Vascular Endothelial Growth Factor-C and C-C Chemokine Receptor 7 in Tumor Cell–Lymphatic Cross-talk Promote Invasive Phenotype

The most common carcinomas spread to distant sites via lymphatic vessels, but the mechanisms of tumor cell escape into lymphatic vessels remain unclear. Most likely, lymphatic metastasis involves both invasion toward surrounding lymphatic vessels and either induction of new lymphatic sprouts into the tumor or expansion of peritumoral lymphatic vessels (1, 2). In human biopsies, tumors are often seen invading their surrounding stromal tissue and entering peritumoral lymphatics. Tumor expression of lymphangiogenic factors, specifically vascular endothelial growth factor-C (VEGF-C) and VEGF-D, has been correlated with lymphatic metastasis both in humans and in animal models (3–6). For example, one report showed serum VEGF-C levels in patients with squamous cell carcinoma of the esophagus to be on the order of 16 to 22 ng/mL, compared with 11 ng/mL for healthy patients (7). Because the expression of the primary receptor of VEGF-C and VEGF-D, VEGF receptor-3 (VEGFR-3), is largely restricted to lymphatic endothelial cells (LEC; ref. 1), it has been assumed, and in many cases reported, that VEGF-C–induced tumor lymphangiogenesis is the primary mechanism underlying VEGF-C–enhanced lymphatic metastasis (4, 6, 8).

However, other studies have reported that whereas VEGF-C correlates with metastasis and invasion, tumor lymphangiogenesis does not necessarily (9–14), suggesting alternative functional roles for VEGF-C and VEGFR-3 signaling. For example, Hoshida and colleagues (15) showed, using intravital microscopy, that VEGF-C increases invasion of tumor cells into draining lymphatic vessels. Furthermore, several recent studies have shown that both VEGFR-3 and VEGF-C are expressed by metastatic tumor cells (16–21), suggesting autocrine stimulation mechanisms that may include tumor cell proliferation (19, 22) and invasiveness (21, 23, 24) of tumor cells, directly or indirectly by increasing flow into hyperplastic peritumoral lymphatics (11, 15). These findings suggest that VEGF-C may act in multiple ways to promote lymphatic invasion.

Another factor strongly correlated to lymphatic metastasis is C-C chemokine receptor 7 (CCR7), which plays a critical role in lymphocyte homing toward lymphatic vessels and to secondary lymphoid organs including the lymph nodes (25–27). Dendritic cells up-regulate CCR7 on activation and during maturation to respond to lymphatic-secreted CCL21 (also known as SLC and 6cKine), which binds CCR7 to elicit directional migration. In CCR7-deficient mice, peripheral dendritic cells fail to migrate to the draining lymph nodes following activation (25). Lymph nodes are also a rich source of the CCR7 ligands CCL21 and CCL19 (also known as ELC and exodus-3), secreted by reticular stromal cells, high endothelial venules, and other immune cells (26). Because CCR7 expression has been correlated with lymph node metastasis of cervical, colorectal, gastric, mammary, esophageal, lung, and prostate carcinomas (28–33), it is hypothesized that tumor cells use the same mechanisms as immune cells to home to lymphatics, namely, with CCR7 guiding them up CCL21 gradients (34, 35). Additionally, we have recently shown that tumor cells can use CCR7 and autologous secretion of CCL19 and CCL21 as a “flow sensing” mechanism to home to lymphatics because the pericellular distributions can become biased in the flow direction, causing downstream gradients that then guide the tumor cell to the draining lymphatic (36).

The aims of this study were twofold: First, to determine the mechanisms of enhancement of tumor invasiveness by CCR7 and VEGF-C, and second, to determine whether any correlation or synergy exists between CCR7 and VEGF-C expression by tumor cells in promoting invasion toward lymphatics. We report here that VEGF-C enhances tumor cell chemoinvasion toward lymphatics in two ways: (a) by inducing lymphatics to increase their secretion of CCL21 to enhance paracrine signaling by LECs to CCR7-expressing tumor cells, and (b) by autologous ligation of tumor VEGFR-3, which is expressed in low levels when tumor cells are in a three-dimensional environment, leading to an increase in the proteolytic activity of the tumor cell and thereby enhancing its migration potential. These effects of VEGF-C both on tumor cells themselves and on lymphatic endothelium to enhance tumor chemoinvasion offer an alternative mechanism to explain why VEGF-C and CCR7 both correlate with enhanced metastatic potential of tumors to lymph nodes.

Neutralizing antibodies and human recombinant proteins. Neutralizing antibodies against human VEGFR-2 (IMC-1121a) and VEGFR-3 (hf4-3C5) along with a matching control IgG were kind gifts from ImClone Systems, Inc. The antibodies were used at concentrations of 10 μg/mL for hf4-3C5 and 20 μg/mL for the control IgG and IMC-1121a. Goat anti-human CCL21 and mouse anti-human CCR7 neutralizing antibodies (AF366 and MAB197, respectively; R&D Systems) were used at concentrations of 4 and 5 μg/mL, respectively. Recombinant human CCL21 and rhVEGF-C (both from R&D Systems) were used at concentrations of 350 and 100 ng/mL, respectively. For the in vivo study, rhVEGF-C and rhVEGF-A165 (R&D Systems) were injected into mouse skin at 10 ng/mouse.

Cell lines and culture. MDA-MB-435s cells were obtained from American Type Culture Collection (ATCC), and two stably transfected versions of these cells, VEGF-C–overexpressing, green fluorescent protein (GFP)–transfected MDA-MB-435 cells and their control-transfected counterparts, referred to herein as VTs and CTs, were a kind gift from Mihaela Skobe (6). They were maintained in DMEM supplemented with 10% fetal bovine serum (FBS) and 1% penicillin-streptomycin-amphotericin B supplemented with 600 μg/mL zeocin and 400 μg/mL geneticin (all from Life Technologies, Inc.). ZR75-1 (invasive mammary carcinoma) cells were obtained from ATCC and were maintained in RPMI 1640 supplemented with 10% FBS, 1% penicillin-streptomycin, 1.0 mmol/L sodium pyruvate, 10 mmol/L HEPES, and 2.5 g/L d-glucose (Life Technologies). U2OS (osteosarcoma) cells, a kind gift from Ivan Stamenkovic, were maintained in DMEM supplemented with 10% FBS and 1% penicillin-streptomycin. Human dermal LECs were isolated from neonatal foreskin as described (37) and maintained in endothelial basal medium (Cambrex) supplemented with 20% FBS (Life Technologies), 1% penicillin-streptomycin-amphotericin B (Life Technologies), 1 μg/mL hydrocortisone, and 50 μmol/L DBcAMP (both from Sigma); they were used between passages 6 and 9. Human umbilical vein endothelial cells (HUVEC; Promocell) were cultured in the same media as LECs and used between passages 6 and 8.

Three-dimensional chemoinvasion assay. To determine relative chemoinvasiveness between tumor cells and endothelial cells and the effects of tumor secretion of VEGF-C on these interactions, we used 12-mm, 8-μm-pore cell culture inserts (Millipore) with CTs, VTs, HUVECs, and LECs [the latter two being labeled with cell tracker red (CMRA, Invitrogen)]. The underside was seeded with one cell type (attracting cell) whereas the upper side was seeded with the migratory cell type within a three-dimensional collagen matrix. Specifically, the underside of each well was seeded with 200,000 cells (of the chemoattracting cell type) and incubated for 4 h to allow cells to attach. Then the inserts were filled with 100 μL of type I collagen (2.0 mg/mL; BD Biosciences) containing 100,000 cells (of the migratory cell type) in basal medium supplemented with 2% FBS. The following cocultures were used (upper-lower): HUVEC-CT; HUVEC-VT; LEC-CT; LEC-VT; CT-HUVEC; CT-LEC; VT-HUVEC; VT-LEC; and the controls, HUVEC-null, LEC-null, CT-null, and VT-null. After incubation for 24 h in a 37°C/5% CO₂ incubator, nonmigrated cells were removed and the inserts were fixed in 2% paraformaldehyde. The membrane was removed and mounted with Vectashield containing 4′,6-diamidino-2-phenylindole (Vector Labs); images were captured using a Zeiss Axiovert 220 fluorescence microscope with an Axiocam MRm camera; and the number of migrated cells was counted to allow calculation of normalized migration.

To determine the chemotactic functionality of tumor CCR7 and VEGFR-3, another panel of chemoinvasion assays were carried out as above, with tumor cells seeded in three-dimensional collagen gels within Millipore inserts and 350 ng/mL CCL21 or 100 ng/mL VEGF-C added to the bottom well after serum starvation overnight. In some wells, blocking antibody was added to both top and bottom chambers: 10 μg/mL hF4-3C5, 20 μg/mL IMC-1121a, or both. Analysis of migrated cells was identical to that described above.

Horizontal three-dimensional migration model. A second type of three-dimensional migration assay was established to directly observe interactions between tumor cells and lymphatic or blood endothelial cells (LECs and HUVECs). Using a Lab-Tek four-chambered cover glass system (Nalge Nunc), two adjacent collagen gels (2.5 mg/mL) were cast in each chamber by blocking half the well with a precast PDMS plug while the first gel set, then removing it to pour the second gel. Each gel contained either tumor cells, LECs, or HUVECs at a cell density of 5 × 10⁵/mL, or gel alone. HUVECs and LECs were loaded with cell tracker red CMRA (Invitrogen) according to the manufacturer's instructions before being suspended in the gel to differentiate them from the GFP-transfected tumor cells. After polymerization, basal DMEM supplemented with 2% FBS was added along with one of the following: (a) buffer, (b) 100 ng/mL VEGF-C, (c) 10 μg/mL hF4-3C5, (d) 4 μg/mL neutralizing anti-CCL21, or (e) 5 μg/mL neutralizing anti-CCR7. The entire setup was placed in a 37°C, 5% CO₂ humidified incubator for 24 h and then imaged and quantified for cell invasion into the other compartment. Imaging was done with the use of a Zeiss LSM 510 Meta confocal microscope.

Immunoprecipitation and Western blot. Tumor cells were seeded at 10⁶/mL into 300 μL of either type I collagen (2.5 mg/mL) or T25 flasks (for two-dimensional comparisons) and maintained in basal media for 24 h. Samples were then lysed with modified radioimmunoprecipitation assay (RIPA) buffer and separated on a 7.5% [for VEGFR-2, VEGFR-3, and neuropilin-2 (Nrp-2)] or 12.5% (for CCR7 and VEGF-C) polyacrylamide gel. Protein was transferred onto a polyvinylidene difluoride membrane and the following antihuman antibodies were used for detection: VEGFR-2 (0.1 μg/mL,), VEGF-C (0.1 μg/mL), CCR7 (0.2 μg/mL), Nrp-2 (1 μg/mL; all from R&D Systems), and VEGFR-3 (0.4 μg/mL; sc-321, Santa Cruz). Blots were then probed using an appropriate horseradish peroxidase–conjugated secondary antibody (Bio-Rad) and developed using Western Pico ECL substrate kit (Pierce). For immunoprecipitation, lysates were precleared with protein A agarose beads (Sigma), then incubated in the presence of agarose beads and the corresponding antibody overnight at 4°C. Samples were then run on a gel as described above.

PCR. Tumor cells were seeded in 2.5 mg/mL type I collagen matrices (BD Biosciences) at 10⁶/mL and maintained for 24 h. RNA was extracted using Trizol (Invitrogen) following the manufacturer's protocol. Reverse transcription was done on 1 μg of total RNA using iScript cDNA Synthesis Kit (Bio-Rad). For real-time PCR, a mix of 10 ng cDNA and GoTaq DNA Polymerase (Promega) was prepared for the amplification. Amplification product was then loaded on a 2% agarose gel. The primers used for detecting expression of VEGFR-3 were targeted against base pairs 736–846, which is in the extracellular domain of the receptor (CGCTGGAGCTGCTGGTAGG and CCCGCTCTGCCTGCTTCC for the forward and reverse sequences, respectively). CCL21 expression in human LECs was determined using the forward and reverse primer sequences CAAGACTGGCAAGAAAGGAAAGGG and GGCTGCTCACTGGGCTATGG, respectively. All gene expression data were normalized to the average expression of two housekeeping genes, GAPDH (with forward sequence CACCCACTCCTCCACCTTTGAC and reverse sequence GTCCACCACCCTGTTGCTGTAG) and EF1a1 (AGCAAAAATGACCCACCAATG and GGCCTGGATGGTTCAGGATA for forward and reverse sequences, respectively).

For CCL21 expression in mouse tissue, CCCTGCTTCAACCATTACATCTGC and CCTGCTGTCTCCTTCCTCATTCC were used for the forward and reverse sequences, respectively. Gene expression was normalized to P0 (CGAAAATCTCCAGAGGCACCATTG and GTTCAGCATGTTCAGCAGTGTGG for forward and reverse sequences, respectively).

CCL21 secretion in vitro. CCL21 protein secretion was quantified from LECs cultured at 850,000/cm² on a growth-factor–reduced Matrigel matrix (BD Biosciences) in basal medium for 24 h. On each gel surface, 500,000 cells were seeded. After culture, the apical, basal, and cell compartments were analyzed for CCL21 by ELISA (Quantikine ELISA kit, R&D Systems) after using Cell Recovery Solution (BD Biosciences) for matrix digestion and standard RIPA buffer protocols (Sigma) for cell lysis. The different compartments were separated as follows: First, medium was removed (apical component) and then Cell Recovery Solution was added to the cells and subcellular matrix together, followed by repeated pipetting to digest the matrix. The solution containing cells and matrix was centrifuged for 5 min at 1,500 rpm; the supernatant (basal compartment containing noncellular components) was removed for analysis and the pellet (cellular component) was lysed with RIPA buffer. In each of the three compartments, CCL21 expression was quantified as described above. In addition to evaluating protein secretion, in separate wells, CCL21 expression was evaluated by real-time PCR as described above on the cells lysed with Trizol.

CCL21 secretion in vivo. All protocols were approved by the Veterinary Authorities of the Canton Vaud according to Swiss law. To assess the effects of VEGF-C on CCL21 secretion by lymphatic vessels in an in vivo model, 6- to 8-wk-old C57 mice (Charles River Labs) were injected intradermally with 0.1 mL of saline in one shoulder and 0.1 mL of 100 ng/mL VEGF-C or VEGF-A in the other shoulder. After 24 h, mice were sacrificed; shoulder skin was cryosectioned into 10-μm-thick sections; and sections were stained with rabbit anti-mouse LYVE-1 (Upstate) and anti-mouse CCL21 (R&D Systems) with Alexa Fluor–conjugated IgG secondary antibodies (Invitrogen). These sections were observed and imaged with a Zeiss Axiovert 220 fluorescence microscope with an Axiocam MRm camera, and for each LYVE-1⁺ lymphatic vessel identified, the number of CCL21⁺ pixels relative to the number of LYVE-1⁺ pixels was quantified as a measure of relative CCL21 secretion (because CCL21 is strongly matrix-binding to the basal lamina of the lymphatic vessels ref. 38). CCL21 expression in whole mouse skin (injected region) was quantified by real-time PCR as described above after extracting RNA from mouse skin digests using Cell Recovery Solution.

Tumor cell proteolysis assay. To quantify the effects of autocrine VEGFR-3 signaling on tumor cell invasiveness and proteolysis, we used a “real-time zymography” assay. DQ collagen (Invitrogen) was diluted in 2.5 mg/mL type I collagen at a ratio of 1:100. Cells were loaded with cell tracker red (Invitrogen) to differentiate GFP from the cells with the fluorescent DQ collagen, which fluoresces green on cleavage and thus can reveal proteolytic activity within a gel. The setup was placed in a humidified 37°C, 5% CO₂ incubator for 24 h and then imaged using a Zeiss LSM 510 Meta confocal microscope. Images were treated using MetaMorph (MDS), Photoshop, and Image J to measure both the intensity and size of the proteolytic areas around each cell, as distinguished by the number of pixels positive for green but negative for red (because cells were both red and green). These values were normalized to both cell area and cell number.

Gelatin zymography. To further test whether proteases were affected by the activation of VEGFR-3 by VEGF-C and the activation of CCR7 by CCL21, we used gelatin zymography to detect changes in matrix metalloproteinase (MMP) activity. Cells were left overnight in either minimal medium or minimal medium supplemented with 100 ng/mL rhVEGF-C or 350 ng/mL rhCCL21 and, in some cases, additionally with 10 mg/mL hf4-3C5 (anti–VEGFR-3 neutralizing antibody). Medium from each sample was then mixed with Laemmli sample buffer (Bio-Rad), in a ratio of 2:1, and analyzed by electrophoresis with a 10% Novex zymogram gel (Invitrogen) for 90 min. The gel was developed according to the manufacturer's instructions, stained using SimplyBlue Safestain (Invitrogen), and imaged using Quantity One Imaging System (Bio-Rad). Densitometry (using ImageJ, NIH) was done, subtracting background and normalizing to the unstimulated samples of each tumor cell type (CT, U2OS, and ZR75-1).

Statistics. For determining statistical significance in all quantifications, ANOVA followed by Tukey honestly significant difference tests were used; all data are presented as mean ± SE. Data were considered significant when P < 0.05.

VEGF-C enhances tumor cell chemoattraction to LECs. Tumor-secreted VEGF-C is generally considered a lymphangiogenic growth factor, but we found that VEGF-C also strongly promotes tumor cell attraction toward LECs. Specifically, we compared the migration of tumor cells to that of LECs, and vice-versa, in modified three-dimensional Boyden chambers using HUVECs for controls to determine lymphatic-specific effects. When tumor cells were used as a chemoattracting source, the migration of both LECs and HUVECs was low, even toward VEGF-C–secreting tumor cells (VTs). This contrasts with chemoattraction in a two-dimensional surface, with LEC chemoattraction readily seen toward VTs; the difference in response is likely due to the relatively low invasiveness of LECs in a three-dimensional gel. On the other hand, when endothelial cells were used as the chemoattracting source, VTs were nearly thrice more chemoinvasive toward LECs than were CTs, although both CTs and VTs showed no chemoattraction toward HUVECs (Fig. 1A). These data suggest that VEGF-C secretion by tumor cells enhances tumor cell invasion specifically to LECs.

VEGF-C–enhanced tumor cell migration depends on both VEGFR-3 and CCR7. We next established a horizontal three-dimensional migration chamber with two adjacent gels, each initially containing one cell type, to directly observe invasion into the neighboring gel (Fig. 1B). In this way, we could determine relative chemoinvasiveness of tumor cells toward LECs and vice-versa, as well as the effects of blocking antibodies on tumor cell-LEC interactions. We found that VTs invaded into the adjacent LEC-seeded compartment at a six times higher rate than baseline (i.e., random migration into an adjacent acellular collagen gel), whereas CT invasion was unaffected by the presence of LECs (Fig. 1B). This effect was VEGFR-3 dependent because when anti–VEGFR-3 blocking antibodies were added to the chambers, VT chemoinvasion toward LECs was knocked down to baseline (i.e., the same as VT random migration, and the same as CT migration toward LECs; Fig. 1C). Furthermore, when 100 ng/mL rhVEGF-C protein was added to the wells, it increased CT chemoinvasion toward LECs nearly 10-fold (P < 0.001) while unaffecting VTs, indicating that any effect of VEGF-C secreted by VTs was already saturating (Fig. 1C). This shows that VEGF-C mediates cross-talk between tumor cells and LECs to increase tumor cell chemoattraction and invasiveness.

Because we saw that VEGF-C increased tumor cell invasion toward LECs but not toward HUVECs, used as a control, we hypothesized that this was due to CCL21-mediated chemoattraction. Indeed, we found that blocking either CCL21 or CCR7 could completely block tumor cell chemoattraction to LECs (Fig. 1D). This suggests that VEGF-C secretion by tumor cells and CCL21 secretion by LECs are interconnected and both contribute to their cross-talk. Furthermore, this result suggests that tumor cells express the lymphocyte homing receptor CCR7, which, on activation by CCL21, results in the chemoattraction of tumor cells toward LECs via a mechanism similar to that used by dendritic cells in the process of homing to lymphatic vessels.

VEGF-C increases CCL21 secretion by lymphatics, but CCL21 does not affect VEGFR-3 expression by tumor cells. We tested three tumor cell types—MDA-MB-435s (with two sublines, CT and VT), ZR75-1, and U2OS—and found them all to express the lymphocyte homing receptor CCR7 (Fig. 2A), which has been correlated with lymphatic metastasis (28–33) and which is required for dendritic cell homing to lymphatic vessels. LECs secrete CCL21 to guide CCR7-expressing cells, including dendritic cells (25–27) and some tumor cells (34). Because the cross-talk between VEGF-C and CCR7 in tumor cell-LEC interactions to promote tumor cell invasion was now obvious, we sought to determine mechanisms by which these factors interact.

Because tumor cells were more invasive only in the presence of VEGF-C, we first tested the hypothesis that VEGFR-3 activation in LECs caused up-regulation of CCL21, which would in turn increase CCR7-dependent chemoattraction of tumor cells. To test this, we cultured LECs on a growth-factor–reduced Matrigel matrix and added 0, 25, or 100 ng/mL rhVEGF-C protein for 24 hours. We found, using ELISA, that 100 ng/mL VEGF-C significantly enhanced LEC secretion of CCL21 basally, but not apically (Fig. 2B). This increase in CCL21 expression by VEGF-C was comparable to increases induced by Oncostatin M on LECs (39). Matrigel was used because CCL21 is strongly matrix-binding, particularly to sulfated proteoglycans (38), and unlike reconstituted collagen, Matrigel is rich in sulfated proteoglycans. This finding is consistent with the physiologic situation in which CCL21 is secreted by lymphatics into the basement membrane, where it binds to presumably counteract fluid flow draining into the lymphatic vessel. Additionally, real-time PCR confirmed this increased CCL21 expression in LECs exposed to VEGF-C (Fig. 2C).

Because our finding that VEGF-C caused CCL21 to be up-regulated by LECs was so significant in explaining the link between VEGF-C and CCR7 in tumor cell invasion of lymphatics, we further examined its validity in vivo. We injected C57/Bl6 intradermally, in the back skin over the shoulder, with 10 ng of VEGF-C or VEGF-A on one side and the same volume of saline in the other side as a control. Animals were sacrificed 24 hours later. A portion of the skin samples were digested for PCR analysis of total CCL21 in skin, which showed consistent results from our in vitro studies: CCL21 was roughly twice in VEGF-C–injected skin than in saline-injected skin, and VEGF-A had no effect (Fig. 2C). The remaining skin samples were thin-sectioned, and lymphatic-associated CCL21 was examined by immunostaining, together with LYVE-1 as a lymphatic marker, which is possible because of its strong binding to the basal lamina. Qualitatively, there were obvious differences in CCL21 staining intensity relative to LYVE-1 in the VEGF-C–treated animals (Fig. 2D). We quantified these by comparing the area of CCL21⁺ pixels around LYVE-1⁺ structures with the area of LYVE-1⁺ pixels and saw a >2-fold increase (P < 0.05) in VEGF-C–treated mice (Fig. 2D). Therefore, VEGF-C secretion by tumor cells up-regulates CCL21 expression by LECs, which would explain the CCR7 dependence of the VEGF-C effects on tumor invasion we saw in Fig. 1D.

Three-dimensional environment affects tumor cell expression of VEGFR-3. We next examined whether VEGF-C had autocrine effects on the tumor cells because others have reported correlations between tumor VEGFR-3 and metastasis (17, 18, 20, 23, 40), although expression of VEGFR-3 in some tumor cell lines has been debated. We saw no expression in any cell line when cultured in two-dimensional conditions. However, in three-dimensional gels, both MDA-MB-435s sublines (CTs and VTs) and U2OS cells expressed VEGFR-3, whereas VEGFR-3 could not be detected in ZR75 cells (Fig. 3A). The expression levels were much lower in tumor cells than in LECs, but VEGFR-3 phosphorylation was clearly seen only when VEGF-C was present (Fig. 3A,, middle). VEGFR-2 was not detected in any of the tumor cell lines (Fig. 3A). Interestingly, CTs and VTs also expressed Nrp-2 (Supplementary Fig. S1A), a VEGF-C receptor that lacks an intracellular signaling domain.

Importantly, VEGFR-3 activation did not seem to affect CCR7 expression on tumor cells. CTs and VTs were seeded in collagen gels in minimal medium supplemented with VEGF-C, with or without neutralizing antibodies against VEGFR-3. In all cases, CCR7 expression seemed to be unaffected (Fig. 3A , bottom).

It is now well appreciated that the three-dimensional microenvironment is critical to recapitulating key features of cell behavior (41), particularly for studying tumorigenesis and invasion (42–45). Tumor cell invasion toward lymphatic vessels is a three-dimensional migration process that is likely governed by extracellular matrix (ECM) proteolysis, cell-ECM adhesion, and migration strategies that are different than those used to crawl along a two-dimensional surface—features that all require a three-dimensional environment. Our finding that VEGFR-3 is expressed only when the tumor cells are cultured in three-dimensional environments implies that its function on tumor cells is related to a three-dimensional process such as matrix proteolysis and invasion.

Autocrine VEGFR-3 signaling enhances tumor cell motility and proteolysis. The presence of functional VEGFR-3 on tumor cells raised the question of possible autocrine effects of VEGF-C on tumor cells. Given that ECM proteolysis has been identified as a critical factor in tumor cell invasion (46, 47), we tested whether VEGF-C directly affected invasiveness and proteolytic activity. To this end, tumor cells were seeded in collagen gels with DQ collagen, which fluoresces on ECM cleavage. After 24 hours, the extent of proteolysis around each cell was imaged under a confocal microscope and quantified. VTs were visibly more proteolytically active than CTs (Fig. 3B), and this effect could be completely blocked in VTs by VEGFR-3 blocking or completely recapitulated in CTs by addition of 100 ng/mL rhVEGF-C. The VEGF-C–mediated increase in proteolytic activity was further shown by gelatin zymography, wherein mainly MMP-9, and to a lesser extent MMP-2, was modulated by VEGFR-3 ligation (Fig. 3C). This was true on all the tested tumor cells except ZR75-1, which did not express VEGFR-3 (Fig. 3A). We found statistically significant differences in MMP-9 levels when cells were treated with VEGF-C, but not in any cells treated with CCL21 (Fig. 3C).

In each case, the percentage of cells that were proteolytically active (i.e., those that colocalized with green DQ collagen) and migratory (defined as cells that moved at least one cell diameter within a 12-hour period during live cell tracking) were quantified. We found that VEGF-C, via VEGFR-3 signaling, significantly and substantially enhanced both (Fig. 3D). On the other hand, of those cells that did move, VEGF-C did not affect average cell speed (Supplementary Fig. S1B). These results are consistent with recent reports that autocrine signaling of VEGF-C via VEGFR-3 in cancer cells promotes cell motility and invasion in vitro and in vivo (15, 21, 23).

Thus, in addition to its paracrine effects on LEC secretion of CCL21, which increases chemoinvasion of tumor cells via CCR7, VEGF-C has autocrine effects on tumor cell VEGFR-3 that leads to increased proteolytic activity, which increases their ability to migrate through a three-dimensional matrix.

Autocrine VEGFR-3 signaling increases tumor cell chemoinvasion up a CCL21 gradient. Having established that tumor cells are chemoattracted to LECs via CCR7 (enabling lymphatic-secreted CCL21 gradients to be sensed) and VEGF-C secretion (binding VEGFR-3 on LECs to further enhance CCL21 secretion, and also binding tumor VEGFR-3 to increase their proteolytic activity and thus invasiveness), we then asked whether VEGF-C affects tumor cell chemoinvasion toward CCL21 in the absence of LECs, to confirm autocrine VEGF-C effects of chemoinvasion. To test this, three-dimensional Boyden chemoinvasion assays were used to test chemoattraction toward VEGF-C and CCL21 in the presence of either VEGF-C or blocking antibodies against VEGFR-2, VEGFR-3, or CCR7.

As expected, we saw that CTs were chemoattracted to VEGF-C gradients, and that this chemoattraction could be blocked totally by neutralizing antibodies against VEGFR-3 but not VEGFR-2 or control antibody (Fig. 4A , first five white columns). As expected, VTs were not chemoinvasive toward a VEGF-C gradient, and this can be explained by the fact that their own secreted VEGF-C would dominate an exogenous gradient.

In addition, whereas both CTs and VTs were chemoattracted to CCL21 above baseline values, VTs were more chemoinvasive than CTs, which is consistent with VTs being more proteolytically active than CTs (Fig. 3). This was validated by the fact that CT chemoinvasion toward CCL21 was increased in the presence of VEGF-C to levels seen in VTs (Fig. 3). Chemoinvasion toward CCL21 of both CT + VEGF-C and VT could be blocked by CCR7 blocking antibodies (Fig. 4A). Finally, both ZR75-1 and U2OS cells were chemoattracted to CCL21, indicating that CCR7 was functional, but only U2OS showed enhanced chemoattraction with VEGF-C (Supplementary Fig. S1C), which is consistent with the lack of VEGFR-3 on ZR75-1 cells (Fig. 3A).

Tumor lymphangiogenesis is widely reported to be involved in lymphatic metastasis of solid tumors. Activation of lymphatic VEGFR-3 is required for lymphangiogenesis (1, 4, 48), and although tumor-secreted VEGF-C is correlated with incidence of lymph node metastasis (3, 7, 8, 49), the requirement of lymphangiogenesis for lymphatic escape of tumors has not been directly shown (11, 14). Recently, it has become evident that many tumors may also express VEGFR-3 (11, 15–24), suggesting the presence of an autocrine mechanism that might promote invasion. Furthermore, the most widely correlated factor with lymph node metastasis in human tumors has been CCR7 (28–33), a receptor required for lymphatic homing of dendritic cells (25–27).

We introduce here a new mechanism in which VEGF-C secretion by tumors and CCR7 expression are positively coupled to further increase metastasis. We show that tumor-secreted VEGF-C not only promotes tumor cell invasion by autocrine mechanisms, to make tumor cells more proteolytic (particularly through MMP-9 secretion) and migratory through a three-dimensional matrix, but also up-regulates CCL21 in lymphatic endothelium to enhance chemoattraction of CCR7-expressing tumor cells toward lymphatics. In this way, VEGF-C secretion by tumor cells has two functions that enhance invasiveness: (a) It gives tumor cells a more invasive phenotype, and (b) it increases chemotaxis toward LECs. These functions add to the previously shown effects of VEGF-C inducing chemoattraction of LECs toward the tumor, which would expand the peritumoral lymphatic network and possibly enhance tumor cell entry; however, in our experiments, this mechanism seems to be a weaker effect in a relevant three-dimensional environment than that on tumor cell invasiveness.

We note that the presence of VEGFR-3 on MDA-MB-435 cells has been debated, as well as the possibility of autocrine VEGF-C signaling in these cells; however, our data highlight the importance of the three-dimensional environment for both expression and receptor phosphorylation. The implications of autocrine VEGFR-3 signaling are significant because these could provide a rationale to treat cancer invasion and metastasis by targeting VEGFR-3 directly on the tumor, in addition to any potential anti-lymphangiogenic effects.

Thus, in summary, VEGF-C secretion and CCR7 expression by tumor cells are positively coupled to synergistically direct and enhance tumor cell invasion toward lymphatic vessels. This has implications not only for tumor invasion in lymphatics but for tumor interactions with CCR7-expressing immune cells as well.

No potential conflicts of interest were disclosed.

Grant support: Swiss National Science Foundation grant 107602 (M.A. Swartz), Department of Defense grant W81XWH-05-1-0492 (M.A. Swartz), Oncosuisse grant 02114-08 (M.A. Swartz), Avon Foundation (M.A. Swartz), and Novartis Foundation grant 06C71 (M.A. Swartz).

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.

We thank Véronique Borel, Joseph Rutkowski, Amanda Knudson, Olga Sazonova, and Daniel Sedehi for valuable assistance; Mihaela Skobe (Department of Oncological Sciences, Mt. Sinai School of Medicine, New York, NY) for the GFP- and VEGF-C–transfected MDA-MB-435s cells; Ivan Stamenkovic (Institute of Pathology, University of Lausanne, Lausanne, Switzerland) for the U2OS cells; and Bronek Pytowski, Dan Hicklin, and ImClone Systems, Inc., for the neutralizing antibodies against VEGFR-2 and VEGFR-3.