Lymphatic invasion and accumulation of continuous collagen bundles around tumor cells are associated with poor melanoma prognosis, but the underlying mechanisms and molecular determinants have remained unclear. We show here that a copy-number gain or overexpression of the membrane-type matrix metalloproteinase MMP16 (MT3-MMP) is associated with poor clinical outcome, collagen bundle assembly around tumor cell nests, and lymphatic invasion. In cultured WM852 melanoma cells derived from human melanoma metastasis, silencing of MMP16 resulted in cell-surface accumulation of the MMP16 substrate MMP14 (MT1-MMP) as well as L1CAM cell adhesion molecule, identified here as a novel MMP16 substrate. When limiting the activities of these trans-membrane protein substrates toward pericellular collagen degradation, cell junction disassembly, and blood endothelial transmigration, MMP16 supported nodular-type growth of adhesive collagen-surrounded melanoma cell nests, coincidentally steering cell collectives into lymphatic vessels. These results uncover a novel mechanism in melanoma pathogenesis, whereby restricted collagen infiltration and limited mesenchymal invasion are unexpectedly associated with the properties of the most aggressive tumors, revealing MMP16 as a putative indicator of adverse melanoma prognosis. Cancer Res; 75(10); 2083–94. ©2015 AACR.

Melanoma is the most aggressive skin cancer in humans, occurring at a younger age than other common malignancies (1). The prognosis of metastatic melanoma remains unfavorable due to melanoma plasticity and lack of effective treatments. Currently, the most accurate predictors of melanoma progression are tumor thickness, ulceration, mitotic index, and sentinel lymph node (SLN) metastasis (2). Patients with >1-mm-thick tumors are directed for SLN biopsy (2). However, small melanomas can also develop lethal metastases. Moreover, SLN biopsy procedure is associated with increased morbidity, and the impact on survival is questionable (3). Therefore, a better understanding of the mechanisms and early markers of melanoma progression is essential. Previously, lymphatic vessel density, lymphatic invasion, and extracellular matrix (ECM) assembly in parallel bundles and networks have been implicated as additional parameters correlating with poor patient survival (4–7). Although multiple ECM-modifying enzymes, adhesion receptors, and signaling pathways regulate these tumor properties, the specific underlying mechanisms and molecular determinants have remained elusive.

Membrane-type matrix metalloproteinases (MT-MMP) are pericellular ECM-degrading proteases with substrate specificities ranging from fibrillar collagens and fibrin to basement membrane (BM) components (8, 9). They also cleave cell-surface–associated molecules and activate secreted MMPs (10). MMP14/MT1-MMP, commonly induced in invasive cancers including melanoma, is the most widely expressed of six human MT-MMPs (11). Interestingly, overexpression of the related protease MMP16/MT3-MMP has been reported in the most aggressive nodular melanoma subtype, which grows as adhesive nodules within dermis, rather than spreading radially, and metastasizes early/aggressively into lymph nodes (12, 13).

Unlike MMP14, MMP16 is inefficient type I collagenase (14). Although, for example, collagen III, laminin, fibrin, CD44, and LRP1 are cleaved by both MT-MMPs in vitro, potential in vivo functions of MMP16 in tumor cells have remained unclear (15). We reported that MMP16 functions as a matrix composition-dependent regulator of melanoma cell invasion and 3D growth (16). While promoting fibrin invasion, MMP16 also associates with MMP14 in hetero-oligomers cleaving and posttranslationally suppressing this proinvasive protease (16). These observations suggest an MMP16 function in melanoma pathogenesis. Here, we assessed MMP16 expression and copy number in melanoma tumors and cell lines. Intrigued by significant induction and a novel association with aggressive melanoma properties, we explored MMP16 functions using cultured cells and WM852 melanoma xenografts. The results revealed a functional contribution of MMP16 to the aggressive melanoma properties.

Cell culture

WM852 and WM165 cells from melanoma metastasis (Wistar Institute, Philadephia, PA), as well as Bowes melanoma cells and COS1 cells (ATCC) were cultivated as described (16, 17). Human foreskin lymphatic endothelial cells (LEC; PromoCell), and blood endothelial cells (BEC; dermal HDBECs and umbilical vein endothelial cells HUVECs, PromoCell) were cultured in Endothelial Cell Growth Medium MV containing gentamicin (50 μg/mL).

Clinical melanoma samples

Paraffin-embedded human melanomas were obtained from the Skin and Allergy Hospital of Helsinki University Hospital (Helsinki, Finland). Benign nevi were kindly provided by Dr. Paula Kujala, Fimlab, Tampere University Hospital, Tampere, Finland (see Supplementary Data).

RNAi, cDNAs, and qPCR

siRNAs were transfected using Lipofectamine 2000 (Invitrogen). shRNAs were transduced lentivirally (18). For RNAi and cDNA constructs, see Supplementary Data. Total mRNA was extracted from punch biopsies of paraffin-embedded human melanomas and benign nevi using High Pure RNA Paraffin Kit (Roche), and from cultured cells using RNeasy Mini Kit (Qiagen), followed by reverse transcription with iScript cDNA Synthesis Kit (Bio-Rad) and qPCR (see Supplementary Data).

Cell adhesion, transendothelial migration, and morphology assessment

WM852 cells prestained with green fluorescent Vybrant CFDA SE Cell Tracer (Invitrogen) were seeded on LEC or BEC monolayers. After 3 hours, adhered cells/well were quantified using Cellomics View software (Thermo Scientific). For transmigration, prestained WM852 cells were seeded on LEC or HUVEC monolayers in cell culture inserts (BD Falcon). Human recombinant L1CAM (L1CAM-Fc, R&D Systems) was added where indicated. Ten percent FCS in the lower chamber served as a chemoattractant. After 12 hours, cultures were fixed, and transmigrated cells quantified. For cell morphology assessment, see Supplementary Data.

3D melanoma-endothelial spheroid coculture

LEC and HDBEC spheroids were formed by culturing 4,000 cells/well in agarose-coated round-bottom 96-well plates o/n. Ten spheroids were mixed with 5,000 melanoma cells and 50 μL fibrin (4.5 mg/mL; ref. 16). After 3 to 5 days, the cultures were fixed, subjected to immunofluorescence, and imaged using Leica SP5 confocal microscope. Intravasated melanoma cells were quantified using Anduril workflow (see Supplementary Data; ref. 19).

Antibody array, immunoblotting, immunoprecipitation, and gelatin zymography

Serum-free conditioned media (CM) from WM852 cells transfected with siCtrl or siMMP16 were collected for 48 hours and subjected to Human Soluble Receptor Antibody Array, Non-Hematopoietic Panel (R&D Systems). Selected proteins were assessed by gelatin zymography, immunoblotting and immunoprecipitation using HA-agarose (Sigma; see Supplementary Data; refs. 16, 18). The dot/band intensities were quantified using ImageJ.

Kaplan–Meier survival analysis

The clinical and copy-number data for 192 skin cutaneous melanoma samples were downloaded from The Cancer Genome Atlas (TCGA) and analyzed with Anduril workflow (see Supplementary Data).

In vivo xenografts

Experiments were approved by the State Provincial Office of Southern Finland. In two independent experiments, shScrambled or shMMP16 WM852 cell pools (1.3 × 106 or 2.0 × 106 cells/mouse) were implanted into abdominal subcutis of 7-week-old ICR-SCID female mice and followed for 7 to 9 weeks.

IHC and immunofluorescence

Mouse xenografts were fixed, dehydrated, and embedded in paraffin, or in OCT Compound (Tissue-Tek) for frozen sections. For IHC, immunofluorescence, and primary antibodies, see Supplementary Data.

Statistical analysis

Numerical values represent mean±SD unless stated otherwise. Statistical significance was determined using the Mann–Whitney U test.

MMP16 overexpression and gain are associated with poor melanoma outcome

To shed light on the possible role of MMP16 in melanoma pathogenesis, we analyzed the corresponding mRNA and DNA copy-number alterations using Oncomine (www.oncomine.org) and TCGA (20). MMP16 mRNA expression was significantly higher in melanoma cell lines compared with 17 other types of cancer cell lines (Fig. 1A; www.oncomine.org; ref. 21). MMP14, encoding the related and widely cancer-associated transmembrane protease and, according to some of the datasets, also MMP1, MMP2, MMP8, and MMP17 out of the 23 human MMPs were likewise highly expressed in melanoma cell lines (Supplementary Fig. S1A and Supplementary Table S1). Outlier analysis results by Oncomine showed significant MMP16 overexpression in part of clinical human melanomas (Supplementary Table S2; P < 0.0001 in 5/6 mRNA datasets with n ≥ 20; ref. 22).

Figure 1.

MMP16 overexpression and copy-number gain are associated with poor melanoma outcome. A, MMP16 mRNA is overexpressed in melanoma and CNS cancer cell lines (Oncomine.org; ref. 48). *, P = 3.78 × 10−16; **, P = 2.83 × 10−16. The number of cell lines is indicated in brackets. B and MMP16 expression versus overall patient survival (Oncomine.org). Primary melanoma (B), n = 81 (49), and melanoma metastases (C), n = 44 (50); outlier line (90%) marks significant MMP16 overexpression (P = 0.0001). Red square highlights MMP16 expression in poor outcome group. D, MMP14 and MMP16 DNA copy-number variations (Oncomine.org; TCGA). E, Kaplan–Meier survival curve visualizes the correlation of MMP16 gain with patient survival (TCGA), P = 0.008.

Figure 1.

MMP16 overexpression and copy-number gain are associated with poor melanoma outcome. A, MMP16 mRNA is overexpressed in melanoma and CNS cancer cell lines (Oncomine.org; ref. 48). *, P = 3.78 × 10−16; **, P = 2.83 × 10−16. The number of cell lines is indicated in brackets. B and MMP16 expression versus overall patient survival (Oncomine.org). Primary melanoma (B), n = 81 (49), and melanoma metastases (C), n = 44 (50); outlier line (90%) marks significant MMP16 overexpression (P = 0.0001). Red square highlights MMP16 expression in poor outcome group. D, MMP14 and MMP16 DNA copy-number variations (Oncomine.org; TCGA). E, Kaplan–Meier survival curve visualizes the correlation of MMP16 gain with patient survival (TCGA), P = 0.008.

Close modal

To consider the clinical significance of the MMP16 induction, we compared tumor mRNA expression with overall patient survival. Unexpectedly, melanomas with MMP16 overexpression segregated into poor outcome groups (Fig. 1B; primary and Fig. 1C; metastasis), whereas high MMP14 expression across the biopsies did not vary according to survival (Supplementary Fig. S1B and S1C). Moreover, MMP16 copy number was increased in a group of melanomas (Fig. 1D and Supplementary Table S2). Significantly, the MMP16 gain was associated with poorer survival of melanoma patients as compared with the no-gain group (Fig. 1E; P = 0.008), consistent with a function of MMP16 in melanoma progression.

MMP16 overexpression is associated with assembly of continuous collagen bundles and lymphatic invasion in human melanoma

Our previous study demonstrated the association with and cleavage of the major cell-surface collagenase MMP14 by MMP16 in melanoma cells, suggesting an intimate interplay between these proteases (16). To dissect their potential functions in clinical melanoma, we compared MMP14 and MMP16 expression with histopathology of cutaneous melanoma biopsies (Breslow > 1 mm; n = 19; 18 primary tumors, one skin metastasis). MMP14 was >1.7-fold overexpressed in eight and MMP16 in six melanoma biopsies compared with mean expression in benign nevi (Fig. 2A and Supplementary Table S3; nevi n = 7). Remarkably, all five tumors with high MMP14 and MMP16 coexpression (MMP14high/MMP16high), and only one of nine other tumors available for representative IHC, displayed invasion of S100B-positive melanoma cells into podoplanin-positive lymphatic vessels (Fig. 2A–C and Supplementary Fig. S2 and Supplementary Tables S3 and S4; no correlation with lymphatic or blood vessel densities). Two MMP14high/MMP16high patients also had skin metastases, representing local lymphatic spread (23; Supplementary Table S3). Moreover, MMP14high/MMP16high tumors displayed a distinct pattern of adhesive cell nests confined by continuous collagen networks, which further assembled into tumor nodules surrounded by collagenous ECM (Fig. 2D; Supplementary Fig. S2). Lymphatic and blood vessels were enriched within these ECM structures (Fig. 2E and F). Melanoma cells remained aggregated inside lymphatic vessels (Fig. 2B). In contrast, no lymphatic vessel invasion (LVI) was observed in tumors with high MMP14 and low MMP16 (MMP14high/MMP16low), where confined cell nests were limited to areas close to epidermis (Fig. 2C and D). In deeper dermis, MMP14high/MMP16low melanoma cells were instead intermingled with fragmented collagen, suggestive of pericellular collagenolytic and tissue-infiltrative activities (Fig. 2D and Supplementary Table S3). Therefore, MMP16 overexpression in the MMP14-expressing melanomas may contribute to the collagen assembly into continuous bundles and networks, cell growth as aggregates, and LVI, which are the properties associated with aggressive melanoma dissemination.

Figure 2.

Human melanomas with MMP16 overexpression exhibit LVI and growth in collagen-surrounded nests. A, MMP14 and MMP16 mRNAs in human benign nevi and melanoma specimens were assessed by qPCR. The mean expression in benign nevi was set to 1. B and C, podoplanin (brown) and S100B (brown) IHC of the indicated melanoma tissue sections. Arrows, tumor cells inside lymphatic vessels. D, Herovici staining (red) visualizes the collagen-confined, nest-like morphology of representative MMP14high/MMP16high tumor as opposed to collagen-infiltrative MMP14high/MMP16low tumor. E, podoplanin IHC visualizes lymphatic vessels enriched in the ECM structures of MMP14high/MMP16high tumors. F, CD34-positive blood vessels in the tumors.

Figure 2.

Human melanomas with MMP16 overexpression exhibit LVI and growth in collagen-surrounded nests. A, MMP14 and MMP16 mRNAs in human benign nevi and melanoma specimens were assessed by qPCR. The mean expression in benign nevi was set to 1. B and C, podoplanin (brown) and S100B (brown) IHC of the indicated melanoma tissue sections. Arrows, tumor cells inside lymphatic vessels. D, Herovici staining (red) visualizes the collagen-confined, nest-like morphology of representative MMP14high/MMP16high tumor as opposed to collagen-infiltrative MMP14high/MMP16low tumor. E, podoplanin IHC visualizes lymphatic vessels enriched in the ECM structures of MMP14high/MMP16high tumors. F, CD34-positive blood vessels in the tumors.

Close modal

Silencing of MMP16 switches WM852 melanoma cells from lymphatic to blood vessel invasion

To define functional contribution of MMP16 in melanoma growth and vascular invasion, lentiviral shRNAs (shMMP16) were used for MMP16 silencing in nodular melanoma metastasis-derived WM852 cells followed by subcutaneous cell implantation into immunocompromised mice (Supplementary Fig. S3A and S3B). The control shScr-transduced WM852 cells express high levels of MMP14 and MMP16 (16). After approximately 8 weeks, shMMP16 tumors expressed 70% less MMP16 than shScr tumors (Fig. 3A). ShScr tumors grew relatively slowly, and MMP16 silencing increased the growth rate (Fig. 3B and C and Supplementary Fig. S3C). In both tumor groups, intratumoral density of CD31-positive blood vessels was >40-fold higher than that of LYVE-1–positive lymphatic vessels (Supplementary Fig. S4A–S4C). Importantly, shScr tumors displayed prominent LVI by cell collectives, which was decreased by >80% after MMP16 silencing (Fig. 3D and E). In contrast, modest blood vessel invasion (BVI) in shScr tumors was increased by approximately 10-fold after MMP16 knockdown (Fig. 3D and E).

Figure 3.

MMP16-silencing shifts vascular invasion from lymphatic to blood vessels. A and B, MMP16 mRNA expression (A) and final weight of control (shScr; B) and MMP16-depleted (shMMP16) WM852 xenografts (pooled from two experiments). C, tumor growth in independent experiments. D, the percentage of intratumoral and peritumoral lymphatic and blood vessels containing intravasated tumor cells. C and D, error bars, ± SE. E, xenograft sections stained for mouse Lyve-1 (red) and CD31 (red). Arrowheads and arrows, tumor cells inside lymphatic and blood vessels, respectively. *, P < 0.002; **, P < 0.05.

Figure 3.

MMP16-silencing shifts vascular invasion from lymphatic to blood vessels. A and B, MMP16 mRNA expression (A) and final weight of control (shScr; B) and MMP16-depleted (shMMP16) WM852 xenografts (pooled from two experiments). C, tumor growth in independent experiments. D, the percentage of intratumoral and peritumoral lymphatic and blood vessels containing intravasated tumor cells. C and D, error bars, ± SE. E, xenograft sections stained for mouse Lyve-1 (red) and CD31 (red). Arrowheads and arrows, tumor cells inside lymphatic and blood vessels, respectively. *, P < 0.002; **, P < 0.05.

Close modal

Silencing of MMP16 increases pericellular collagen degradation

Because MMP16 could regulate melanoma growth and vascular invasion by altering tumor ECM directly and/or via posttranslational MMP14 downregulation, we compared the ECM composition of shScr with shMMP16 tumors (16). BM collagen IV thickness and mouse collagen I remained unaltered (Fig. 4A and Supplementary Fig. S5). In contrast, prominent tumor cell–derived human collagen I fibers displayed in shScr tumors were markedly decreased, and overall intratumoral collagenous ECM was diminished by approximately 80% in shMMP16 tumors (Fig. 4B and C). Collagen mRNA was not decreased (Fig. 4D), suggesting that MMP16 silencing enhanced pericellular collagen degradation, consistently with the reported increase in cell-surface MMP14 in shMMP16 cells (16). Fibrin accumulation was instead enhanced (Fig. 4E), consistently with fibrinolytic MMP16 activity (16, 24).

Figure 4.

MMP16 silencing increases collagen degradation and fibrin accumulation. A and B, IHC of frozen sections visualize mouse collagen IV (red; A) and human collagen I (red; B) coupled with mouse CD31 (green) in shScr and shMMP16 WM852 xenografts. BM collagen IV thickness and overall collagen I intensity are shown below the micrographs. C, Herovici staining (red) visualizes the reduction of collagenous ECM after MMP16 silencing. Collagen coverage/microscope field was calculated. D, human collagen 1A1 and mouse collagen 1A1 and 3A1 mRNAs in the xenografts. Mean expression in shScr tumors was set to 1. E, fibrin and CD31 IHC visualize increased interstitial fibrin and fibrin around intravasated tumor cells after MMP16 silencing. Fibrin coverage/microscope field; *, P ≤ 0.021; **, P = 0.055.

Figure 4.

MMP16 silencing increases collagen degradation and fibrin accumulation. A and B, IHC of frozen sections visualize mouse collagen IV (red; A) and human collagen I (red; B) coupled with mouse CD31 (green) in shScr and shMMP16 WM852 xenografts. BM collagen IV thickness and overall collagen I intensity are shown below the micrographs. C, Herovici staining (red) visualizes the reduction of collagenous ECM after MMP16 silencing. Collagen coverage/microscope field was calculated. D, human collagen 1A1 and mouse collagen 1A1 and 3A1 mRNAs in the xenografts. Mean expression in shScr tumors was set to 1. E, fibrin and CD31 IHC visualize increased interstitial fibrin and fibrin around intravasated tumor cells after MMP16 silencing. Fibrin coverage/microscope field; *, P ≤ 0.021; **, P = 0.055.

Close modal

MMP16 promotes melanoma cell–cell adhesion

Considering the association of collective invasion with LVI (25) and the collective cell patterns in MMP16-expressing human melanoma and xenograft tumors, we next assessed the effects of MMP16 on WM852 cell growth and migration, and their relation to cell–cell adhesion in culture. As reported, shScr cells displayed limited invasion in 3D collagen, which was increased by MMP16 silencing (S6A and S6B; ref. 16), and MMP16 overexpression reduced collagen invasion of MMP14-expressing Bowes melanoma cells (Supplementary Fig. S6C). The rates of shScr and shMMP16 cell wound migration and proliferation remained comparable in 2D (Supplementary Fig. S6D and S6E). Notably, shScr cells grew as collective colonies, whereas MMP16 silencing induced a switch to single-cell phenotype, previously associated with BVI (Fig. 5A; ref. 25). MMP16 siRNAs decreased ZO-1, N-cadherin, and EphA2-containing cell junctions, whereas MMP14 silencing strengthened the junctions (Supplementary Fig. S6F). Consistently, MMP16 overexpression increased cell–cell contacts in Bowes cells (Supplementary Fig. S6G).

Figure 5.

MMP16 silencing induces melanoma cell junction disassembly and BEC transmigration but inhibits intravasation into LEC spheroids. A, light micrographs visualize adhesive morphology of WM852 shScr cells and single-cell phenotype of shMMP16 cells overlaid by indicated 3D matrices. B, WM852 cell adhesion to LEC or BEC monolayers in 3 hours. C, WM852 cell transmigration across LECs and BECs in 12 hours. D, relative BEC or LEC spheroid intravasation by WM852 cells. The mean number of intravasated shScr cells was set to 100. E and F, representative confocal images of WM852 cells (E; green) and WM165 cells (F; green) transfected to overexpress MMP16 where indicated, and cocultured with LEC spheroids (red) in fibrin for 5 days. Arrows, melanoma cells inside the spheroids. MMP16-r1 and MMP16-r2 contain silent mutations in shMMP16 sequence. Relative number of intravasated melanoma cells/spheroid is presented on the right. The mean number of shScr cells was set to 100. *, P = 0.001; **, P < 0.04.

Figure 5.

MMP16 silencing induces melanoma cell junction disassembly and BEC transmigration but inhibits intravasation into LEC spheroids. A, light micrographs visualize adhesive morphology of WM852 shScr cells and single-cell phenotype of shMMP16 cells overlaid by indicated 3D matrices. B, WM852 cell adhesion to LEC or BEC monolayers in 3 hours. C, WM852 cell transmigration across LECs and BECs in 12 hours. D, relative BEC or LEC spheroid intravasation by WM852 cells. The mean number of intravasated shScr cells was set to 100. E and F, representative confocal images of WM852 cells (E; green) and WM165 cells (F; green) transfected to overexpress MMP16 where indicated, and cocultured with LEC spheroids (red) in fibrin for 5 days. Arrows, melanoma cells inside the spheroids. MMP16-r1 and MMP16-r2 contain silent mutations in shMMP16 sequence. Relative number of intravasated melanoma cells/spheroid is presented on the right. The mean number of shScr cells was set to 100. *, P = 0.001; **, P < 0.04.

Close modal

MMP16 exhibits ECM environment–dependent and -independent regulatory activities toward blood endothelial transmigration and lymphatic intravasation

To further investigate the MMP16-dependent mechanisms of vascular invasion, melanoma-endothelial cell interactions were assessed in 2D and 3D. When shScr cells were plated as a suspension atop endothelial cell monolayers, they adhered efficiently to BECs, but poorly to LECs (Fig. 5B). MMP16 silencing increased adhesion to and transmigration across BECs, consistently with the increased BVI of shMMP16 cells in vivo, whereas adhesion to and transmigration across LECs remained modest in both shScr and shMMP16 cells (Fig. 5B and C). The approximately 16-fold more efficient shScr cell transmigration across BECs than LECs was different from the more prominent LVI than BVI in vivo, most likely due to poor adhesion to LECs in the absence of 3D tissue microenvironment. Indeed, when WM852 and WM165 melanoma cells were embedded with preformed endothelial spheroids in 3D fibrin, shScr cells invaded more readily into LEC than BEC spheroids (Fig. 5D–F and and Supplementary Fig. S7A). MMP16 silencing inhibited LEC spheroid intravasation, plausibly via effects related to fibrin-invasive MMP16 activity and stronger chemoattraction by LECs than BECs (Fig. 5D–F and Supplementary Fig. S7A; refs. 16, 26). Consistently, inefficient LEC spheroid intravasation of shMMP16 cells, and of Bowes cells devoid of endogenous MMP16, was markedly increased after MMP16 overexpression (Fig. 5E–F and Supplementary Fig. S7B–C). Altogether, these results suggest that MMP16 impedes tumor–blood endothelial cell adhesion and BEC transmigration, whereas LEC interactions and transmigration are supported by 3D ECM environment and MMP16.

L1CAM is a novel MMP16 substrate involved in transendothelial migration

To search for melanoma cell–surface substrates of MMP16 involved in vascular invasion, we assessed the release of 119 proteins to the CM of control and MMP16-depleted WM852 cells. Out of the strongly detected proteins, MMP2 and EMMPRIN were decreased most efficiently by MMP16 silencing (Fig. 6A; Supplementary Fig. S8A and Supplementary Table S5). Soluble form of Neural cell adhesion molecule L1 (L1CAM), a receptor previously linked to transendothelial migration, was reduced by approximately 50% (Fig. 6A; ref. 27). To test whether MMP16 releases soluble L1CAM by proteolytic shedding, COS1 cells devoid of endogenous MMPs were transfected to express L1CAM alone or with MMP16. Besides full-length 220- to 240-kDa L1CAM in cell lysates, soluble ectodomain of approximately 200 kDa in size was detected in CM (Fig. 6B and C). MMP16 increased the ectodomain shedding, coincident with the detection of approximately 30 kDa C-terminal L1CAM fragment and L1CAM coimmunoprecipitation with MMP16 (Fig. 6B and C). As expected, the MMP inhibitor GM6001 inhibited this MMP16-mediated shedding (Fig. 6B and C).

Figure 6.

MMP16-mediated shedding of L1CAM is involved in blood endothelial transmigration. A, Human Soluble Receptor Antibody Array analysis of WM852 cells transfected with control (siCtrl) and MMP16 (siMMP16) siRNAs. Equal dot intensities in siCtrl and siMMP16 cell CM were set to zero. Boxes highlight the indicated proteins with >20% decrease after MMP16 silencing in the shown chemiluminescence detection. B and C, L1CAM was expressed alone or together with HA-tagged MMP16 in COS1 cells, followed by immunoprecipitation using HA-agarose and immunoblotting. MMP inhibitor GM6001 (10 μmol/L) was added o/n where indicated. D and E, CM of WM852 cells transfected with indicated siRNAs or transduced with shMMP16 (D) and CM of shMMP16 cells transfected with MMP16-r1 and MMP16-r2 rescue plasmids were subjected to immunoblotting (E). F, relative transmigration of WM852 cells across LEC or BEC monolayers in 12 hours. G, relative LEC spheroid intravasation by WM852 cells. The mean number of intravasated shScr cells was set to 100. H, full-length L1CAM was detected from xenograft tumor lysates by immunoblotting. L1CAM band intensities normalized with tubulin are indicated below each lane. I, relative expression of L1CAM protein and mRNA in the xenografts. J and K, L1CAM IHC coverage/microscopic field in human melanoma (J) and xenograft tissue sections (K). L and M, IHC visualizes abundant cell-surface L1CAM in human tumors with low MMP16 (L; see Supplementary Fig. S9) and shMMP16 xenografts (M). *, P < 0.02.

Figure 6.

MMP16-mediated shedding of L1CAM is involved in blood endothelial transmigration. A, Human Soluble Receptor Antibody Array analysis of WM852 cells transfected with control (siCtrl) and MMP16 (siMMP16) siRNAs. Equal dot intensities in siCtrl and siMMP16 cell CM were set to zero. Boxes highlight the indicated proteins with >20% decrease after MMP16 silencing in the shown chemiluminescence detection. B and C, L1CAM was expressed alone or together with HA-tagged MMP16 in COS1 cells, followed by immunoprecipitation using HA-agarose and immunoblotting. MMP inhibitor GM6001 (10 μmol/L) was added o/n where indicated. D and E, CM of WM852 cells transfected with indicated siRNAs or transduced with shMMP16 (D) and CM of shMMP16 cells transfected with MMP16-r1 and MMP16-r2 rescue plasmids were subjected to immunoblotting (E). F, relative transmigration of WM852 cells across LEC or BEC monolayers in 12 hours. G, relative LEC spheroid intravasation by WM852 cells. The mean number of intravasated shScr cells was set to 100. H, full-length L1CAM was detected from xenograft tumor lysates by immunoblotting. L1CAM band intensities normalized with tubulin are indicated below each lane. I, relative expression of L1CAM protein and mRNA in the xenografts. J and K, L1CAM IHC coverage/microscopic field in human melanoma (J) and xenograft tissue sections (K). L and M, IHC visualizes abundant cell-surface L1CAM in human tumors with low MMP16 (L; see Supplementary Fig. S9) and shMMP16 xenografts (M). *, P < 0.02.

Close modal

WM852 and WM165 shScr cell CM contained endogenous L1CAM ectodomain (Fig. 6D and and Supplementary Fig. S8B and S8C). Unexpectedly, silencing of ADAM10, the previously identified L1CAM sheddase, neither decreased the ectodomain nor altered the full-length L1CAM (Fig. 6D and Supplementary Fig. S8B–S8D; ref. 28). In contrast, the ectodomain was decreased by MMP16 silencing (Fig. 6D). Concomitantly, full-length L1CAM was increased in WM852 cells (Supplementary Fig. S8D and S8E). Moreover, overexpression of MMP16 increased L1CAM shedding in WM852 and WM165 shMMP16 cells (Fig. 6E and Supplementary Fig. S8F). Silencing of MMP14 instead increased the 200-kDa ectodomain, coincidentally decreasing an additional approximately 160-kDa ectodomain fragment in WM165 cells (Supplementary Fig. S8C). In COS1 cells, MMP14 likewise cleaved L1CAM into differently sized fragments (Supplementary Fig. S8G). These results indicate that, while MMP14 and MMP16 can both cleave L1CAM, MMP16 is the major endogenous L1CAM sheddase in these cells.

To examine whether L1CAM shedding contributed to the MMP16-dependent WM852 cell adhesion and vascular invasion, shMMP16 cell morphology and transendothelial migration were assessed after L1CAM knockdown. L1CAM siRNA did not alter the defective homotypic shMMP16 cell–cell adhesion (Supplementary Fig. S8H). Although the increased yet inefficient 2D LEC transmigration of shMMP16 cells was somewhat suppressed, intravasation of the cells into LEC spheroids remained unaltered after L1CAM silencing (Fig. 6F and G). In contrast, L1CAM silencing rescued the shMMP16 cell transmigration across BECs to the lower level of shScr cells (Fig. 6F). Recombinant L1CAM ectodomain (L1CAM-Fc) did not revert the phenotype (Supplementary Fig. S8I), suggesting that increased full-length L1CAM in shMMP16 cells facilitates BEC transmigration. Consistent with these in vitro results, the shift from LVI to BVI was coupled with >2-fold increased full-length L1CAM without changes in corresponding mRNA in shMMP16 xenografts (Fig. 6H and I). The intensity of cell-surface L1CAM likewise showed an inverse pattern with MMP16 expression in human melanoma and the xenografts (Fig. 6J–M and Supplementary Fig. S9).

MMP16 controls melanoma cell morphology and collagenase expression via MMP14 regulation

The MMP16-dependent suppression of single-cell phenotype, BVI, and collagen infiltration, all properties promoted by MMP14 (9, 29, 30), raised the possibility that MMP16 supports the alternative melanoma cell aggregation- and collagen assembly-associated LVI phenotype indirectly by limiting cell-surface MMP14. To test this hypothesis, MMP14 protein and mRNA were assessed and related to collagen patterns in WM852 xenografts. Indeed, shScr tumors with high MMP14 and MMP16 displayed mainly intracellular MMP14 in collagen-surrounded cell nests (Fig. 7A). Moreover, while MMP14 mRNA remained unaltered, prominently cell surface-localized MMP14 was significantly increased in conjunction with the pattern of more loosely organized cell masses and fragmented collagen in shMMP16 tumors (Fig. 7A and B; P < 0.01). Considering the MMP expression in melanoma cells (Supplementary Table S1), and MMP16-dependent changes in MMP2 and the MMP-regulator EMMPRIN (Fig. 6A), we also analyzed soluble MMP1 and MMP2 in WM852 cell CM. MMP2, detected as 72-kDa zymogen, was decreased by 30% after MMP16 silencing (Supplementary Fig. S8J). Interestingly, the activated form of MMP1 was enhanced by MMP14 silencing, and conversely diminished coincidently with the MMP14 increase in shMMP16 cells (Fig. 7C). However, silencing of neither MMP14 nor the soluble MMPs altered LEC spheroid intravasation in 3D fibrin (Fig. 7D), consistent with limited direct regulation of melanoma–LEC interactions, and sufficiency of MMP16 for WM852 cell fibrin invasion. In vivo, collective cell phenotype favors LVI over BVI (25), and the shift from LVI to BVI after MMP16-depletion was associated with single WM852 cell invasion. Therefore, we also tested the impact of increased MMP14 on shMMP16 cell–cell adhesion. Importantly, MMP14 siRNAs rescued the cell–cell adhesive phenotype (Fig. 7E).

Figure 7.

MMP16 silencing increases cell surface MMP14 involved in cell junction disassembly and collagen degradation. A, IHC for MMP14 in shScr and shMMP16 WM852 xenografts (brown). Arrowheads, cytoplasmic MMP14; arrows, membranous MMP14. Herovici staining visualizes collagen bundles (red) around cell nests in shScr and dispersed cell growth in shMMP16 xenografts (bottom). B, MMP14 staining intensity and MMP14 mRNA expression in the xenografts. C, immunoblotting of shScr and shMMP16 cells after siRNA transfections as indicated. The ratio of activated approximately 46 kDa and latent approximately 54-kDa MMP1 is expressed below each lane. D, LEC spheroid intravasation by WM852 cells. The mean number of intravasated siCtrl-transfected shScr cells was set to 100. E, phalloidin staining of the cells (F-actin). F, schematic model: MMP16 cleaves/downregulates MMP14 and L1CAM thus limiting their activities on collagen fragmentation, cell junction disassembly, BM degradation, MMP1 activation, and BVI, leading to cell–cell adhesion, expansive growth of collagen-surrounded cell nests, and passive LVI. *, P < 0.01.

Figure 7.

MMP16 silencing increases cell surface MMP14 involved in cell junction disassembly and collagen degradation. A, IHC for MMP14 in shScr and shMMP16 WM852 xenografts (brown). Arrowheads, cytoplasmic MMP14; arrows, membranous MMP14. Herovici staining visualizes collagen bundles (red) around cell nests in shScr and dispersed cell growth in shMMP16 xenografts (bottom). B, MMP14 staining intensity and MMP14 mRNA expression in the xenografts. C, immunoblotting of shScr and shMMP16 cells after siRNA transfections as indicated. The ratio of activated approximately 46 kDa and latent approximately 54-kDa MMP1 is expressed below each lane. D, LEC spheroid intravasation by WM852 cells. The mean number of intravasated siCtrl-transfected shScr cells was set to 100. E, phalloidin staining of the cells (F-actin). F, schematic model: MMP16 cleaves/downregulates MMP14 and L1CAM thus limiting their activities on collagen fragmentation, cell junction disassembly, BM degradation, MMP1 activation, and BVI, leading to cell–cell adhesion, expansive growth of collagen-surrounded cell nests, and passive LVI. *, P < 0.01.

Close modal

Aggressive melanomas are characterized by tumor cell aligning collagen patterns, high lymphatic vessel density, LVI, and early lymph node metastasis (4–7). These characteristics are accompanied by a specific gene expression profile, including overexpression of cell–cell adhesion molecules and the membrane-anchored protease MMP16 in nodular melanomas (12). Despite these results, the molecular and cellular mechanisms behind adverse clinical outcome remain poorly understood. Here, we show that the overexpression of as yet poorly characterized MMP16 is associated with both the growth pattern of tumor cell nests toward outlining continuous collagen and LVI in primary human melanomas. Human WM852 melanoma xenografts recapitulated a similar tumor phenotype that was dependent on MMP16, since its silencing switched the tumors toward a more dispersed growth pattern devoid of continuous collagen or LVI. Our results indicate that MMP16 confers melanoma cells with the properties linked to aggressive tumor behavior by activities including regulation of its cell-surface substrates MMP14 and L1CAM (Fig. 7F). Supportive of such a causal relationship, the MMP16 gain correlated with poor survival of melanoma patients.

MMP16 is located in chromosome 8q21. Copy-number gains involving chromosome 8 have been previously linked to poor prognosis in uveal melanoma, where specific collagen patterns also associate with disease aggressiveness (6, 31). The herein identified MMP16 functions help to explain these correlations, and further suggest that increased MMP16 expression by, for example, copy-number gain predicts aggressive progression in cutaneous melanoma. The gain also affects other chromosome 8q genes, including MYC with established functions toward melanoma progression. However, unlike MMP16, MYC was widely overexpressed across melanoma patients independent of the 8q gain or survival status, similarly to MMP14. The large size of the amplicon, approximately 500 genes, further implicates that MYC alone does not explain decreased survival of patients harboring the 8q gain. In contrast, the identified MMP16-dependent tumor properties suggest that MMP16 overexpression serves as an important mechanism behind melanoma pathogenesis.

The transcripts for both MMP14 and MMP16 were overexpressed in the primary human melanomas and WM852 shScr xenografts with continuous collagen patterns coupled to nodular-type growth and LVI. MMP14 is the major pericellular collagenase, which also cleaves cell-surface proteins, thus enhancing junctional disassembly and mesenchymal invasion of tumor cells across tissue barriers (8, 29, 32). Our results reveal a novel concept, whereby MMP16, rather than inducing cell dissociation and mesenchymal invasion, regulates tumor progression by impeding pericellular collagen degradation and supporting tumor cell–cell adhesion via posttranslational MMP14 downregulation. In WM852 xenografts, this was coupled with assembly of continuous collagen bundles and networks containing blood and lymphatic vessels around tumor cell nests, the morphology prevailing also in the MMP14high/MMP16high human tumors. This was distinct from tumor cell–destructed collagen in both shMMP16 xenografts, with increased cell-surface MMP14, and the MMP14high/MMP16low clinical samples, or from the cohesive noncollagen-infiltrative morphology of the MMP14low/MMP16low tumors devoid of LVI. Although MMP16 reduces cell-surface MMP14 (16), the remaining activity was sufficient for the growth of cell nests and LVI. However, the faster growth of MMP16-silenced xenografts can reflect the enhanced degradation of invasion- and growth-restricting collagen by MMP14 (9, 33). Interestingly, MMP16-expressing control cells instead produced higher levels of activated MMP1 and proMMP2. These soluble MMPs, although inefficient for invasion of tumor cells across cross-linked ECM barriers (9), degrade interstitial collagen and other ECM components. They have been linked to, for example, lung metastasis signature of breast carcinomas as well as signaling, adhesion, and motility toward metastasis in melanoma (34–36). Therefore, these proteases provide the MMP16-expressing melanomas with an alternative mechanism for ECM remodeling within the tumor nests to support growth perpendicularly toward the outlining collagen bundles.

Lymphatic vessels have discontinuous BM that permits intravasation of collectively invading tumor cells (25, 37). Therefore, pressure from growing cell nests inside continuous collagen bundles can push the cells into the permissive vessels, explaining the prominent LVI in MMP16-expressing xenografts and human tumors. Previously, collagen linearization and cross-linking have been found to generate force that promotes breast cancer metastasis (38). Consistent with such an ECM-dependent mechanism of intravasation, transmigration of cultured melanoma cells across LECs was dependent on 3D matrix. In our fibrin model, MMP16 induced extensive melanoma cell intravasation particularly to LEC spheroids. Although we did not directly compare chemoattraction by LECs and BECs, this result can reflect the essential activity of MMP16 for melanoma cell invasion in fibrin, coupled with strong chemoattraction by LECs (16, 26). MMP16 silencing instead enhanced both BEC transmigration in the absence of fibrin and BVI in vivo in conjunction with increased L1CAM and MMP14, both known to promote BVI (27, 30). Because L1CAM depletion reverted the BEC transmigration of shMMP16 cells back to the low control level, the identification of L1CAM as an endogenous MMP16 substrate may prove relevant in the regulation of BVI. Indeed, the mesenchymally invading shMMP16 cells were able to infiltrate both blood vessels and collagen barriers in vivo, thus avoiding the generation of pressure and LVI. Because L1CAM silencing had minor effects on LEC spheroid intravasation, it may not directly mediate LVI. However, by limiting the ability of melanoma cells to invade into blood vessels, shedding of L1CAM by MMP16 could indirectly enhance their invasion into lymphatic vessels (Fig. 7F). These results are consistent with LVI being relatively passive process, induced by intratumoral pressure and chemokines secreted by LECs (26, 39). Because poor melanoma prognosis is associated with the continuous collagen structures and LVI, rather than BVI, our results reveal a novel paradigm, where the MMP16-dependent downregulation of both mesenchymal invasion- and BVI-promoting MMP14 and L1CAM is coupled with LVI and the other properties of the most aggressive tumors (4–7). These mechanisms, together with findings linking high MMP16 and low MMP14 expression to stem-like properties in prostate cancer, and deregulated MMP16-targeting miRNAs to malignant properties in several cancers, are relevant also considering targeting options against alternative cancer invasion/dissemination programs (40–42).

Whether LVI represents a functional route for melanoma spread or just a marker of metastatic switch remains unclear (43). Because of slow fluid flow of lymphatic vessels, LVI provides tumors with a potential dissemination route where cell clusters may grow and adhere. Low shear flow and filtering into subcapsular sinuses can further promote melanoma cell aggregation, which correlates with metastatic ability and lung colonization in mouse models (44, 45). However, it is unclear whether tumor cell aggregates from lymph nodes can access systemic blood circulation to allow distant metastasis (43). Alternatively, LVI may simply reflect more metastatic aggregation-prone cell phenotype. In this case, the correlation with poor prognosis can be explained by, for example, protection of circulating tumor cell clusters from host immunity, their enhanced trapping in capillaries of distant organs, or decreased dormancy due to survival signals and tumor-propagating properties brought by cell–cell adhesion (45–47). Nevertheless, by demonstrating that concurrently with poor survival of melanoma patients MMP16 promotes LVI of tumor cell aggregates, our results reveal a novel mechanism of melanoma progression. Because we show that MMP16 is associated with aggressive melanoma properties already in primary tumors, it is an excellent prognostic marker candidate that could also guide clinical decisions such as SLN biopsies to predict aggressiveness of, for example, small cutaneous melanomas.

No potential conflicts of interest were disclosed.

Conception and design: O. Tatti, J. Keski-Oja, K. Lehti

Development of methodology: O. Tatti, P. Pekkonen, P.M. Ojala, K. Lehti

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): O. Tatti, E. Gucciardo, T. Holopainen, P. Maliniemi, A. Ranki

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): O. Tatti, E. Gucciardo, R. Louhimo, P. Repo, J. Lohi, V. Rantanen, S. Hautaniemi, A. Ranki, P.M. Ojala, J. Keski-Oja, K. Lehti

Writing, review, and/or revision of the manuscript: O. Tatti, E. Gucciardo, P. Pekkonen, R. Louhimo, P. Maliniemi, S. Hautaniemi, K. Alitalo, A. Ranki, P.M. Ojala, J. Keski-Oja, K. Lehti

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P. Pekkonen, P. Repo, J. Keski-Oja

Study supervision: K. Alitalo, P.M. Ojala, J. Keski-Oja, K. Lehti

The authors thank Beatriz Martins, Anne Remes, and Sami Starast for excellent technical assistance, Dr. Paula Kujala (Fimlab Laboratories, Tampere), and Prof. Kai Krohn for providing part of the human tissue samples, and Biomedicum Imaging Unit (Biomedicum Helsinki) and Light Microscopy Unit (Institute of Biotechnology) for imaging facilities. Current results are in part based upon data generated by Oncomine.org and The Cancer Genome Atlas project established by the NCI and NHGRI (http://cancergenome.nih.gov/).

This work was supported by the Academy of Finland, University of Helsinki Research Foundation, Sigrid Juselius Foundation, Finnish Cancer Foundation, Finnish Cancer Institute, Helsinki University Central Hospital Fund, Magnus Ehrnrooth Foundation, Doctoral Programme in Biomedicine, Instrumentarium Foundation, Finska Läkaresällskapet, Finnish Medical Foundation, and Paulo Foundation.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1.
Dadras
SS
,
Lange-Asschenfeldt
B
,
Velasco
P
,
Nguyen
L
,
Vora
A
,
Muzikansky
A
, et al
Tumor lymphangiogenesis predicts melanoma metastasis to sentinel lymph nodes
.
Mod Pathol
2005
;
18
:
1232
42
.
2.
Balch
CM
,
Gershenwald
JE
,
Soong
SJ
,
Thompson
JF
,
Atkins
MB
,
Byrd
DR
, et al
Final version of 2009 AJCC melanoma staging and classification
.
J Clin Oncol
2009
;
27
:
6199
206
.
3.
Emmett
MS
,
Symonds
KE
,
Rigby
H
,
Cook
MG
,
Price
R
,
Metcalfe
C
, et al
Prediction of melanoma metastasis by the Shields index based on lymphatic vessel density
.
BMC Cancer
2010
;
10
:
208–2407–10–208
.
4.
Alitalo
A
,
Detmar
M
. 
Interaction of tumor cells and lymphatic vessels in cancer progression
.
Oncogene
2012
;
31
:
4499
508
.
5.
Egger
ME
,
Gilbert
JE
,
Burton
AL
,
McMasters
KM
,
Callender
GG
,
Quillo
AR
, et al
Lymphovascular invasion as a prognostic factor in melanoma
.
Am Surg
2011
;
77
:
992
7
.
6.
Warso
MA
,
Maniotis
AJ
,
Chen
X
,
Majumdar
D
,
Patel
MK
,
Shilkaitis
A
, et al
Prognostic significance of periodic acid-Schiff-positive patterns in primary cutaneous melanoma
.
Clin Cancer Res
2001
;
7
:
473
7
.
7.
Mäkitie
T
,
Summanen
P
,
Tarkkanen
A
,
Kivelä
T
. 
Microvascular loops and networks as prognostic indicators in choroidal and ciliary body melanomas
.
J Natl Cancer Inst
1999
;
91
:
359
67
.
8.
Willis
AL
,
Sabeh
F
,
Li
XY
,
Weiss
SJ
. 
Extracellular matrix determinants and the regulation of cancer cell invasion stratagems
.
J Microsc
2013
;
251
:
250
60
.
9.
Rowe
RG
,
Weiss
SJ
. 
Navigating ECM barriers at the invasive front: the cancer cell-stroma interface
.
Annu Rev Cell Dev Biol
2009
;
25
:
567
95
.
10.
Tam
EM
,
Morrison
CJ
,
Wu
YI
,
Stack
MS
,
Overall
CM
. 
Membrane protease proteomics: Isotope-coded affinity tag MS identification of undescribed MT1-matrix metalloproteinase substrates
.
Proc Natl Acad Sci U S A
2004
;
101
:
6917
22
.
11.
Hofmann
UB
,
Westphal
JR
,
Zendman
AJ
,
Becker
JC
,
Ruiter
DJ
,
van Muijen
GN
. 
Expression and activation of matrix metalloproteinase-2 (MMP-2) and its co-localization with membrane-type 1 matrix metalloproteinase (MT1-MMP) correlate with melanoma progression
.
J Pathol
2000
;
191
:
245
56
.
12.
Jaeger
J
,
Koczan
D
,
Thiesen
HJ
,
Ibrahim
SM
,
Gross
G
,
Spang
R
, et al
Gene expression signatures for tumor progression, tumor subtype, and tumor thickness in laser-microdissected melanoma tissues
.
Clin Cancer Res
2007
;
13
:
806
15
.
13.
Poetsch
M
,
Dittberner
T
,
Woenckhaus
C
. 
Can different genetic changes characterize histogenetic subtypes and biologic behavior in sporadic malignant melanoma of the skin
?
Cell Mol Life Sci
2003
;
60
:
1923
32
.
14.
Shimada
T
,
Nakamura
H
,
Ohuchi
E
,
Fujii
Y
,
Murakami
Y
,
Sato
H
, et al
Characterization of a truncated recombinant form of human membrane type 3 matrix metalloproteinase
.
Eur J Biochem
1999
;
262
:
907
14
.
15.
Sternlicht
MD
,
Werb
Z
. 
How matrix metalloproteinases regulate cell behavior
.
Annu Rev Cell Dev Biol
2001
;
17
:
463
516
.
16.
Tatti
O
,
Arjama
M
,
Ranki
A
,
Weiss
SJ
,
Keski-Oja
J
,
Lehti
K
. 
Membrane-type-3 matrix metalloproteinase (MT3-MMP) functions as a matrix composition-dependent effector of melanoma cell invasion
.
PLoS ONE
2011
;
6
:
e28325
.
17.
Airola
K
,
Karonen
T
,
Vaalamo
M
,
Lehti
K
,
Lohi
J
,
Kariniemi
AL
, et al
Expression of collagenases-1 and -3 and their inhibitors TIMP-1 and -3 correlates with the level of invasion in malignant melanomas
.
Br J Cancer
1999
;
80
:
733
43
.
18.
Tatti
O
,
Vehvilainen
P
,
Lehti
K
,
Keski-Oja
J
. 
MT1-MMP releases latent TGF-beta1 from endothelial cell extracellular matrix via proteolytic processing of LTBP-1
.
Exp Cell Res
2008
;
314
:
2501
14
.
19.
Ovaska
K
,
Laakso
M
,
Haapa-Paananen
S
,
Louhimo
R
,
Chen
P
,
Aittomaki
V
, et al
Large-scale data integration framework provides a comprehensive view on glioblastoma multiforme
.
Genome Med
2010
;
2
:
65
.
20.
Cancer Genome Atlas Research Network
. 
Comprehensive genomic characterization defines human glioblastoma genes and core pathways
.
Nature
2008
;
455
:
1061
8
.
21.
Rhodes
DR
,
Yu
J
,
Shanker
K
,
Deshpande
N
,
Varambally
R
,
Ghosh
D
, et al
ONCOMINE: a cancer microarray database and integrated data-mining platform
.
Neoplasia
2004
;
6
:
1
6
.
22.
MacDonald
JW
,
Ghosh
D
. 
COPA–cancer outlier profile analysis
.
Bioinformatics
2006
;
22
:
2950
1
.
23.
Squires
MH
 III
,
Delman
KA
. 
Current treatment of locoregional recurrence of melanoma
.
Curr Oncol Rep
2013
;
15
:
465
72
.
24.
Hotary
KB
,
Yana
I
,
Sabeh
F
,
Li
XY
,
Holmbeck
K
,
Birkedal-Hansen
H
, et al
Matrix metalloproteinases (MMPs) regulate fibrin-invasive activity via MT1-MMP-dependent and -independent processes
.
J Exp Med
2002
;
195
:
295
308
.
25.
Giampieri
S
,
Manning
C
,
Hooper
S
,
Jones
L
,
Hill
CS
,
Sahai
E
. 
Localized and reversible TGFbeta signalling switches breast cancer cells from cohesive to single cell motility
.
Nat Cell Biol
2009
;
11
:
1287
96
.
26.
Wiley
HE
,
Gonzalez
EB
,
Maki
W
,
Wu
MT
,
Hwang
ST
. 
Expression of CC chemokine receptor-7 and regional lymph node metastasis of B16 murine melanoma
.
J Natl Cancer Inst
2001
;
93
:
1638
43
.
27.
Zhang
H
,
Wong
CC
,
Wei
H
,
Gilkes
DM
,
Korangath
P
,
Chaturvedi
P
, et al
HIF-1-dependent expression of angiopoietin-like 4 and L1CAM mediates vascular metastasis of hypoxic breast cancer cells to the lungs
.
Oncogene
2012
;
31
:
1757
70
.
28.
Beer
S
,
Oleszewski
M
,
Gutwein
P
,
Geiger
C
,
Altevogt
P
. 
Metalloproteinase-mediated release of the ectodomain of L1 adhesion molecule
.
J Cell Sci
1999
;
112
:
2667
75
.
29.
Sugiyama
N
,
Gucciardo
E
,
Tatti
O
,
Varjosalo
M
,
Hyytiainen
M
,
Gstaiger
M
, et al
EphA2 cleavage by MT1-MMP triggers single cancer cell invasion via homotypic cell repulsion
.
J Cell Biol
2013
;
201
:
467
84
.
30.
Perentes
JY
,
Kirkpatrick
ND
,
Nagano
S
,
Smith
EY
,
Shaver
CM
,
Sgroi
D
, et al
Cancer cell-associated MT1-MMP promotes blood vessel invasion and distant metastasis in triple-negative mammary tumors
.
Cancer Res
2011
;
71
:
4527
38
.
31.
van Beek
JG
,
Koopmans
AE
,
Vaarwater
J
,
de Rooi
JJ
,
Paridaens
D
,
Naus
NC
, et al
The prognostic value of extraocular extension in relation to monosomy 3 and gain of chromosome 8q in uveal melanoma
.
Invest Ophthalmol Vis Sci
2014
;
55
:
1284
91
.
32.
Sugiyama
N
,
Varjosalo
M
,
Meller
P
,
Lohi
J
,
Hyytiainen
M
,
Kilpinen
S
, et al
Fibroblast growth factor receptor 4 regulates tumor invasion by coupling fibroblast growth factor signaling to extracellular matrix degradation
.
Cancer Res
2010
;
70
:
7851
61
.
33.
Hotary
KB
,
Allen
ED
,
Brooks
PC
,
Datta
NS
,
Long
MW
,
Weiss
SJ
. 
Membrane type I matrix metalloproteinase usurps tumor growth control imposed by the three-dimensional extracellular matrix
.
Cell
2003
;
114
:
33
45
.
34.
Minn
AJ
,
Gupta
GP
,
Siegel
PM
,
Bos
PD
,
Shu
W
,
Giri
DD
, et al
Genes that mediate breast cancer metastasis to lung
.
Nature
2005
;
436
:
518
24
.
35.
Blackburn
JS
,
Liu
I
,
Coon
CI
,
Brinckerhoff
CE
. 
A matrix metalloproteinase-1/protease activated receptor-1 signaling axis promotes melanoma invasion and metastasis
.
Oncogene
2009
;
28
:
4237
48
.
36.
Foley
CJ
,
Luo
C
,
O'Callaghan
K
,
Hinds
PW
,
Covic
L
,
Kuliopulos
A
. 
Matrix metalloprotease-1a promotes tumorigenesis and metastasis
.
J Biol Chem
2012
;
287
:
24330
8
.
37.
Pflicke
H
,
Sixt
M
. 
Preformed portals facilitate dendritic cell entry into afferent lymphatic vessels
.
J Exp Med
2009
;
206
:
2925
35
.
38.
Levental
KR
,
Yu
H
,
Kass
L
,
Lakins
JN
,
Egeblad
M
,
Erler
JT
, et al
Matrix crosslinking forces tumor progression by enhancing integrin signaling
.
Cell
2009
;
139
:
891
906
.
39.
Azzali
G
. 
Tumor cell transendothelial passage in the absorbing lymphatic vessel of transgenic adenocarcinoma mouse prostate
.
Am J Pathol
2007
;
170
:
334
46
.
40.
Oktem
G
,
Sercan
O
,
Guven
U
,
Uslu
R
,
Uysal
A
,
Goksel
G
, et al
Cancer stem cell differentiation: TGFbeta1 and versican may trigger molecules for the organization of tumor spheroids
.
Oncol Rep
2014
;
32
:
641
9
.
41.
Xu
XT
,
Xu
Q
,
Tong
JL
,
Zhu
MM
,
Nie
F
,
Chen
X
, et al
MicroRNA expression profiling identifies miR-328 regulates cancer stem cell-like SP cells in colorectal cancer
.
Br J Cancer
2012
;
106
:
1320
30
.
42.
Lin
F
,
Wang
X
,
Jie
Z
,
Hong
X
,
Li
X
,
Wang
M
, et al
Inhibitory effects of miR-146b-5p on cell migration and invasion of pancreatic cancer by targeting MMP16
.
J Huazhong Univ Sci Technolog Med Sci
2011
;
31
:
509
14
.
43.
Sleeman
JP
,
Thiele
W
. 
Tumor metastasis and the lymphatic vasculature
.
Int J Cancer
2009
;
125
:
2747
56
.
44.
Updyke
TV
,
Nicolson
GL
. 
Malignant melanoma cell lines selected in vitro for increased homotypic adhesion properties have increased experimental metastatic potential
.
Clin Exp Metastasis
1986
;
4
:
273
84
.
45.
Glinsky
VV
,
Glinsky
GV
,
Glinskii
OV
,
Huxley
VH
,
Turk
JR
,
Mossine
VV
, et al
Intravascular metastatic cancer cell homotypic aggregation at the sites of primary attachment to the endothelium
.
Cancer Res
2003
;
63
:
3805
11
.
46.
Gupta
GP
,
Massague
J
. 
Cancer metastasis: building a framework
.
Cell
2006
;
127
:
679
95
.
47.
Ocana
OH
,
Corcoles
R
,
Fabra
A
,
Moreno-Bueno
G
,
Acloque
H
,
Vega
S
, et al
Metastatic colonization requires the repression of the epithelial-mesenchymal transition inducer Prrx1
.
Cancer Cell
2012
;
22
:
709
24
.
48.
Barretina
J
,
Caponigro
G
,
Stransky
N
,
Venkatesan
K
,
Margolin
AA
,
Kim
S
, et al
The Cancer Cell Line Encyclopedia enables predictive modelling of anticancer drug sensitivity
.
Nature
2012
;
483
:
603
7
.
49.
Xu
L
,
Shen
SS
,
Hoshida
Y
,
Subramanian
A
,
Ross
K
,
Brunet
JP
, et al
Gene expression changes in an animal melanoma model correlate with aggressiveness of human melanoma metastases
.
Mol Cancer Res
2008
;
6
:
760
9
.
50.
Bogunovic
D
,
O'Neill
DW
,
Belitskaya-Levy
I
,
Vacic
V
,
Yu
YL
,
Adams
S
, et al
Immune profile and mitotic index of metastatic melanoma lesions enhance clinical staging in predicting patient survival
.
Proc Natl Acad Sci U S A
2009
;
106
:
20429
34
.

Supplementary data