Unexpected contribution of lymphatic vessels to promotion of distant metastatic tumor spread

Lymphangiogenesis occurs in metastasis-bearing lungs

Tumor-associated lymphangiogenesis is commonly seen in primary tumors and sentinel lymph nodes and is associated with poor prognosis (1, 2). We wondered whether lymphangiogenesis might also occur in distant organ metastases. Therefore, we orthotopically implanted spontaneously metastasizing 4T1 breast cancer cells into the fourth mammary fat pad and removed the primary tumors 21 days later. A further 21 days after primary tumor removal, when most mice had developed overt lung metastasis, we compared the lymphatic density in the control lungs and non-nodule covered area in the lungs of 4T1-injected mice. Thy1 (15–17) was used to identify pulmonary LVs, and cytokeratin was used to identify 4T1 cells (Fig. 1A and fig. S1). The lung area occupied by LVs was significantly larger in 4T1-injected mice compared with naïve lungs (4.86 ± 2.00% versus 2.35 ± 0.73%, P < 0.01, Fig. 1B). We next investigated the tracheobronchial lymph nodes of 4T1-injected mice. The weight of lung-draining lymph nodes in these mice was significantly greater than in normal controls (2.29 ± 0.90 mg versus 0.86 ± 0.40 mg, P < 0.001, Fig. 1C). Because a weight increase of tumor-draining lymph nodes is not an exact measure for metastatic colonization, we next performed immunofluorescence staining of the lymph nodes for cytokeratin. Tumor cells were detected in about 30% of the lung-draining lymph nodes (Fig. 1D). Comparable results were obtained in an experimental melanoma metastasis model, in which B16F10 melanoma cells expressing luciferase (luc2) and the fluorescent reporter tdTomato were injected into the tail vein. The resulting metastases in the lungs were accompanied by an increased tdTomato fluorescence signal in the draining lymph nodes (Fig. 1E), and there was a significant correlation between the extent of lung metastases (quantified by bioluminescence) and the weight of the lung-draining lymph nodes (Spearman correlation, P = 0.018; nonlinear regression, r² = 0.42; Fig. 1F). Together, these data reveal that, as in the primary tumor, lymphangiogenesis also occurred in metastases. Therefore, we next investigated whether lymphangiogenesis at sites of metastasis might contribute to further tumor spread to other organs and correlate with poor patient outcome.

High lymphatic density and lymphatic invasion in metastasis-bearing lungs correlate with poor prognosis in melanoma patients

To study the prognostic relevance of lymphangiogenesis at sites of metastasis, we conducted a retrospective study of 266 melanoma patients with lung metastasis (fig. S2A). Tumor cells in lung-draining lymph nodes (fig. S2B) were detected in 60.2% of the 266 patients. Surgically resected lung samples were available from 26 of these patients (fig. S2C). LVs stained for podoplanin with the D2-40 antibody were quantified around (Fig. 2A and fig. S3A) and within metastases in the lungs (Fig. 2B and fig. S3, B and C). The area covered by lymphatics (Fig. 2C) and LV density (the number of LVs per square millimeter) (fig. S3D) were significantly greater around lung metastases in patients who also had metastases in lung-draining lymph nodes (3.73 ± 2.09% versus 2.05 ± 1.50%, P < 0.05, Fig. 2C; 33.93 ± 19.33 LV/mm² versus 18.64 ± 12.49 LV/mm², P < 0.05, fig. S3D). Similarly, the lymphatic area was greater within metastases in patients who also had metastases in lung-draining lymph nodes (2.09 ± 1.70% versus 1.00 ± 0.47%, P < 0.05, Fig. 2D). The LV density showed the same trend without reaching significance (30.94 ± 25.45 LV/mm² versus 19.6 ± 14.97 LV/mm², P = 0.22, fig. S3E). Patients with lymphatic area and lymphatic density values around metastases above the median had a significantly shorter survival time than those with values below the median (14.07 ± 2.16 months versus 38.69 ± 5.44 months, P = 0.001, Fig. 2E; 17.02 ± 3.17 months versus 36.47 ± 6.08 months, P = 0.019, fig. S3F). Similarly, high values for lymphatic area within metastases correlated with poor prognosis (16.87 ± 2.97 months versus 35.87 ± 6.54 months, P = 0.045, Fig. 2F). The difference in LV density was not significant but followed the same trend (21.09 ± 5.62 months versus 33.45 ± 5.73 months, P = 0.16, fig. S3G).

In melanoma, lymphatic invasion in the primary tumor is frequently related to sentinel lymph node metastasis and correlates with poor prognosis (18, 19). We next analyzed the incidence of lymphatic invasion (Fig. 2G) at the site of lung metastases. Lymphatic invasion, defined by the presence of tumor cells within D2-40–positive LVs, was seen more frequently in patients with metastases in lung-draining lymph nodes than in those without (12 of 15 versus 2 of 11, P = 0.0043, Fig. 2H). Furthermore, patients with lymphatic invasion had a shorter survival time (19.56 ± 2.94 months versus 36.03 ± 7.21 months, P = 0.085, Fig. 2I). These data provide the first clinical evidence that lymphatic density within and around metastases and lymphatic invasion correlate with poor prognosis. These findings also raise the possibility that LVs support the growth of metastases and/or facilitate further tumor spread to other organs.

Immune cell infiltration is commonly seen in many types of tumors, and its prognostic value has been reported (20, 21). Immune cells are important players in the tumor microenvironment, actively interacting with other tumor-surrounding components, including tumor-associated LVs. To study the potential correlation between immune cell infiltration and lymphangiogenesis, the degree of immune cell infiltration in the lung sections with melanoma metastases was scored semiquantitatively by a board-certified pathologist (fig. S4A), and the LV density (percentage of tissue area covered by LVs) was analyzed for each sample. There was a clear trend for an increased LV density in the group of patients with a higher degree of immune cell infiltration; however, no statistical significance was reached [P = 0.052, one-way analysis of variance (ANOVA); fig. S4B]. We also compared the degree of immune cell infiltration between patients with and without lung-draining lymph node metastasis (fig. S4C). Statistical analysis (χ² test) did not detect a significant difference (P = 0.18); however, the patients with the highest degree of immune cell infiltration all had metastases in the lung-draining lymph node.

Inducible transgenic overexpression of VEGF-C in the lungs of mice results in increased LV density

To study the function of LVs at sites of distant metastasis in more detail, we used an inducible, Clara cell promoter-driven transgenic mouse model to overexpress VEGF-C in the airway (CCSP-rtTA × tet-O–VEGF-C; CCSP × VEGFC) (22, 23).

In pilot studies, we induced VEGF-C overexpression in 8-week-old mice by applying different concentrations of doxycycline for 2 weeks to identify a dose that reliably induces LV growth without causing chylothorax. On the basis of these studies, a doxycycline concentration of 1 mg/ml was chosen for mice in the C57BL/6 background and 2 mg/ml for mice in the BALB/c background. After 2 weeks of doxycycline treatment of C57BL/6 mice, VEGF-C mRNA expression in the lung was increased more than 25-fold compared to control wild-type × tet-O–VEGF-C (WT × VEGFC) mice (P < 0.001, Fig. 3A). The VEGF-C protein concentration in the lung increased to 141.7 ± 22.24 pg/mg tissue compared to 80.67 ± 23.5 pg/mg tissue in control mice (P = 0.001, Fig. 3B). Accordingly, the area of VEGFR3⁺ (VEGF receptor 3–positive) lymphatics was almost threefold greater than in controls (8.66 ± 1.95% versus 2.98 ± 0.25%, P < 0.01; Fig. 3, C and D). Lung lymphatics around bronchi and pulmonary veins were more abundant but otherwise appeared normal. Lymphatics were also more abundant beneath the visceral pleura of CCSP × VEGFC mice but were sparse or absent in controls (Fig. 3E). The morphology (Fig. 3F) and density (Fig. 3G) of blood vessels were not affected by VEGF-C overexpression.

Metastasis in lungs and other organs is greater in mice with VEGF-C overexpression in the lungs

To determine whether lymphatic growth induced by VEGF-C overexpression in the lung increased metastases, 5 × 10⁵ B16F10-luc2-tdTomato tumor cells were injected via the tail vein of doxycycline-treated CCSP × VEGFC mice, and 14 days later, the metastatic load in the lungs was evaluated by measuring the tdTomato [red fluorescent protein (RFP)]–positive area (Fig. 4A). WT × VEGFC mice were used as controls. This analysis revealed greater metastatic growth (20.69 ± 8.54% versus 14.42 ± 3.82%, P < 0.05; Fig. 4, B and C) in the lungs of CCSP × VEGFC mice. Measurements of tdTomato-positive cells at 24 hours after intravenous injection indicated that the difference did not reflect greater initial tumor cell homing to the lungs of CCSP × VEGFC mice (Fig. 4D). Fourteen days after tumor cell injection, the number of metastatic nodules in the lung was also comparable (fig. S5A). In VEGF-C–overexpressing mice, there were slightly, but not significantly, more large- and medium-sized nodules compared with control mice (fig. S5B). Most metastatic nodules were near bronchi, whereas in control mice, metastatic nodules were scattered throughout the lungs, including near the pleural surface. The incidence of lymphatic invasion was greater in VEGF-C–overexpressing lungs (6 of 10 versus 0 of 10, P = 0.0034, Fig. 4E), as was the weight of the lung-draining lymph nodes (3.59 ± 1.34 mg versus 2.45 ± 0.67 mg, P < 0.05, Fig. 4F). Metastases were also more numerous in other organs of mice with VEGF-C overexpression in the lungs. In the CCSP × VEGFC group, 10 of 15 mice had metastases in other organs beside the lungs, while in the control group, only 3 of 15 mice also had metastases in other organs, indicating that CCSP × VEGFC mice are 7.4 times more likely to have metastasis in other organs compared to the control group (odds ratio, 7.4; P = 0.025, Fisher’s exact test). Furthermore, we quantified the number of organs with metastasis including the liver, intestine, brain, and kidney in each mouse (Fig. 4G) and performed a quasi-Poisson regression analysis to estimate the expected number of affected organs per animal. CCSP × VEGFC mice had 3.6 times more organs affected than control mice (1.2 organs per mouse in CCSP × VEGFC mice versus 0.33 in control mice; P = 0.025), suggesting a greater spread of melanoma cells through lung lymphatics.

VEGF-C expressed in the lung could reach the circulation and promote lymphatic growth and metastasis in other organs. We addressed this issue by measuring VEGF-C mRNA expression, protein levels, and LV density and morphology in the liver and intestine, two common sites of metastasis. After 2 weeks of VEGF-C induction, VEGF-C mRNA expression (1.19 ± 0.66 versus 1.00 ± 0.27, P = 0.52, fig. S6A) and protein levels were comparable in the liver (3.31 ± 0.59 μg/mg tissue versus 3.37 ± 0.84 μg/mg tissue, P = 0.89, fig. S6B) and intestine (mRNA: 1.04 ± 0.25 versus 1.00 ± 0.12, P = 0.75, fig. S6A; protein: 47.99 ± 17.94 pg/mg tissue versus 54.79 ± 13.42 pg/mg tissue, P = 0.47, fig. S6B). VEGF-C protein levels were slightly higher in the serum (24.85 ± 12.69 pg/ml versus 13.52 ± 6.00 pg/ml, P = 0.14, fig. S6B). The density and morphology of LVs in the intestine and liver were comparable in CCSP × VEGFC and control mice (fig. S6, C and E). The width of intestinal lacteals was similar in CCSP × VEGFC and control mice (fig. S6D).

As mechanisms of lung colonization after tail vein injection of B16F10 cells are likely to differ from spontaneously forming lung metastases, we next investigated the effects of increased lung lymphatic density in the 4T1 orthotopic breast cancer model. CCSP × VEGFC mice, bred into the BALB/c background, showed a similar extent of lung lymphangiogenesis as CCSP × VEGFC mice on the C57BL/6 background when a higher dose (2 mg/ml) of doxycycline was used (fig. S7). As before, 1 × 10⁵ 4T1-luc2 cells were orthotopically injected and primary tumors were removed after 21 days. Four days after the surgery, VEGF-C overexpression was induced in the lung. On days 7, 14, and 21, in vivo IVIS imaging was performed to detect the metastases, and 21 days after the surgery, lungs and other organs were resected after euthanization and were quantified ex vivo (Fig. 5A). The primary tumor growth (fig. S8, A and B) and the incidence of lung metastasis (fig. S8C) at the time of primary tumor removal were comparable. Twenty-one days after primary tumor removal (17 days after induction of VEGF-C), overt lung metastasis was detectable in 8 of 16 mice in the control group. Of these, one mouse had to be euthanized before the predefined study end point. In mice with VEGF-C overexpression in the lung, 11 of 16 mice showed detectable lung metastases, of which 3 mice reached the euthanization criteria before the end point. Because of the high variability of spontaneously formed metastasis, we did not observe a significant difference in nodule number and size (fig. S9). Strikingly, the metastatic load over the whole body, monitored by bioluminescence imaging, was much greater in CCSP × VEGFC mice (Fig. 5, B and C). CCSP × VEGFC mice were 4.8 times more likely to have metastasis in other organs compared to the control group (7 of 11 mice in the CCSP × VEGFC group versus 2 of 8 mice in the control group; odds ratio, 4.8; P = 0.17, Fisher’s exact test). Quasi-Poisson regression analysis estimated that 2.9 times more organs were affected in CCSP × VEGFC mice than in control mice (1.45 organs per mouse in CCSP × VEGFC mice versus 0.5 in control mice; P = 0.13; Fig. 5D and fig. S10). Together, these data further support the concept that metastasis-associated lymphatics might facilitate the spread of lung metastases to other organs.

Mice

C57BL/6J and BALB/c WT mice were obtained from Janvier. CCSP-rtTA × tet-O–VEGF-C mice (22, 23) were backcrossed into the C57BL/6 and BALB/c backgrounds. Prox1-GFP mice were provided by Y.-K. Hong, University of Southern California (29). All mice were housed under specific pathogen–free conditions. The overexpression of VEGF-C in CCSP-rtTA × tet-O–VEGF-C mice was induced by administration of doxycycline (1 mg/ml, C57BL/6 mice; 2 mg/ml, BALB/c mice) in drinking water (containing 5% sucrose) for 2 weeks, starting at 8 weeks of age.

Tumor studies

For the B16F10 metastasis model, 5 × 10⁵ B16F10-luc2-tdTomato cells were suspended in 200 μl of phosphate-buffered saline (PBS) and injected into the tail vein. The mice were monitored three times a week for up to 3 weeks. For the 4T1-luc2 spontaneous breast cancer metastasis model, 1 × 10⁵ tumor cells were suspended in 50 μl of PBS and injected into the fourth mammary fat pad of BALB/c mice. The primary tumor growth was monitored every 3 days for 3 weeks until the primary tumors were surgically removed. After the surgery, mice were monitored three times a week for another 3 weeks. The mice were euthanized with an overdose of anesthesia [ketamine (1000 mg/kg), Ketasol, Graeub; medetomidine (3.5 mg/kg), Domitor, Provet], and tissues were collected for analysis. All tumor studies were performed in agreement with the regulations of the local ethical board of Kantonales Veterinäramt Zürich, under license ZH012/15.

Retrospective lung metastasis study

The information on melanoma patients and lung samples was collected at the Department of Dermatology at the University Hospital Zurich. Resected lung samples of patients with melanoma metastasis (n = 26) were sectioned at a thickness of 4 μm and stained for podoplanin (clone D2-40, BioLegend). The LV hotspots around and within metastases were imaged on a Zeiss Axioskop 2 MOT Plus microscope with a 20× objective. The LV number and lymphatic area percentages were quantified in a blinded manner from patient information. Five hotspots per sample were analyzed, and the average value was calculated for each sample. The incidence of lymphatic invasion was determined on the basis of the entire sample. The local ethics committee approved the written informed consent for tissue storage including the retrospective analysis with collection of clinical/laboratory/histological information before collection (Kantonale Ethikkommission Zürich, Biobank/Sammlung von Tumorgewebe, KEK-ZH-Nr. 647).

Tissue processing and immunofluorescence staining

For lung staining, the mouse lungs were filled with 2% low-melt agarose (Bio-Rad) through the trachea before dissection, chilled down in ice-cold PBS, and fixed in 4% paraformaldehyde (PFA) for 2 hours. For sectioning, agarose-filled lungs were embedded in 2% low-melt agarose and sectioned by Vibratome (Leica VT1000 S) into 200-μm-thick slices before staining. In the studies to compare the pulmonary lymphatic density in CCSP × VEGFC with WT × VEGFC mice, transverse sections from the part near the left bronchial branch of the left lobe were used for staining. In the study to compare the pulmonary lymphatic density in the lungs of control and 4T1-injected mice, transverse sections were made at the position of the detected nodules and corresponding positions in the control lungs. For the lungs without visible nodules on the surface, sections from the part near the left bronchus of the left lobe were used. For lacteal staining, approximately 1 cm of small intestine after the pancreatic curve was dissected, and the tube was cut longitudinally and pinned down (inside up) in silicone gel–coated six-well plates and fixed in 4% PFA for 2 hours at 4°C before staining. For liver staining, the right lobe of the mouse liver was dissected and fixed in 4% PFA for 2 hours at 4°C. Longitudinal sections (200 μm) were cut by Vibratome (Leica VT1000 S) before staining. Lymph nodes and other organs were dissected, embedded in OCT (optimal cutting temperature) compound, snap-frozen, and stored at −80°C until preparation of cryosections (7 μm).

For immunofluorescence stainings, sections were fixed with 4% PFA for 20 min, rinsed with PBS, and blocked in PBS with 0.2% bovine serum albumin, 5% donkey serum, and 0.3% Triton X-100 (blocking solution). Primary antibodies suspended in the blocking solution were incubated at room temperature for 2 hours or at 4°C overnight, followed by extensive washing and incubation with secondary antibodies. The antibodies used were Thy1.2 (rat, eBioscience), cytokeratin (wide spectrum, rabbit, Dako), LYVE-1 (rabbit, AngioBio), αSMA-Cy3 (mouse, Sigma-Aldrich), VEGFR3 (goat, R&D Systems), MECA-32 (rat, BD), RFP (rabbit, Rockland), podoplanin (goat, R&D Systems), and CD31 (rat, BD Pharmingen).

Image analysis

Images were taken on a Zeiss Axioskop 2 MOT Plus microscope with 10× or 20× objective, or a Zeiss LSM 880 upright confocal microscope with 20× objective. Image analysis was performed using ImageJ (National Institutes of Health). For the quantification of the Thy1⁺ lymphatic area, structures smaller than 150 μm² were excluded.

Bioluminescence and stereomicroscope imaging

Bioluminescence imaging was performed with an IVIS Spectrum system (PerkinElmer). Fifteen minutes after intraperitoneal injection of d-luciferin (10 µl/g; 15 mg/ml in PBS; PerkinElmer), mice were anesthetized by inhalation of 2% isoflurane in O₂, placed on the temperature control platform of the IVIS machine, and imaged in vivo. After euthanization, mice were dissected and organs were imaged ex vivo. The intensity of luminescence signals was quantified by Living Image Software (PerkinElmer). This luminescence imaging method can detect as few as 50 tumor cells in the lymph node. Stereomicroscope imaging was performed with a Zeiss Axio Zoom.V16 microscope and a QImaging optiMOS sCMOS camera (QImaging), combined with a light-emitting diode illumination system (pE-4000, CoolLED Ltd.).

Cell lines

B16F10 cells expressing luciferase (luc2) and tdTomato were generated and described previously (30). Cells were cultured in Dulbecco’s modified Eagle medium (DMEM) containing GlutaMAX, pyruvate, 10% fetal bovine serum (FBS), and penicillin/streptomycin (Gibco/Thermo Fisher). G418 (1.5 mg/ml; Roche) was added to the culture medium to ensure stable expression of the tdTomato transgene. 4T1 mammary carcinoma cells expressing luciferase (luc2) (PerkinElmer) were maintained in DMEM supplemented with l-glutamine, 10% FBS, and penicillin/streptomycin under standard culture conditions (37°C, 5% CO₂).

RNA extraction and qPCR

RNA from snap-frozen tissues was extracted using the NucleoSpin RNA Kit (Macherey-Nagel) according to the manufacturer’s instructions. The intestinal tissue was snap-frozen after removal of the fat tissue, and RNA was extracted using the RNeasy Lipid Tissue Mini Kit (Qiagen). RNA was reverse-transcribed using the High-Capacity cDNA Kit (Applied Biosystems/Thermo Fisher). qPCR analyses were performed on a 7900HT FAST instrument (Applied Biosystems/Thermo Fisher) in triplicate using SYBR Green (Roche). RPLP0 was used as an internal reference gene. Relative expression (RE) was calculated according to the formula RE_geneX = 2−(CtgeneX − CtRPLP0) and was expressed as fold change normalized to the control condition. The primer sequences used for qPCR were as follows: RPLP0, 5′-AGATTCGGGATATGCTGTTGG-3′ (forward) and 5′-TCGGGTCCTAGACCAGTGTTC-3′ (reverse); VEGFC, 5′-GGGGGCGAGGTCAAGGCTTTT-3′ (forward) and 5′-CCTGGTATTGAGGGTGGGCTGC-3′ (reverse).

Enzyme-linked immunosorbent assay

Tissue samples were resected, snap-frozen, and kept at −80°C. Blood (500 μl) was collected in Capiject tubes (Terumo), left at room temperature for 20 min, and centrifuged at 1200g for 10 min. The serum was collected, snap-frozen, and kept at −80°C. For each tissue sample, 100 mg of tissue was homogenized in 1 ml of PBS and diluted by 1:250 (lung and intestine) or 1:5000 (liver) with PBS before measurements. Mouse VEGF-C levels were measured using a mouse VEGF-C ELISA kit (Cusabio), following the manufacturer’s instructions.

Lung homing assay

B16F10-luc2-tdTomato cells (1 × 10⁶) were injected into the tail vein. After 24 hours, the lung was perfused, resected, minced, and digested in a collagenase solution (Collagenase II, 2 mg/ml, Sigma-Aldrich; deoxyribonuclease I, 40 μg/ml, Roche) for 45 min at 37°C. The digested tissue was passed through a 40-μm cell strainer before erythrocyte lysis with Pharm Lyse (BD). After washing and a second filtration step, the cell suspension was labeled with CD45-APC/Cy7 (103116, BioLegend). 7-AAD (7-aminoactinomycin D; 420404, BioLegend) was used for life/dead discrimination. CD45-negative, tdTomato-positive cells were counted as tumor cells homed into the lungs.

Statistical analyses

Statistical analyses were performed with GraphPad Prism 5 (GraphPad Software Inc.). Graphs represent means ± SD. Student’s t test was used to compare two groups. One-way ANOVA was used to compare three groups. Other types of tests are indicated in the text or the figure legends. A P value of <0.05 was considered statistically significant (indicated by asterisks).