may be the critical drivers of their malignant progression. At a more practical level, CSCs have been defined by their most central trait: the ability to seed new tumors when experimentally implanted into appropriate animal hosts. Though the representation of CSCs, assessed in this way, may fluctuate wildly from one tumor to another (13, 14), it is clear that carcinoma cells can reside in two, or even three, alternative states of differentiation within a tumor. Such subpopulations, when separated by fluorescence-activated cell sorting according to distinct cell-surface antigen profiles, show drastic differences in tumor-initiating ability. Because these cell fractionations stratify tumor cell populations that have not previously been subjected to experimental manipulation, it is clear that residence in these alternative states preexisted in primary tumors before their excision from hosts. At one level, the critical role of CSCs in metastasis is obvious: Tumor initiation by experimentally implanted cells is theoretically analogous to tumor initiation by disseminated cancer cells; both processes depend on the ability of cancer cells to function as founder cells that spawn essentially unlimited numbers of descendants. Hence, the very traits that are used to define CSCsselfrenewal and tumor-initiating abilitywould seem to be inextricable elements of successful metastasis formation. Less intuitive is the observation that CSCs exhibit yet other traits that are relevant to metastasis: notably motility, invasiveness, and heightened resistance to apoptosis (1012). This implies that there is a multifaceted cell-biological program packaged together to empower cancer cells within primary tumors to execute multiple steps of the invasion-metastasis cascade. The epithelial-to-mesenchymal transition and metastasis. Over the past three decades, developmental biologists have defined a cell-biological programthe epithelial-to-mesenchymal transition (EMT)that plays critical roles in early embryonic morphogenesis (15). This transdifferentiation program, driven by EMT-inducing transcription factors (EMT-TFs), is deployed during a number of critical steps of morphogenesis, enabling cells of epithelial phenotype to generate mesenchymal derivatives. Importantly, in many embryonic contexts, the EMT is reversible; thus, cells that were recently induced to assume a mesenchymal phenotype may revert back to an epithelial state via mesenchymal-to-epithelial transitions. Recent studies have demonstrated that the EMT can induce non-CSCs to enter into a CSClike state (16, 17 ). As such, the EMT confers on epithelial cells precisely the set of traits that

1560

25 MARCH 2011

VOL 331

SCIENCE

www.sciencemag.org

Downloaded from www.sciencemag.org on March 11, 2012

SPECIALSECTION
cancer cells that are in intimate contact with induced CSCs may be a feature of tumors able standing how cells are able to survive in the adjacent reactive stroma. This implies that ex- to recruit a reactive stroma capable of inducing circulation. Importantly, the presence of CTCs pression of an EMT program cannot possibly an EMT. This raises the question of whether the and of extravasated, disseminated tumor cells be discerned by examining the altered genome two types of CSCs play identical roles in tumor (DTCs), such as those in the bone marrow of patients with breast cancer, correlates with inprogression and differ only in their origins. of an unfractionated tumor sample. Plasticity between epithelial and mesenchy- creased metastatic burden, aggressive disease, A clear resolution of the role of EMTs in high-grade malignancies is complicated by one mal states will surely be important as we try to and a decreased time to relapse (22). In the longer term, a more thorough underfurther consideration: Although this program is understand the dynamics of how metastatic coloften depicted as a bi-stable switch that causes onies are initiated after disseminated carcinoma standing of CTC and DTC biology may gencells to flip from one state into the other, the cells extravasate into tissue parenchyma. Such erate highly useful diagnostic and prognostic biological reality is likely to be more subtle. In carcinoma cells may have previously undergone measurements (23). Whether these cells will remany tumors, epithelial carcinoma cells appear a partial or complete EMT within primary tu- veal important etiologic mechanisms remains to advance only partway down the road toward mors, having been induced to do so via hetero- unclear; for example, the detection of large numthe mesenchymal state. This yields cancer cells typic signals originating in the tumor-associated bers of CTCs may simply reflect a high level of that concomitantly express epithelial and mesen- stroma; components of the EMT program may aggressiveness of a primary tumor, rather than chymal markers and thus persist in a pheno- then have enabled their physical dissemination. revealing the particular cells that serve as the typic state that is not encountered in normal tissues. Intrinsic Induced Intrinsic versus induced cancer stem cells. The discussions above imply that (i) the stem-cell state is an integral part of the development of metastases, (ii) in many types of EMT carcinomas, entrance into this state is facilitated by passage through an EMT, and (iii) EMTs can be induced in carcinoma cells by contexCells poised to Recruitment of tual signals received, for example, undergo EMT reactive stroma Transition to from the tumor-associated reactive CSC-like state stroma. Taken together, these noCSC with tions imply that CSCs can be formed metastatic potential de novo through the actions of these signals. At the same time, these inResponsive Differentiated Transitioning duced CSCs cannot represent the differentiated cancer cell cancer cell Cancer stem cell Stromal cell Inflammatory cell cancer cell only source of cells in the CSC pool within a tumor: Intrinsic CSCs are likely to exist within tumors from Fig. 2. Acquisition of the metastatic phenotype. Tumors are heterogeneous populations of cells. CSC subtheir very inception, long before re- populations are particularly well poised to complete the metastatic cascade. Two alternative means of generating active stroma and EMTs become im- CSCs are depicted here. Intrinsic CSCs are thought to exist in primary tumors from the very early stages of tumorigenesis and may be the oncogenic derivatives of normal-tissue stem or progenitor cells. Induced CSCs may portant (Fig. 2). We speculate that the presence arise as a consequence of the EMT. In this case, carcinoma cells initially recruit a variety of stromal cells, such as of intrinsic and induced subtypes fibroblasts, myofibroblasts, granulocytes, macrophages, mesenchymal stem cells, and lymphocytes. Together these of CSCs within a tumor may partly cells create a reactive microenvironment that releases factors (e.g., Wnt, transforming growth factorb, fibroblast growth factor) that cause the neighboring cancer cells to undergo the EMT and acquire CSC-like characteristics. explain the heterogeneity evident in clinical tumor pathology, where highly aggressive tumors (such as the claudin- However, it is plausible that after extravasat- key intermediaries between primary tumors and low and basal-type breast cancers) exhibit a ing, these cells will not encounter an activated metastases. Our understanding of CTCs and the roles normal mammary stem cell gene profile and thus stroma and, in the absence of associated strocontain high numbers of intrinsic CSCs, whereas mal signaling, may well lapse back to a fully that they play in metastatic dissemination is still luminal-type breast cancers correlate with a ma- epithelial state that lacks CSC function. Still, clouded by some major unresolved biological ture mammary luminal cell phenotype and thus such cells cannot afford to jettison CSC function issues and by technical issues arising from decontain low numbers of intrinsic CSCs (20, 21). completely if they are to serve as founders of a tection sensitivities of these rare cells. Circulating As such, in some subtypes of cancer, intrinsic metastatic tumor; this suggests that complex carcinoma cells have diameters (20 to 30 mm) preneoplastic SCs are likely to be present al- mechanisms operate to maintain the mesenchymal/ that are far too large to allow them to pass ready during the early stages of tumorigenesis. CSC state, even in those CSCs whose recent his- through the bores of capillaries (~8-mm diameter), Such SCs, which already should possess some tory suggests that they are induced rather than such as those present in the capillary beds of the lungs (4). By all rights, within minutes of being of the EMT-associated phenotypes, may play intrinsic CSCs. Circulating tumor cells. The blood of many released by primary tumors into the venous cirprominent roles in disseminating carcinoma cells long before frankly malignant tumors have patients with advanced primary carcinomas con- culation, CTCs should be trapped in these captains circulating cancer cells, at least a subset of illaries during their first pass through the heart. developed. Such thinking confines the induced CSCs to which may be in transit from the primary tumor Yet some persist for far longer periods of time later stages of tumorigenesis when the extensive to sites of future metastasis. These circulating tu- [with half-lives of 1 to 2.4 hours (24 )], which reactive stroma is first apparent. Accordingly, mor cells (CTCs) offer the prospect of under- suggests the possibility that only exceptionally www.sciencemag.org SCIENCE VOL 331 25 MARCH 2011

1561

Downloaded from www.sciencemag.org on March 11, 2012

Cancer Crusade at 40
small or physically plastic CTCs can elude the sieving action of the pulmonary microvasculature and thereby accumulate to substantial steadystate concentrations in the blood. Moreover, if CTCs occasionally travel in multicell clumps of far larger diameter, such clumps must lodge almost immediately in microvessels and thus have extremely short lives in the circulation; such clumps would be strongly underrepresented by current methods that tally CTC numbers in patients. In addition, the tissue factor protein displayed on the surface of individual cancer cells attracts clouds of aggregating platelets (25, 26), which also increase the effective diameters of the cancer cells. This suggests that the CTCs found in the circulation may represent special subpopulations of cancer cells that, for one reason or another, do not trigger platelet aggregation. These platelet cloaks may also complicate the detection of CTCs by occluding the cellsurface marker antigens that are used to identify and separate CTCs from the million-fold greater numbers of nonepithelial cells in the circulation. Epithelial-specific cell-surface markers are in widespread use to detect CTCs (2729). This represents a potential problem, because it is likely that carcinoma cells that have passed through a partial or complete EMT are no longer detectable by epithelial-specific antigens. Enrichment and detection of CTCs via depletion of hematopoietic cells (using antibodies specific for CD45) may represent one way of circumventing this issue in the future (23). Despite these complications, the ability to isolate CTCs from the circulation may become a powerful tool to study the biology of migrating CSCs, especially if those cells differ from CSCs that reside in the primary tumor site. Moreover, they offer the prospect of creating a highly useful (and relatively noninvasive) diagnostic parameter: By monitoring longitudinally the concentrations of CTCs in a cancer patient, oncologists may be able to determine, from day to day, how effectively an applied therapy has been in reducing the burdens of primary tumors that are presumably the sources of these CTCs. Homing. Once lodged in the capillary bed of a foreign tissue, CTCs may soon extravasate and invade the foreign parenchyma, or they may proliferate intraluminally and eventually rupture the wall of the microvessels in which they are lodged (3032). The formation of metastases in certain favored target organs may be influenced by structural differences in the capillaries of various tissues. For example, the sinusoid capillaries in the bone marrow are formed from single layers of endothelial cells and are devoid of supporting mural cells; this design is thought to facilitate the normal trafficking of hematopoietic cells in and out of the bone marrow (33). This route may also present a path of least resistance to carcinoma cells, and thus may help to explain why the bone marrow is a favored site of metastasis by cancer cells that originate in diverse primary tumors (e.g., breast, prostate, lung, and gastric cancers) (34). In certain tumor types, the layout of the circulation may be the strongest determinant of metastatic tropism. Most frequently cited is the behavior of colorectal carcinomas (CRCs), which have a strong preference for generating liver metastases. In fact, the disseminating CRC cells may be intrinsically poorly adapted for survival in the liver microenvironment. However, because of the portal circulation, which drains from the mesentery directly into the liver, myriad carcinoma cells may be dumped over extended periods of time into the liver microvasculature; on rare occasion, an otherwise low-probability event may then generate a liver metastasis. In these various cases, homing to a particular organ can be considered to be a passive process that is determined by circulation patterns and the physical properties of the vasculature rather than by particular biological properties of the disseminating cancer cell. Organ-specific homing may also constitute an active process, where tissue and cancer cell specific features determine metastatic dissemination. The expression by metastasizing cancer cells of specific proteins (for example, integrins) seems to play a key role in this process (3537). We envisage that homing of metastasizing cancer cells may be a combination of both mechanical trapping of cancer cells in the microvasculature of distant organs and cancer cellmediated adhesion to specific luminally displayed components of the vasculature (Fig. 3). Colonization-Adaptation of the Disseminated Cell to the Microenvironment at the Metastatic Site Although EMT programs may prove critical to the physical dissemination of carcinoma cells, the multiple powers of these programs would not seem capable of addressing the problem of colonization. Instead, colonization seems to represent a far more complex set of phenomena and a relatively small number of unifying, generalizable principles. This complexity can be judged, perhaps simplistically, by listing common tumors and their known metastatic tropisms. For example, as noted above, prostate carcinomas preferentially metastasize to bone, whereas CRCs preferentially metastasize to the liver (38). Nonetheless, these and other primary tumors can also form metastases at additional, alternative tissue sites. In each case, the tissue microenvironment of a primary tumor is likely to differ markedly from that of the secondary site of dissemination, necessitating substantial adaptive moves by recently arrived cancer cells. The details of these adaptive programs would seem to be dictated by the microenvironment of the starting point (the primary tumor) and the microVOL 331 SCIENCE environment of the landing site (the tissue parenchyma in which a metastasis is founded). This logic suggests that the number of distinct adaptive programs can be gauged by a simple calculation: the product of multiplying the number of metastasizing primary tumor types by the number of distinct sites of dissemination. However, clearly some adaptive moves, such as the activation of certain ensembles of genes, may simultaneously confer an ability to colonize multiple distinct tissues, reducing the diversity of adaptive programs. Moreover, we know that certain lung cancers metastasize quickly to multiple sites, whereas others, such as breast and prostate carcinomas, often take years to develop metastatic colonies and then only in a relatively limited number of sites (38). This suggests that the differentiation programs of certain normal cells, such as those in the lungs, position their neoplastic descendants to adapt readily to foreigntissue microenvironments, whereas other cancer cell types must laboriously cobble together far more complex shifts in gene-expression programs. Moreover, it is unclear how often colonization depends on epigenetic changes versus genetic mutations in cell genomes. Nonetheless, solid progress is being made toward understanding the biochemical adaptations that carcinoma cells must make to thrive in distant tissues. Gene sets have been defined in breast tumor xenografts that can predict homing and colonization of breast cancer cells specifically to the lung, bone, or brain (3941); conversely, genomic profiling of metastases has been successfully used to predict the sites of primary tumor origin (42). These gene-expression patterns suggest that carcinoma cells within primary breast tumors acquire patterns that enable their subsequent colonization preferentially to specific target organs. These findings are supported by recent studies demonstrating that genetically distinct subpopulations of cells present in primary tumors are responsible for forming metastases (43, 44). Precisely how these gene-expression programs are acquired by the cells within primary tumors is not yet clear. They may be dictated by the differentiation programs of normal cells of origin, by somatic genetic and epigenetic changes that have been selected during multistep tumor progression, or simply stochastically. Interestingly, the discovery of reseeding of primary tumors by their derived metastases (45) raises a fourth possibility: Metastatic cells may contaminate the gene-expression patterns of the corresponding parental primary tumor by introducing gene-expression programs that were selected during earlier metastatic colonization. In general, colonization is an extremely inefficient process, and most cancer cells that successfully translocate from the primary tumor to a secondary site undergo apoptosis within 24 hours of extravasation (4648). Experimental and clinical data support the notion that the survivors per-

1562

25 MARCH 2011

www.sciencemag.org

Downloaded from www.sciencemag.org on March 11, 2012

SPECIALSECTION
sisting in metastatic sites can exist in at least three alternative states: (i) as solitary viable cancer cells in a quiescent, nonproliferative state (dormancy); (ii) as micrometastases, which remain as small lesions (probably due to a balance between proliferation and apoptosis); or (iii) as actively growing macrometastatic lesions (Fig. 3). Our current understanding of the mechanisms governing the entrance into these three states and the transitions between them is limited because of the experimental challenges of studying dormancy, the extended times required for these processes to reach completion, the paucity of appropriate models, and the technical challenges surrounding the analysis of single cells in vastly larger living tissues (49). Successful colonization is presumed to include the ability to acquire mitogenic stimulation from growth factors and cytokines that are naturally present in the alien microenvironment to self-renew and generate a large flock of descendantsto recruit the necessary supporting stroma, including an appropriate blood supply (4, 50). Micrometastases composed of actively proliferating cells represent attractive venues for cancer cells to develop complex colonization programs. Thus, we imagine that active cell division is essential for the generation of genetic and epigenetic alterations; once the resulting variants arise in these micrometastases, their novel phenotypes can be tested for an ability to confer selective advantage in the presence of highly demanding, otherwise-inhospitable microenvironments. It is also possible that the tissue microenvironment at the secondary site may change over time as a result of aging, disease, or wounding and that these environmental changes function as triggers that induce the release of disseminated cells from the dormant state. Several models of mouse metastasis now suggest that factors derived from primary tumors can educate distant sites in preparation for, and before the arrival of, metastasizing cancer cells (5153). In these studies, factors secreted by the primary tumors (e.g., VEGF-A, PlGF, PSAP) are thought to mobilize bone marrow derived cells that are subsequently attracted to premetastatic sites. The cells of this premetastatic niche then release factors (e.g., SDF-1, S1000A8, S100A9) that can attract disseminating tumor cells (Fig. 3). This interesting concept is still in the early stages of investigation. Consequences of successful colonization. The outcome of successful colonization is a rapidly expanding macrometastasis that can now serve as the focus for disseminating a shower of secondary metastases. Importantly, many of the cancer cells that are dispatched from this recently successful metastasis may be invested with a functional colonization program that may empower them to colonize either a limited subset of sites throughout the body or, alternatively, multiple distinct tissue types. The throngs of secondary metastases derived from this shower will soon eclipse the single initiating metastasis that spawned them. Certain microenvironments are almost guaranteed to provide hospitable sites for such disseminating, colonization-competent cancer cells: sites of wound-healing and stroma of alreadyestablished tumors. Strangely enough, one insight has come from dentists who have observed tumors growing out of extracted tooth sockets within weeks of oral surgery (54); the tumors arising in these sites of active wound healing represented the first clinical manifestations of previously undetected, widespread metastatic site for settling and colonization by disseminated CTCs, including those deriving from the contralaterally implanted tumors and those generated by the tumor itself. These diverse observations underscore the notion that, in contrast to normal tissue stroma, the stromata of aggressive tumors and sites of wound healing can serve as readily colonizable microenvironments that make few adaptive demands on the cells that have disseminated to them. Conclusions One of the most revealing findings of the past decade in cancer research is that cells within any

Homing
Mechanical trapping A Site-specific adhesion or B chemoattraction

Colonization
Micrometastasis Macrometastasis

D
Quiescence

C
Differentiated cancer cell

Pre-metastatic niche

Cancer stem cell

Stromal cell

Inflammatory cell

Extracellular matrix

Fig. 3. Adaptation of metastatic cells to a foreign environment. Homing and colonization of a cancer cell to a distant organ are complex processes with many questions still unanswered. CTCs transiting from the primary tumor to a metastatic site can arrive at their destination via a variety of mechanisms: (A) CTCs may become lodged in the capillary beds of specific organs due to size. (B) CTCs may display specific adhesion molecules that enable them to adhere to microvessels in specific organs, or they may respond to a chemoattractive gradient arising from a particular tissue. (C) CTCs may preferentially home to organs where a premetastatic niche has prepared a microenvironment conducive to their survival. (D) Once cancer cells have exited the blood stream (extravasated) they may first experience a period of quiescence (dormancy) while they adapt to their newfound microenvironment. (E) Dormant cells may progress to micrometastatic deposits (perhaps in response to the recruitment of an appropriate stroma or an enhanced ability to respond to proliferative signals present in the host microenvironment) where their size is kept in check because of a balance in proliferation, apoptosis, and phagocytosis by the host-tissue immune system. (F) To develop into a macrometastasis, cancer cells must recruit an adequate blood supply (necessary for growth beyond 1 to 2 mm). The signals or mechanisms responsible for the transition from dormancy to micrometastasis to macrometastasis remain largely unknown.

disease in these patients. Indeed, the similarity between wound-healing environments and the hospitable stroma of tumors was encapsulated in an observation made years ago that tumors are like wounds that will not heal (55). This theme has been extended by the recent observations cited above (45) that the cells from contralaterally implanted aggressive tumors in mice can metastasize to one another and at apparently high efficiency. The suggestion here is that the reactive stroma that arose in each of these tumors (and contributed to their locally aggressive phenotypes) also generated a hospitable SCIENCE VOL 331

given carcinoma display a great deal of heterogeneity. This intratumoral heterogeneity is due, in no small part, to subpopulations of cells that are phenotypically distinct but genetically identical; for instance, cancer cells that are less and more differentiated can share a common set of genetic alterations. These distinct phenotypic states, involving CSCs and non-CSCs, could hold important implications for our understanding of the biology of tumor progression and clinical therapy. For example, the biological attributes of the tumor as a whole may be strongly influenced by its subpopulation of CSCs. These cells may

www.sciencemag.org

25 MARCH 2011

1563

Downloaded from www.sciencemag.org on March 11, 2012

Cancer Crusade at 40
drive the self-renewal of the tumor cell population and, in the context of the present discussion, be responsible for the tumors invasiveness and metastatic dissemination. Moreover, the observations that (i) EMT programs can drive carcinoma cells into states that approximate the CSC state and (ii) EMTs can be activated by contextual signals experienced by carcinoma cells lead to the notion that great plasticity is likely to exist between the non-CSCs and CSCs within a tumor. Likewise, these various observations could hold important implications for strategies aimed at reducing metastasis and eradicating minimal residual disease, including dormant tumor cells, micrometastases, and macrometastatic deposits. Conventional therapeutics, which efficiently target actively proliferating cells within the primary tumor, have little impact on quiescent or slowly proliferating cancer cells and, thus, on those cells that reside in many micrometastatic colonies (56, 57). The difficulty of treating this residual disease may be compounded by the physiology of certain target organs. For example, the blood-brain barrier may shield metastases within the brain from drugs delivered through the circulatory system (58). CSCs can erect additional barriers to successful treatment. Rapidly accumulating evidence indicates that CSCs exhibit a heightened resistance to drug-induced death (59). This resistance may stem from two sources: (i) SCs often retreat reversibly from the active growth-and-division cycle into states of quiescence, and (ii) even more important may be the intrinsic drug resistance exhibited by the mesenchymal cancer cells that are the products of the EMT and exhibit traits associated with CSCs (60). Thus, various observations of drug-resistant carcinoma cell subpopulations confirm that the surviving cells often exhibit a more mesenchymal phenotype (61). Without eradicating carcinoma cells that have entered into the mesenchymal/CSC state, the oncologist is confronted with neoplastic subpopulations that are capable of regrowing primary tumors and, in addition, dispatching metastatic travelers to distant organ sites. The most elusive aspect of the invasionmetastasis cascade involves the fates of carcinoma cells after they have disseminated and extravasated into the parenchyma of distant organs. We presume that the self-renewal ability of recently disseminated cells is an essential prerequisite to their successful colonization of such distant tissues, but at present the evidence for this is only indirect. Interestingly, in the absence of a reactive stroma and EMT-inducing signals in such distant tissues, recently disseminated cells that may have arrived in a quasi-mesenchymal/CSC state may lapse back to a fully epithelial state and thereby forfeit the stem-ness that would seem to be essential for their successful founding of metastases. At present, we do not know when and how carcinoma cells acquire the abilities to colonize distant organ sites. Clearly, certain cell-biological programs that have been established and operated in primary tumors may prove advantageous when disseminated carcinoma cells initially confront the microenvironment of distant tissues. Accordingly, we do not know whether the adaptations required for colonization are largely carried by these traveling cells to their sites of metastasis or whether they are instead cobbled together later in sites of dissemination as carcinoma cells struggle to survive and proliferate in foreign, potentially hostile tissue microenvironments. These questions have proven difficult to address because the process of colonization in the great majority of tumors is extraordinarily inefficient. It is unclear whether fully dormant cancer cells are invariably doomed to eventual elimination or whether they may reawaken after months and years and suddenly spawn exuberant tumors. We suspect that only those micrometastases containing proliferating cells are capable of exploring multiple alternative phenotypic states until they stumble on one that enables them to flourish; at present, this notion is deduced from first principles rather than being demonstrated experimentally. The multiplicity of adaptive programs is an issue of great interest: they may be shared by many tumors and at many sites of dissemination. Alternatively, as argued here, these programs represent ad hoc solutions that are dictated by the origins of disseminated cancer cells and the nature of their newfound homes; if so, we may confront myriad distinct adaptive programs that are difficult to rationalize in terms of a common set of underlying biochemical mechanisms. The therapeutic implications of these different scenarios remain to be determined. Still, on a positive note, it is clear that the pace of discovery is rapid and that paths of future exploration are in sight. We now understand much more about the metastatic process than we did even a few short years ago.
References and Notes
1. I. J. Fidler, Nat. Rev. Cancer 3, 453 (2003). 2. J. A. Joyce, J. W. Pollard, Nat. Rev. Cancer 9, 239 (2009). 3. D. Hanahan, R. A. Weinberg, Cell 100, 57 (2000). 4. A. F. Chambers, A. C. Groom, I. C. MacDonald, Nat. Rev. Cancer 2, 563 (2002). 5. J. P. Thiery, J. P. Sleeman, Nat. Rev. Mol. Cell Biol. 7, 131 (2006). 6. L. Foulds, Cancer Res. 14, 327 (1954). 7. D. Bonnet, J. E. Dick, Nat. Med. 3, 730 (1997). 8. M. Al-Hajj, M. S. Wicha, A. Benito-Hernandez, S. J. Morrison, M. F. Clarke, Proc. Natl. Acad. Sci. U.S.A. 100, 3983 (2003). 9. L. E. Ailles, I. L. Weissman, Curr. Opin. Biotechnol. 18, 460 (2007). 10. E. Charafe-Jauffret et al., Cancer Res. 69, 1302 (2009). 11. R. Pang et al., Cell Stem Cell 6, 603 (2010). 12. P. Marcato et al., Stem Cells 29, 32 (2011). 13. E. Quintana et al., Nature 456, 593 (2008). 14. K. Ishizawa et al., Cell Stem Cell 7, 279 (2010). 15. J. P. Thiery, H. Acloque, R. Y. Huang, M. A. Nieto, Cell 139, 871 (2009). 16. S. A. Mani et al., Cell 133, 704 (2008). 17. A. P. Morel et al., PLoS ONE 3, e2888 (2008). 18. A. Gal et al., Oncogene 27, 1218 (2008). 19. J. Yang, R. A. Weinberg, Dev. Cell 14, 818 (2008). 20. E. Lim et al.; kConFab, Nat. Med. 15, 907 (2009). 21. A. Prat et al., Breast Cancer Res. 12, R68 (2010). 22. S. Braun et al., N. Engl. J. Med. 353, 793 (2005). 23. K. Pantel, C. Alix-Panabires, S. Riethdorf, Nat. Rev. Clin. Oncol. 6, 339 (2009). 24. S. Meng et al., Clin. Cancer Res. 10, 8152 (2004). 25. E. Camerer et al., Blood 104, 397 (2004). 26. B. Nieswandt, M. Hafner, B. Echtenacher, D. N. Mnnel, Cancer Res. 59, 1295 (1999). 27. M. Lacroix, Endocr. Relat. Cancer 13, 1033 (2006). 28. K. Pantel, R. H. Brakenhoff, B. Brandt, Nat. Rev. Cancer 8, 329 (2008). 29. M. Mego, S. A. Mani, M. Cristofanilli, Nat. Rev. Clin. Oncol. 7, 693 (2010). 30. S. Ito et al., Int. J. Cancer 93, 212 (2001). 31. C. W. Wong et al., Am. J. Pathol. 161, 749 (2002). 32. E. Sahai, Nat. Rev. Cancer 7, 737 (2007). 33. H. G. Kopp, S. T. Avecilla, A. T. Hooper, S. Rafii, Physiology (Bethesda) 20, 349 (2005). 34. C. Alix-Panabires, S. Riethdorf, K. Pantel, Clin. Cancer Res. 14, 5013 (2008). 35. M. Abdel-Ghany, H. C. Cheng, R. C. Elble, B. U. Pauli, J. Biol. Chem. 276, 25438 (2001). 36. H. Wang et al., J. Cell Biol. 164, 935 (2004). 37. D. M. Brown, E. Ruoslahti, Cancer Cell 5, 365 (2004). 38. K. R. Hess et al., Cancer 106, 1624 (2006). 39. Y. Kang et al., Cancer Cell 3, 537 (2003). 40. A. J. Minn et al., Nature 436, 518 (2005). 41. P. D. Bos et al., Nature 459, 1005 (2009). 42. G. Bloom et al., Am. J. Pathol. 164, 9 (2004). 43. P. J. Campbell et al., Nature 467, 1109 (2010). 44. S. Yachida et al., Nature 467, 1114 (2010). 45. M. Y. Kim et al., Cell 139, 1315 (2009). 46. I. J. Fidler, J. Natl. Cancer Inst. 45, 773 (1970). 47. J. W. Kim et al., Cancer Lett. 213, 203 (2004). 48. K. J. Luzzi et al., Am. J. Pathol. 153, 865 (1998). 49. P. E. Goss, A. F. Chambers, Nat. Rev. Cancer 10, 871 (2010). 50. J. A. Aguirre-Ghiso, Nat. Rev. Cancer 7, 834 (2007). 51. R. N. Kaplan et al., Nature 438, 820 (2005). 52. S. Hiratsuka, A. Watanabe, H. Aburatani, Y. Maru, Nat. Cell Biol. 8, 1369 (2006). 53. S. Y. Kang et al., Proc. Natl. Acad. Sci. U.S.A. 106, 12115 (2009). 54. A. Hirshberg, P. Leibovich, I. Horowitz, A. Buchner, J. Oral Maxillofac. Surg. 51, 1334 (1993). 55. H. F. Dvorak, N. Engl. J. Med. 315, 1650 (1986). 56. G. N. Naumov et al., Breast Cancer Res. Treat. 82, 199 (2003). 57. S. Braun et al., J. Clin. Oncol. 18, 80 (2000). 58. R. J. Weil, D. C. Palmieri, J. L. Bronder, A. M. Stark, P. S. Steeg, Am. J. Pathol. 167, 913 (2005). 59. X. Li et al., J. Natl. Cancer Inst. 100, 672 (2008). 60. E. Buck et al., Mol. Cancer Ther. 6, 532 (2007). 61. C. J. Creighton et al., Proc. Natl. Acad. Sci. U.S.A. 106, 13820 (2009). 62. C.L.C. is supported by the National Health and Medical Research Council of Australia and the Advanced Medical Research Foundation. R.A.W. is supported by the National Cancer Institute, MIT Ludwig Center for Molecular Oncology, and Breast Cancer Research Fund. R.A.W. is a founder and shareholder of Verastem, Inc., a biopharmaceutical company focused on discovering and developing drugs that target cancer stem cells. 10.1126/science.1203543

1564

25 MARCH 2011

VOL 331

SCIENCE

www.sciencemag.org

Downloaded from www.sciencemag.org on March 11, 2012