Background: The adhesion of cancer cells to the endothelial lining of blood vessels, which is important for metastasis, is promoted by the action of interleukin 1 (IL-1) and other cytokines. Purpose: IL-1-producing melanoma cells were used to induce metastases in mice to test whether melanoma metastasis— wherever it occurs— depends on the action of IL-1. Methods: We used the following experimental designs in this study: 1) Male C57BL6J mice were inoculated in the left cardiac ventricle with5×10 4 murine B16 melanoma cells, and no treatment was given (control animals). 2) Mice received an intraperitoneal injection of either saline (control animals) or recombinant human IL-1 receptor antagonist (rHuIL-1Ra) 2 hours before the injection of cancer cells; thereafter, they received an additional injection of saline or rHuIL-1Ra daily for 20 days. 3) Mice received an intravenous injection of either saline or rHuIL-1Ra; 15 minutes later, mice that received saline were given either a second injection of saline (control animals) or an injection of bacterial lipopolysaccharide (LPS) to stimulate host IL-1 production and endothelial cell activation. The mice that received rHuIL-1Ra were also given an injection of LPS at this time. Six hours later, all mice were inoculated with cancer cells, followed by no further treatment. In all experiments, the mice were killed 20 days after the injection of cancer cells, and metastases were counted in multiple organs and bones. Metastasis incidence values (relating to the frequency that a given site was positive for metastasis) and metastasis development index values (relating to the extent of metastasis at a given site) were calculated. A hierarchical cluster analysis was performed to determine whether groups of organs exhibited characteristic changes in their metastasis development index values in response to the three treatments given (i.e., rHuIL-1Ra, LPS, or rHuIL-1Ra plus LPS). Reported P values are two-sided. Results and Conclusions: Treatment with rHuIL-1Ra alone significantly ( P <.05) reduced the occurrence of metastasis in the bone marrow, spleen, liver, lung, pancreas, skeletal muscle, adrenal gland, and heart, indicating that host-andmelanoma-derived IL-1 promoted metastasis in these organs; treatment with rHuIL1Ra had no effect on metastasis in the kidney, testis, brain, skin, and gastrointestinal tract, suggesting that metastasis in these latter organs was IL-1 independent. Treatment with LPS alone significantly ( P <.05) enhanced metastasis in the same organs for which rHuIL-1Ra treatment reduced metastasis, except for the heart and the adrenal gland. Treatment with rHuIL-1Ra 15 minutes before LPS treatment abrogated the LPS-mediated enhancement of metastasis. Two independent organ groups for which IL-1 promoted melanoma metastasis were identified in the cluster analysis. [J Natl Cancer Inst 1997;89:645-51]
Cancer cell adhesion to the microvascular endothelium is promoted by proinflammatory cytokines and plays a key role in metastasis progression ( 1-3 ) . Moreover, the augmentation of metastasis that is mediated by interleukin 1 (IL-1) in the lung and the liver can be prevented by interfering with cytokine action ( 4 , 5 ) or using antibodies to block the cytokineinduced vascular cell adhesion molecule-1 (VCAM-1) expressed on lung endothelium ( 6 ) or the very late antigen-4 (VLA-4) expressed by tumor cells ( 7 ) . Furthermore, IL-1-activated hepatic sinusoidal endothelium also releases tumor proliferation factors that may potentiate metastatic cell growth ( 5 , 8 ) .
Human melanomas and murine experimental melanomas can constitutively produce IL-1 ( 9-12 ) , and they can respond to this and other proinflammatory cytokines ( 13 ) . This ability may represent a powerful tool for blood-borne cancer cells to create their own prometastatic conditions at certain capillary sites to which they are homing. However, melanoma cell lines produce IL-1 heterogeneously ( 9-11 ) , suggesting that IL-1-independent mechanisms of melanoma metastasis may also operate.
The pattern of metastasis exhibited by human melanoma cells is often widespread and unpredictable ( 14 ) . However, on the basis of a cluster analysis ( 15 ) , two predominant patient groupings have been observed. The first group contains patients with visceral metastases that are widely disseminated, and the second group contains patients with central nervous system (CNS) metastases with cerebral spread but metastasis to few other sites. In addition, a highly significant negative correlation between CNS and hepatic metastases has been shown ( 16 ) . In recent years, liver ( 5 ) , lung ( 6 , 17 ) , and bone marrow ( 18 ) have been reported as favorable “ soils” for IL-1-mediated metastasis. No information exists, however, on IL-1-mediated augmentation of metastasis at other common target organs for malignant melanoma, such as the adrenal glands, gonads, brain, skin, or gastrointestinal tract ( 15 ) . It is also not clear whether or not simultaneously produced CNS and reticuloendothelial system (RES) organ metastases share common IL-1-dependent control mechanisms.
The current study was undertaken to determine whether melanoma metastases depend on the action of IL-1 wherever they occur. To this end, we used recombinant human IL-1 receptor antagonist (rHuIL-1Ra) to reduce organ responses to endogenous host and tumor-derived IL-1. In addition, considering the potential pro-metastatic role of creating an inflammatory microenvironment ( 4 , 8 ) , we induced endotoxemia by means of treatment with bacterial lipopolysaccharide (LPS) to increase endogenous IL-1 release and endothelial cell activation prior to cancer cell injection. Then we examined metastasis development under conditions of IL-1 receptor blockade, again using rHuIL-1Ra. We chose the murine B16 melanoma model for this study because of its subpopulation heterogeneity ( 19 ) , which is demonstrated by its ability to produce metastases in both neuroectodermic-and lymphoreticular-derived organs ( 20 ) . We report herein that B16 melanoma cells metastasize to specific organs via IL-1dependent and -independent mechanisms.
Materials and Methods
Tissue culture media, HEPES (4-[2-hydroxyethyl]-1-piperazineethanesulfonic acid) buffer, and LPS ( Escherichia coli endotoxin, serotype 055:B5) were purchased from the Sigma Chemical Co. (St. Louis, MO). Recombinant human IL-1 (rHuIL1), with a specific activity of 10 8 Uprotein, was provided by Biogen, Inc. (Geneva, Switzerland), and rHuIL-1Ra was provided by Upjohn Co. (Kalamazoo, MI). The endotoxin content of preparations of rHuIL-1 and rHuIL-1Ra was less than 20 pgprotein as measured by the Limulus lysate assay (Associates of Cape Cod, Woods Hole, MA).
A highly metastatic B16 melanoma cell line was selected as described previously ( 21 ) . The cells were cultured in Dulbecco'ss modified Eagle medium (DMEM) supplemented with fetal calf serum (10%), penicillin (100 U), and streptomycin (100 g mL). Cultures were maintained at 37 °C in a humidified atmosphere containing 5% CO2.
Left Cardiac Ventricle (LCV) Injection of Melanoma Cells
Male C57BL6J mice 1 (6-8 weeks old) were obtained from IFFA Credo (L'sArbreole, France). The mice (15 per experimental group, with experiments performed in triplicate) were anesthetized with Nembutal (50 mgbody weight) and kept warm at a temperature of 25 °C. The anterior chest wall of each mouse was then shaved and prepared for aseptic surgery by washing it with iodine and 70% ethanol. The ribs over the heart were exposed, and a 30-gauge needle attached to a tuberculin syringe was inserted through the second intercostal space to the left of the sternum and into the left ventricle. When blood entered the tip of the needle,5×10 4 viable melanoma cells in 50 L HEPES-buffered DMEM were injected. The needle was withdrawn slowly, and the muscle and skin were closed with a single suture. After 20 days, the mice were killed by cervical dislocation, and the following organs were removed and the number of melanoticnodules were counted under a dissecting microscope: lungs, liver, pancreas, testes, spleen, adrenals, kidneys, brain, skeletal muscles, skin, gastrointestinal tract, and heart. In addition, the skeletal system of each mouse was completely dissected, and the number of metastatic nodules was recorded for each of the following bones under a dissecting microscope (magnification ×10): spine (cervical, thoracic,
lumbar, and sacral bones), skull (maxilla, mandible, and cranium), thorax (sternum, ribs, and scapula), pelvis (ilium, ischium, and pubis), foreleg (humerus and radius), and hindleg (tibia and femur). On the basis of this inspection, each organ or bone was scored as either containing a metastatic nodule or being free of microscopic tumor. The percentage of organs positive for metastasis was calculated for the total number of mice in each group (metastasis incidence). In addition, a metastasis development index was obtained for each organ and bone segment. To accomplish this, the organs were first fixed in 10% formalin (in phosphate-buffered saline) and embedded in paraffin. Five serially cut, representative, 4-m thick cross-sections were then obtained from each organ and stained with hematoxylin- eosin. A computerized densitometric analysis of digitized, low-magnification (×20) microscopic images from each organ section was used to discriminate, on the basis of relative optical density, between malignant and normal tissue by virtue of differential staining. The surface percentage occupied by metastatic tissue was calculated by use of a previously described morphometric method ( 5 , 21 ) . In contrast, bone segments were studied at the time of cervical dislocation. They were observed directly under a video-camera zoom view (magnification ×10), and the highly contrasted images of bone segments were digitized. Then the same densitometric program was used to discriminate between black, melanotic tissue (metastases) and normal bone tissue and to calculate the percentage of the bone image occupied by metastases. The metastasis development index was determined for each organ or bone segment as follows: the number of recorded metastases per organ or bone segment (maximum, 10) was multiplied by the average surface percentage occupied by metastatic tissue per organ or bone segment (maximum, 100%) and expressed as a relative percentage with respect to a previously defined maximum for each individual organ or bone segment. However, in large organs, such as skin, skeletal muscle, and the gastrointestinal tract, the metastasis development index was based on a metastasis diameter measurement, using an eye reticle calibrated with a stage micrometer as previously described ( 22 ) . To avoid subjective influences on the microscopic study of metastases, the recordings were made in a blinded fashion. Paired and multiple organs were considered as single organ sites, with the incidence values and the metastasis development indices calculated including both or all of the organs, respectively, within an animal.
Mouse Treatment Protocols
In some experiments, mice (n 30) were given a single intraperitoneal injection of either 0.1 mL pyrogen-free saline (15 mice) or 5 mgbody weight rHuIL-1Ra (also in a volume of 0.1 mL) (15 mice) 2 hours before the injection of cancer cells; thereafter, single injections of saline or rHuIL1-Ra were given daily for 20 days. This experimental design was repeated two additional times (i.e., using an additional 60 mice). In other experiments, mice (n
45) first received one intravenous injection of either 0.1 mL pyrogen-free saline (30 mice) or 5 mgbody weight rHuIL-1Ra (15 mice). Fifteen minutes later, the mice in the saline-treated group were divided into two subgroups, one receiving a second injection of 0.1 mL pyrogen-free saline (15 mice) and the other receiving 0.5 mgbody
weight LPS (in a volume of 0.1 mL) (15 mice); the mice in the rHuIL-1Ra-pretreated group also received an injection of 0.5 mgLPS. Six hours after these second injections, cancer cells were injected as described above, and the mice were not treated further. These experiments were performed three times (i.e., using a total of 135 mice).
Two-sided statistical tests were used to analyze the data. Comparisons of metastasis incidence values for treated and control mice were based on Z -statistics. An analysis of variance (ANOVA) was used to make comparisons between average metastasis numbers and metastasis development indices for control and treated mice (Statview 512 software for Macintosh; Abacus Concepts, Inc., Calabasas, CA). The Scheffe F-test was used as a post-ANOVA test. In addition, we applied a hierarchical cluster analysis to determine whether groups of organs with characteristic changes in their melanoma metastasis development indices could be identified under the three treatment conditions (i.e., rHuIL-1Ra, LPS, and rHuIL-1Ra plus LPS). Euclidean distance values were calculated as a measure of average linkages among the studied organs, and a dendrogram was generated to define their aggregations (SPSS software for MS Windows, release 6.0; Professional Statistic, Chicago, IL).
Following LCV injection of B16 melanoma cells, untreated control mice (n 45) became moribund at an average time of 20 ± 2 (mean ± standard deviation) days, and tumors had developed in all animals, with a mean number of 36±4metastases per mouse. The incidence and the degree of development of arterially disseminated metastases in 13 different target organs for these untreated mice is shown in Table 1. Bone (more specifically, bone marrow tissue) and the adrenal glands were the most frequent sites of metastasis. Skeletal muscle, the testes, and the heart also constituted sites of frequent metastasis, with incidence values of around 60%-80%. The frequencies of skin, brain, spleen, and liver metastases were each around 40%-60%. The lung, kidneys, and pancreas formed an organ group of low metastasis incidence (20%40%), and gastrointestinal metastases were the least frequent ones, with a mean incidence of only 13%. Differences in metastasis growth, indicated by the development index parameter, were observed among various organs (Table 1), thus confirming the results of previous studies (20,23,24) on the dissemination of B16 melanoma. B16 melanoma consistently presented the highest metastasis development index in the adrenal glands. High metastasis development indices were also calculated for the lung, skeletal muscle, heart, testes, and gastrointestinal tract. Heart metastases were frequently myocardial and nonpigmented, whereas gastrointestinal tract, testes, and lung metastases were uniformly pigmented and infiltrated the subepithelial lamina propria (intestine) and the interstitial connective tissue among the alveoli (lungs) and the seminiferous tubules (testes). Intermediate metastasis development indices were calculated for the liver, kidneys, brain, and bone marrow. At this last site, two bone subgroups could be defined, resulting from the large variation in metastasis incidence between bone segments: 1) a high incidence metastasis group, involving the maxilla, mandible, spine, ribs, ilium, humerus, scapula, femur, and tibia; and 2) a low incidence metastasis group (having 50% fewer metastases), comprising the radius, pubis, ischium, sternum, and cranium. The organ group with low metastasis development indices included the pancreas, spleen, and skin. At these sites, few and small metastases occurred. In the experiments described below, the multiple saline injections that were given to mice in the groups used as controls did not substantially alter the incidence or the development index parameters relative to the values obtained for control mice (Table 1) not receiving saline injections.
Previously, we reported that rHuIL1Ra (5 mgbody weight), given to mice 2 hours before the intrasplenic injection of B16 melanoma cells and then daily until the animals were killed by cervical dislocation, significantly reduced the development of hepatic metastases ( 5 ) . Here, we report that application of this treatment schedule to mice inoculated in the LCV with B16 melanoma cells significantly (all P <.05) reduced the incidence of metastasis in several organs (11 ± 4 metastases per mouse; P .001 compared with the number of metastases in saline-treated mice), leading to a distinct metastasis pattern (, a). There was a statistically significant (all P <.05) reduction (40%-75%) in metastasis to the bone marrow, spleen, liver, lung, pancreas, adrenals, skeletal muscle, and heart, whereas the differences in metastasis to the kidneys, testes, brain, skin, and gastrointestinal tract were not statistically significant. Moreover, the incidence of metastasis dropped significantly (50%70% reduction; all P <.05) in those organs where metastasis development had decreased, except for the adrenals and skeletal muscle, where only a 20% reduction in incidence was detected.
We ( 8 ) and others ( 4 , 25 ) have reported that the intravenous injection of LPS 6 hours before the injection of melanoma cells significantly enhances the development of hepatic and lung metastases. Using the same working dose of 0.5 mgbody weight LPS given 6 hours before the injection of cancer cells, we found a generalized enhancement of metastasis, which significantly ( P .001) raised the number of metastases per mouse (57 ± 8; n 45 mice) compared with the number in saline-treated mice. However, the enhancement of metastasis was organ specific. As shown in , b, LPS treatment significantly (all P <.05) increased the metastasis development index in the bone marrow, spleen, liver, lungs, pancreas, and skeletal muscles, whereas no significant increase— or even a slight decrease— was observed in the adrenals, heart, kidneys, testes, brain, skin, and gastrointestinal tract. Moreover, the incidence of metastasis increased significantly (all P <.05) in the spleen, liver, and lungs and decreased significantly ( P .04) in the brain, whereas no significant alterations were observed in the rest of the organs studied. In mice receiving intraperitoneal rHuIL-1Ra (5 mgbody weight) 15 minutes before LPS administration, there was a complete abrogation of the above-reported LPS-mediated enhancement of metastasis (32 ± 7 metastases per mouse; P .01 compared with the number of metastases in LPS-treated mice) (, c). The LPS-induced increase in metastasis incidence in the spleen, liver, and lungs was also abrogated (all P <.05). In comparison with LPS-treated mice, no significant variation of these metastasis parameters was observed for the kidneys, testes, brain, skin, and gastrointestinal tract.
A hierarchical cluster analysis was used to define similarities in organ responses to the three treatments. As shown in the dendrogram in , three major organ clusters emerged where euclidean distance values, calculated between organs (within a single cluster) or between organ clusters, defined similarities in metastasis development. Measurement of average linkages between organs within each cluster indicated short-distance associations in most of the cases. In contrast, long-distance values were obtained from intercluster linkages. These features clearly defined two independent clusters entirely comprised of organs with IL-1dependent metastases: the larger cluster being comprised of liver and bone marrow as a nuclear zone, to which heart, spleen, and pancreas were attached with increasing euclidean distance values; the smaller cluster, being composed of lung and skeletal muscle as the nuclear zone, to which the adrenal gland was solely connected. Additionally, there was a third cluster involving only those organs with IL-1-independent metastases. In this case, most of the aggregated organs (skin, kidney, brain, and testis) were primarily interrelated by the same short-distance value, except for gastrointestinal tract that remained secondarily connected.
By blocking the association of IL-1 with its receptors before the arterial dissemination of B16 melanoma cells, both the incidence and the degree of metastasis development were significantly reduced (and occasionally eradicated) in various specific organs of mice, suggesting that IL-1-mediated events may be involved in melanoma metastasis to those organs. This hypothesis is further supported by the finding that stimulating host IL-1 production by LPS treatment before the injection of melanoma cells led to a selective enhancement of metastasis to the same organs, with the exception of the heart and the adrenals, where, although treatment with rHuIL-1Ra reduced metastasis, LPS had no effect. Moreover, rHuIL1Ra abrogated LPS-mediated enhancement of metastasis, indicating that prometastatic conditions in these specific target organs were created by host IL-1 that was released during endotoxemia. In contrast, as was found with the heart and the adrenals, LPS did not affect metastasis development in those organs where metastases were unaffected by rHuIL-1Ra therapy. Thus, rHuIL-1Ra and LPS-induced host IL-1 affected the metastasis of B16 melanoma cells in various specific organs, indicating that these organs share an IL-1-dependent mechanism for melanoma metastasis. It has been reported that the liver ( 5 ) , lungs ( 6 , 17 ) , and bone marrow ( 18 ) are favorable soils for IL-1-mediated metastasis. According to our results, the spleen, pancreas, and skeletal muscle should be added to this list. The adrenal glands and the heart should also be added, although in these cases only tumor-derived IL-1 may be involved, since LPS-induced host IL-1 had no effect. In addition, IL-1-independent metastases also occurred in those organs where neither IL-1 receptor blockade nor LPS-mediated host IL-1 had effects. Thus, another list including the kidneys, testes, brain, skin, and gastrointestinal tract should be considered. In some of these organs, metastases were even slightly inhibited by both Journal of the National Cancer Institute Vol. 89, No. 9, May 7, 1997 IL-1 receptor blockade and LPS-induced host IL-1 effects, suggesting an antimetastatic role for IL-1, as previously reported ( 26 ) . A segregated distribution of major target organs for this murine melanoma was further supported by use of a hierarchical cluster analysis (). Organs with IL-1-dependent metastases were aggregated into two independent clusters composed entirely of organs whose tight interrelationships indicate a similarity in metastasis development in the organs in response to the three treatments. All of the organs with IL-1-independent metastases were similarly linked by short-distance values, defining again a homogeneous response in metastasis development of the organs. Finally, long euclidean distance values, defining the linkage of the organ cluster with IL-1-independent metastases to the two clusters with IL-1-dependent metastases, describe the different responses of the various organs to the three treatments in terms of metastasis development. There may be many mechanisms by which IL-1 can promote metastasis progression, without discounting possible IL-1-dependent antimetastatic effects (12,26,27) . First, melanoma cells may take advantage of an IL-1-mediated increase in the adhesiveness of the vascular endothelium. This mechanism has recently been observed in the lung ( 6 , 7 ) and the liver (8,27; our unpublished results), where increased VCAM-1 expression in the endothelium in response to host or exogenous proinflammatory cytokines resulted in an enhancement of VLA-4mediated melanoma cell adherence and metastasis. Very recently, we have determined metastasis development indices for several organs in LPS-treated mice that had additionally received anti-mouse VCAM-1 antibodies. In comparison with non-antibody-injected LPS-treated mice, there was significant (all P <.05) inhibition of metastasis in the lungs (56%), liver (42%), bone marrow (77%), adrenal glands (50%), skeletal muscle (73%), and pancreas (100%) (our unpublished results). Moreover, in the liver, increased mannose receptor expression on sinusoidal endothelium activated by exogenous or host IL-1 also increased melanoma cell adherence and metastasis ( 8 ,2 8 ). Second, host IL-1 may promote the invasive capacity of metastatic cells through a mechanism that involves both metalloproteinase ( 29 ) and chemokine ( 30 ) production by tumor cells or stromal cells. Third, host IL-1 may induce the release of paracrine andautocrine metastatic-melanoma cell growth factors, such as IL-6 ( 31 ) and IL-8 ( 29 , 32 ) , whose expression levels may be influenced by specific organ microenvironments and may correlate with the metastatic potential of human melanoma cells ( 33 ) . Fourth, host IL-1 may activate intrametastatic neoangiogenesis, since rHuIL-1Ra given to rats significantly reduced inflammation-associated angiogenesis ( 34 , 35 ) .
It has been reported that certain human and murine melanoma cell lines produce IL-1 ( 9-12 ) . In a manner similar to that of host IL-1, melanoma cell-derived IL-1 may also have prometastatic effects. This mechanism may have been operating in the melanoma metastases in the adrenals and the heart observed in our work. Not surprisingly, pretreatment of B16 melanoma cells in vitro with the IL-1-converting-enzyme inhibitor (ICEi), which suppresses melanoma cell IL-1 secretion, also significantly ( P <.01) reduced the development of metastasis by LCV-injected cells in those organs in which metastases were affected by rHuIL-1Ra treatment, including the adrenals and the heart. In contrast, no statistically significant variations were detected in the occurrence of metastases in organs whose metastases are unaffected by rHuIL-1Ra (our unpublished results). While it still remains unclear whether B16 melanoma cells produce IL-1 homogeneously or heterogeneously, they may make use of their constitutive IL-1 expression to metastasize to certain organs. Without discounting a possible autocrine effect of melanoma-derived IL-1, which enhances metastasis competence, prometastatic factors activated in IL-1-responsive target organs may be involved. This hypothesis is supported by the observation that melanoma cell-conditioned medium induced IL-1-mediated expression of endothelial cell adhesion molecules and increased reciprocal adhesiveness of melanoma and endothelial cells in vitro ( 10 ) . The same conditioned medium also substantially increased progelatinase-A secretion by isolated hepatic perisinusoidal fibroblasts and the chemotaxis of B16 melanoma cells in vitro ( 36 ) . Moreover, the B16 melanoma cell-contact-induced increase in reactive oxygen species production, mannose receptor expression, and melanoma proliferation factor secretion by hepatic sinusoidal endothelial cells is juxtacrine IL-1-mediated ( 37 ) . Thus, IL-1 may confer upon B16 melanoma cells the ability to create prometastatic microenvironments for themselves, and this mechanism may be operative in the organ group where metastases were reduced by rHuIL1Ra. However, the murine melanoma model used in this study also revealed the existence of murine metastases produced by an IL-1-independent mechanism, although it is not clear whether these metastases were produced by IL-1-producer or -nonproducer melanoma cells. The details of this hypothetical IL-1 independent mechanism are also not clear. In contrast, melanoma patients having CNS metastases usually do not present metastases elsewhere ( 15 ) . In addition, in melanoma patients there is a highly negative correlation between CNS and hepatic metastases ( 16 ) . CNS metastases may derive from non-IL-1-producer melanoma cells, since, if they produce IL-1, metastases should also occur in organs in which IL-1-dependent metastases occur, and this is not the case.
Both the inherent capacity of B16 melanoma cells to produce IL-1 and the release of prometastatic factors by specific organ microenvironments in response to IL-1 may be crucial to the occurrence of metastasis among organs. Within the classical context of the “ Seed and Soil” theory of metastasis postulated by Paget ( 38 ) , the capacity of melanoma cells to produce IL-1 andto respond to IL-1-dependent factors may contribute to “ seed organophilia” ; the capacity of a given organ microenvironment to create prometastatic conditions in response to IL-1 may contribute to “ soil suitability” for metastasis. Since both aspects seem to be operative in the B16 melanoma model, as shown by its IL-1-dependent and -independent metastases coexisting in the same mice, melanoma cell subpopulations using different mechanisms of metastasis probably participate in the metastatic process. This would further support the concept of subpopulation heterogeneity with regard to the metastatic phenotype ( 39 ) .