Research ArticleCancer

Tumor lymphangiogenesis promotes T cell infiltration and potentiates immunotherapy in melanoma

See allHide authors and affiliations

Science Translational Medicine  13 Sep 2017:
Vol. 9, Issue 407, eaal4712
DOI: 10.1126/scitranslmed.aal4712

Unintentional immunotherapy inhibition

Metastatic spread depends on lymphangiogenesis, and mediators of this pathway are targeted clinically for cancer treatment. Fankhauser et al. used mouse models of melanoma to show that blocking lymphangiogenesis actually disrupted recruitment of naïve T cells and subsequent antitumor immunity. Data from patients enrolled in clinical trials confirmed that indicators of lymphangiogenesis were associated with robust T cell responses. These findings have important implications for the use and predictions of response to immunotherapy.


In melanoma, vascular endothelial growth factor–C (VEGF-C) expression and consequent lymphangiogenesis correlate with metastasis and poor prognosis. VEGF-C also promotes tumor immunosuppression, suggesting that lymphangiogenesis inhibitors may be clinically useful in combination with immunotherapy. We addressed this concept in mouse melanoma models with VEGF receptor–3 (VEGFR-3)–blocking antibodies and unexpectedly found that VEGF-C signaling enhanced rather than suppressed the response to immunotherapy. We further found that this effect was mediated by VEGF-C–induced CCL21 and tumor infiltration of naïve T cells before immunotherapy because CCR7 blockade reversed the potentiating effects of VEGF-C. In human metastatic melanoma, gene expression of VEGF-C strongly correlated with CCL21 and T cell inflammation, and serum VEGF-C concentrations associated with both T cell activation and expansion after peptide vaccination and clinical response to checkpoint blockade. We propose that VEGF-C potentiates immunotherapy by attracting naïve T cells, which are locally activated upon immunotherapy-induced tumor cell killing, and that serum VEGF-C may serve as a predictive biomarker for immunotherapy response.


Many tumors including melanoma are considered lymphangiogenic because they associate with lymphatic vessels (LVs) and can induce lymphatic expansion and activation via vascular endothelial growth factor–C (VEGF-C), and both VEGF-C expression and LV density correlate with poor prognosis in melanoma patients (13). Apart from offering physical routes for metastatic spread (4, 5), recent evidence is emerging that tumor-associated LVs are involved in shaping antitumor immunity (6, 7). On a physical level, they passively drain antigens, cytokines, and danger signals from the tumor to sentinel lymph nodes (LNs), which are essential for the generation of a T cell–inflamed microenvironment in mouse melanoma (8). Moreover, lymphatic endothelial cells (LECs) actively affect immune cell function by releasing immunomodulatory cytokines and by presenting endogenous and exogenous antigens on major histocompatibility complex (MHC) class I and II molecules (916). In this context, it was previously shown that tumor-associated LECs in lymphangiogenic tumors can directly suppress antitumor T cell responses by tolerogenic cross-presentation of tumor antigens (11). Finally, VEGF-C induces LECs to up-regulate CCL21, a chemokine expressed by LN stromal cells that, together with CCL19, guides immune cell subsets into the LN paracortex for education (1719). We have previously shown that tumor-associated CCL21 recruits CCR7+ immune cells into primary mouse melanomas and induces the formation of a lymphoid-like stroma with hallmarks of an immunosuppressive tumor microenvironment (20). Together, these observations have promoted the idea that tumor-associated LECs are one of the many cell types that promote a suppressive microenvironment and help the tumor escape host immunity. A better understanding of the mechanisms that govern immunoregulation by tumor-associated lymphatics should thus enable the rational development of immunotherapeutic strategies.

We asked whether inhibiting tumor-associated lymphangiogenesis, thereby reducing its suppressive effects, would enhance the efficacy of immunotherapy, but we observed the opposite. Instead, we identify a mechanism whereby CCL21-dependent recruitment of naïve T cells into lymphangiogenic melanomas renders the tumor microenvironment more responsive to systemic immunotherapy. We suggest that once in the tumor, naïve T cells are locally primed and activated after immunotherapy-induced tumor cell death, leading to epitope spreading and long-lasting antitumor immunity. These results reveal an unappreciated role of tumor-associated lymphangiogenesis in shaping the tumor immune microenvironment.


VEGF receptor–3 inhibition decreases suppressive features of VEGF-C–expressing B16 melanoma

Lymphangiogenesis was not seen in ovalbumin (OVA)–expressing B16-F10 tumors (B16-OVA), so we modified the cells to express VEGF-C (B16-OVA/VC) (Fig. 1A). We then assessed whether blocking antibodies against VEGF receptor–3 (VEGFR-3) (αR3) altered LEC density after implantation. As expected, VEGF-C increased the density of intratumoral Lyve-1+ LVs, whereas αR3 starting the day of inoculation prevented this lymphangiogenesis (Fig. 1B). After tumor digestion optimized for stromal and immune cell retrieval (21), we confirmed by flow cytometry that LECs (gp38+CD31+), but not blood endothelial cells (BECs; gp38CD31+) or macrophages (Macs; F4/80+), were enriched in lymphangiogenic tumors (Fig. 1, C and D). B16-OVA tumor growth was unaffected by VEGFR-3 blockade, whereas B16-OVA/VC tumors grew somewhat larger than the B16-OVA tumors (Fig. 1E); this increase was eliminated with αR3. No differences were seen among any of the groups in overall survival (Fig. 1E).

Fig. 1. Blocking VEGFR-3 signaling in lymphangiogenic melanomas decreases Treg cell infiltration and delays primary tumor growth.

B16-OVA and B16-OVA/VC tumor–bearing mice were treated with control (Iso) or αR3 starting on the day of inoculation, and tumors were characterized on day 9. (A) Intratumoral VEGF-C concentration (n = 5). N.D., not detected. (B) Immunostained whole-tumor sections [scale bars, 500 μm and 200 μm (in zoomed images)] showing overall lymphatic density [green, Lyve-1; gray, 4′,6-diamidino-2-phenylindole (DAPI)]. (C) Representative flow cytometry plots of tumor cell suspensions for LECs (CD45gp38+CD31+), BECs (CD45gp38CD31+), and Macs (CD45+F4/80+). (D) Quantification of LECs, BECs, and Macs in tumors (n = 5). (E) Growth and survival curves (n ≥ 5). (F) Total infiltrating leukocytes (CD45+). (G) Treg cells (CD4+FoxP3+) and effector CD8+ T (Teff) cells (CD62LCD44+). (H) Gating strategy for myeloid subsets. (I) Quantification of myeloid subsets (mature, m-Myeloid, CD11cCD11b+MHCII+ and immature, imm-Myeloid, CD11cCD11b+MHCII) and MDSCs (granulocytic, G-MDSCs, CD11cCD11b+MHCIILy6G+Ly6Clow, and monocytic, Mo-MDSCs, CD11cCD11b+MHCIILy6G+Ly6Clow). (J) Gating strategy for DC subsets. (K) Quantification of DC subsets: conventional (cDCs; CD11c+CD11b), cross-presenting (CD8+ DCs; CD11c+CD11bCD8+), and myeloid (Myel. DCs; CD11c+CD11b+) (n = 5). All data represent two independent experiments. *P < 0.05, **P < 0.01 by two-tailed Student’s t test.

We speculated that the increased growth rate in lymphangiogenic tumors [that is, B16-OVA/VC tumors treated with isotype control (Iso) antibody] was due to increased inflammatory cell infiltration seen previously (11). We found VEGFR-3–dependent increases in immune cell infiltration (Fig. 1F), most notably in regulatory T (Treg) cells (Fig. 1G), immunosuppressive myeloid subsets, such as immature myeloid cells and myeloid-derived suppressor cells (MDSCs) (Fig. 1, H and I), and antigen-presenting cells, including conventional, cross-presenting (CD8+), and myeloid dendritic cell (DC) subsets (Fig. 1, J and K). We did not detect any significant effects of αR3 on immune cell infiltration in non-lymphangiogenic B16-OVA tumors (Fig. 1, D and F to K). These data demonstrate that VEGF-C expression promotes features of an immunosuppressive tumor microenvironment, whereas inhibiting VEGFR-3 signaling prevents tumor lymphangiogenesis and decreases suppressive cell infiltrates such as Treg cells and MDSCs in B16 melanomas.

Lymphangiogenic melanomas are highly sensitive to immunotherapy

Having confirmed that inhibiting VEGFR-3 signaling and preventing lymphangiogenesis decreases cellular hallmarks of immunosuppression in B16-OVA/VC tumors, we hypothesized that VEGFR-3 blockade would enhance the efficacy of immunotherapy in this model. To test this, we adoptively transferred ex vivo activated OVA-specific CD8+ OT-I cells into tumor-bearing mice on day 9 after tumor inoculation. At early times after adoptive T cell therapy (ATT), αR3-treated tumors responded earlier, with more rapid declines in tumor volume on day 12 (Fig. 2A). However, these tumors reversed course and began progressing again shortly thereafter, whereas lymphangiogenic (Iso-treated) tumors showed a more profound and long-lasting response to ATT. This translated into significantly decreased tumor volume in the progression phase (day 23, P < 0.0001; day 27, P = 0.002) and increased survival of control Iso-treated B16-OVA/VC tumor–bearing mice. In contrast, αR3 treatment had no effect on the efficacy of ATT in B16-OVA tumors, indicating that VEGF-C was necessary to potentiate immunotherapy (fig. S1A). Furthermore, we performed the same experiment in mice lacking dermal LVs (K14-VEGFR-3-Ig mice) (22) and found no difference between B16-OVA and B16-OVA/VC tumor growth or host survival (Fig. 2B), confirming that the VEGF-C–mediated potentiation of ATT was dependent on host lymphangiogenesis.

Fig. 2. VEGF-C/VEGFR-3 signaling increases responsiveness of melanoma to immunotherapy.

Tumor growth and survival of three different melanoma models treated with control (Iso) or αR3-blocking antibodies receiving different immunotherapies (arrows indicate times of administration). (A and B) B16-OVA/VC tumors treated with ATT in (A) WT (n ≥ 15) and (B) K14-VEGFR-3-Ig mice that lack dermal lymphatics (n = 4). ns, not significant. (C to F) B16-OVA/VC tumors in WT mice treated with (C) ex vivo activated DCs (DC vax; n = 6), (D) 50 μg of CpG (n = 6), (E) 10 μg of OVA + 50 μg of CpG (n ≥ 8), and (F) 2 μg of Trp2 peptide–conjugated nanoparticles (NP-Trp2) + 50 μg of CpG (n = 7). (G) B16/VC tumors treated with NP-Trp2 + 50 μg of CpG (n = 6). (H) Tamoxifen-induced tumors in BrafV600E/Pten−/− mice treated with CpG + gp100 peptide (days 8 and 12) and anti–PD-1 antibody (day 12 and every 4 days thereafter). Each panel shows data from one (B to D, F, and G), two (E), or three (A) independent experiments. *P < 0.05, **P < 0.01, ***P < 0.001 by two-tailed Student’s t test for growth curves and log-rank (Mantel-Cox) test for comparing survival curves.

We next asked whether the lymphangiogenic status of B16 melanomas modulates the efficacy toward an immunotherapy approach that relies on raising an endogenous antitumor response. We used a therapeutic DC vaccination (DC vax), in which ex vivo activated, SIINFEKL peptide–pulsed, bone marrow–derived DCs were intravenously injected into tumor-bearing mice on days 4 and 10 after tumor inoculation. As with ATT, we found that αR3-treated tumors began to regress earlier following immunotherapy but then reversed course, whereas lymphangiogenic tumors underwent profound and long-lasting regression (Fig. 2C). Accordingly, the median survival increased from 21 days for αR3-treated mice to more than 2 months for Iso-treated mice bearing B16-OVA/VC tumors.

To test whether the lymphangiogenic status of B16 tumors also modulates non–antigen-specific immunotherapy, we used an adjuvant-only treatment with the TLR9 (Toll-like receptor 9) ligand CpG. Intradermal CpG injection into the hind footpads on days 4, 7, and 10 after tumor inoculation was more effective at controlling tumor growth and enhancing survival in Iso-treated versus αR3-treated mice (Fig. 2D). When CpG was combined with OVA protein, vaccine efficacy was nearly complete (Fig. 2E); in all but two mice, tumors regressed completely. The VEGF-C potentiating effects on immunotherapy were not limited to vaccinating against OVA because a vaccine composed of CpG and nanoparticle-bound endogenous melanoma peptide Trp2 (NP-Trp2) (23) showed similar trends (Fig. 2F). Furthermore, to rule out the possibility that lymphangiogenic potentiation was only effective against tumors expressing the highly immunogenic antigen OVA, we performed a set of experiments with wild-type B16 (B16-WT) and VEGF-C–overexpressing (B16/VC) tumors lacking OVA. As before, VEGFR-3 blockade alone did not affect the growth of B16-WT or B16/VC tumors (fig. S1B), but only VEGF-C–expressing (Iso-treated) tumors showed enhanced response to immunotherapy, either to CpG adjuvant-only therapy (fig. S1C) or to the NP-Trp2 + CpG vaccine (Fig. 2G).

To extend our results beyond transplantable tumor models, we performed immunotherapy in a more clinically relevant, genetically engineered mouse model of melanoma driven by mutated BrafV600E and biallelic deletion of Pten (BrafV600E/Pten−/−) (24). These tumors, induced via topical application of 4-OH-tamoxifen, were naturally lymphangiogenic, and blocking VEGFR-3 led to decreased intratumoral lymphatics (fig. S2, A and B). To raise a potent antitumor immune response, BrafV600E/Pten−/− mice received a combinatorial immunotherapy consisting of a peptide vaccine (CpG + gp100 peptide) combined with anti–PD-1 treatment starting 8 days after tumors were visible. As observed in the B16 model, anti–VEGFR-3–treated BrafV600E/Pten−/− mice responded less well to immunotherapy intervention, whereas immunoglobulin G (IgG)–treated mice showed delayed tumor outgrowth and increased survival (Fig. 2H). Together, these data demonstrate that VEGF-C–mediated lymphangiogenesis potentiates the effects of immunotherapy in several different mouse models of melanoma, despite promoting an immunosuppressive microenvironment.

CCL21 is increased in lymphangiogenic melanomas and drives recruitment of naïve T cells into VEGF-C–overexpressing B16 tumors

We next asked why the more immunosuppressed, invasive tumors would be more responsive to immunotherapy. When examining the immune cell infiltrates before immunotherapy, we found a significant increase (~2.5-fold, P = 0.03) in CD4+FoxP3 T cell density as well as increased CD8+ T cell density in Iso-treated B16-OVA/VC tumors versus αR3-treated tumors (Fig. 3A). The infiltrating subsets responsible for these increases were mainly naïve (CD62L+CD44) T cells, particularly in the CD4+ T cell compartment (Fig. 3B), shifting the ratio between naïve and effector (CD62LCD44+) T cells in lymphangiogenic tumors (Fig. 3C).

Fig. 3. VEGFR-3 signaling increases infiltration of naïve T cells in a CCR7-dependent manner.

(A to D) B16-OVA and B16-OVA/VC tumor–bearing mice were treated with control (Iso) or VEGFR-3–blocking antibodies (αR3) starting on the day of inoculation, and tumors were characterized on day 9 by flow cytometry (data represent two independent experiments, n = 5 each). (A) Quantification of tumor-infiltrating conventional CD4+ T cells (conv CD4+; FoxP3) and CD8+ T cells. (B) Activation status of CD4+ (top) and CD8+ (bottom) T cells. Naïve, CD44CD62L+; effector/effector memory (EM), CD44+CD62L; central memory (CM), CD44+CD62L+. (C) Ratio of naïve/EM in infiltrating CD4+ and CD8+ T cells. (D) CCL21 concentration as assessed by enzyme-linked immunosorbent assay (ELISA) in the tumor, dLNs, and ndLNs. (E) Representative image of a lymphangiogenic B16/VC tumor section immunostained for CD4+ T cells (red), CCL21 (green), LECs (Lyve-1, white), and DAPI (blue). Scale bar, 100 μm. (F) Quantification of CCR7+ T cell subsets in untreated B16-OVA and B16-OVA/VC tumors after 14 days (n ≥ 4). (G to J) B16-OVA/VC tumor–bearing mice were treated with Iso or anti-CCR7 (αCCR7)–blocking antibodies on days 0, 3, and 6, and (G to I) tumors were characterized on day 9 by flow cytometry or (J) mice were given adoptive transfer of 106 naïve OT-I CD8+ T cells (n ≥ 6). (G) Representative flow cytometry plots of T cell activation status. (H) Quantification of naïve and CM fractions of conventional CD4+ (left) and CD8+ (right) T cells. (I) Ratio of naïve/EM subsets. (J) Quantification of intratumoral OT-I cells as percentage of overall CD8+ T cells on day 10 after inoculation. *P < 0.05, **P < 0.01, ***P < 0.001 performed with two-tailed Student’s t test or one-way analysis of variance (ANOVA).

Because naïve, but not effector, T cells express the chemokine receptor CCR7, we assessed tumor expression of the CCR7 ligand CCL21, which is normally expressed by LN stromal cells to guide naïve and memory T cells as well as mature DCs into the LN parenchyma (17, 25). Because CCL21 is expressed by LECs and up-regulated in response to VEGF-C/VEGFR-3 signaling (17, 19, 2628), we were not surprised to find that CCL21 protein was substantially increased in lymphangiogenic B16 (Fig. 3D) and BrafV600E/Pten−/− (fig. S2B) tumors. Furthermore, CCL21 was detected within and around lymphatic endothelium in both tumor models (Fig. 3E and fig. S2C). This effect was restricted to the local tumor microenvironment because no change in CCL21 concentration could be detected in the tumor-draining LNs (dLNs) or non-draining LNs (ndLNs) (Fig. 3D). Accordingly, increased numbers of conventional T cells, but not Treg cells, expressing the CCL21 receptor CCR7 were observed within B16-OVA/VC as compared to B16-OVA tumors (Fig. 3F).

To test whether CCL21/CCR7 signaling was responsible for the increased recruitment of naïve T cells into lymphangiogenic tumors, B16-OVA/VC tumor–bearing mice were treated with CCR7-blocking antibodies (Fig. 3, G to J). CCR7 blockade mainly reduced the infiltration of naïve T cells (Fig. 3, G and H), with a very large reduction in the ratio of naïve versus effector phenotype for both tumor-infiltrating CD4+ and CD8+ T cells (Fig. 3I). We also observed this more directly by adoptive transfer of allogeneic (CD45.2) naïve OT-I cells into mice bearing B16-OVA/VC tumors; after 24 hours, nearly 10-fold fewer OT-I cells were found in αCCR7-treated tumors compared with control-treated tumors (Fig. 3J). Together, these data demonstrate that the CCL21/CCR7 axis drives not only regulatory CD4+ but also naïve T cells into lymphangiogenic B16 melanomas.

In human melanoma, VEGFC expression correlates with CCL21, CCR7, and a T cell signature

We next asked whether the VEGF-C/CCL21 axis was relevant for shaping the immune microenvironment in human melanoma. We first performed immunofluorescence analysis of the lymphatic marker podoplanin in sections from 14 untreated primary human melanomas (Fig. 4A) and found that roughly half were lymphangiogenic, that is, showing substantially higher lymphatic density in the tumor than in the adjacent skin (Fig. 4B). Furthermore, these tumors stained positively for VEGF-C expression (Fig. 4C) as well as CCL21 expression in intratumoral LECs (Fig. 4D).

Fig. 4. Primary human metastatic melanomas contain CCL21-expressing LECs, and expression of VEGFC positively correlates with hallmarks of tumor inflammation.

(A) Representative image of a human primary melanoma immunostained for LVs (green, podoplanin; blue, DAPI). Scale bars, 500 μm (left) and 200 μm (right). (B) Quantification of LV density in tumor (n = 14) and, when present, neighboring skin (n = 7) of primary melanoma tumor sections stratifying patients with elevated intratumor LV density (closed circles), indicating tumor lymphangiogenesis, from those without (open circles). (C) Representative image of a lymphangiogenic melanoma immunostained for VEGF-C (brown). Scale bar, 100 μm. (D) Representative image of an intratumoral LV (podoplanin, green) expressing CCL21 (red). Blue, nuclei (DAPI). Scale bar, 10 μm. (E and F) Correlations of gene expression data of human primary cutaneous metastatic melanoma patients from TCGA. (E) Heat map showing correlation between the expression of 30 genes indicative of T cell inflammation versus VEGFC, FIGF (VEGFD), and VEGFA. Colors indicate minimum and maximum r values using nonparametric Spearman’s test. (F) Dot plots of genes of interest (n = 103) shown with linear regression correlations using nonparametric Spearman’s test.

Several recent reports have described genetic signatures that stratify patient response to immunotherapy according to T cell infiltration, where high-expression levels are seen in responders (24, 2933). We thus analyzed primary tumors from 469 metastatic melanoma patients from The Cancer Genome Atlas (TCGA) database and found strong correlations between VEGFC, but not VEGFA or FIGF (VEGFD), and genes correlating with immunotherapy response (Fig. 4E, left). In addition, several genes previously correlated with immunotherapy resistance, including CTNNB1 and MYC (24), were inversely correlated with VEGFC (Fig. 4E, right). In line with our findings in mice, gene expression of VEGFC correlated with that of CCL21 and CCR7 in human primary melanoma, whereas expression of the other main VEGFR-3 ligand, FIGF (VEGFD), showed no correlation, and VEGFA showed an inverse correlation (Fig. 4F, top two rows). Similarly, expression of CD8, CD4, CD11c, FoxP3, and CD127 (expressed by naïve T cells), but not CD44 (expressed by activated T cells), correlated with that of VEGFC (Fig. 4F, bottom three rows). The same trends were observed in metastatic (secondary) tumors from the same database (fig. S3). Together, these data show that VEGFC and CCL21 expression are strongly correlated in human melanoma and are consistent with the notion that VEGF-C/CCL21 up-regulation shifts the immune microenvironment to drive T cell infiltration.

VEGF-C correlates with response to immunotherapy in human metastatic melanoma patients

We next sought to determine whether VEGF-C correlated with response to immunotherapy in human melanoma patients. Using sera stored from an earlier clinical study of 20 patients who underwent Melan-A analog vaccination (34), we measured VEGF-C concentrations and found direct correlations with T cell response to immunotherapy, in terms of both numbers of circulating Melan-A–specific CD8+ T cells (Fig. 5A) and their expression of the effector cytokine interferon-γ (IFN-γ) (Fig. 5B). Furthermore, Melan-A–specific T cells in patients with high VEGF-C concentration displayed superior polyfunctionality (Fig. 5C).

Fig. 5. Serum VEGF-C correlates with antitumor immune responses and PFS after immunotherapy in human metastatic melanoma patients.

(A to C) Correlations of magnitude and quality of T cell responses with serum VEGF-C concentrations (n = 20) in patients enrolled in a phase 1 clinical study (NCT00112229) evaluating an antitumor Melan-A/MART-1 peptide vaccine. T cell responses reflect peak values across four weekly blood samples in Melan-A tetramer+ CD8+ T cells, and serum VEGF-C was measured before therapy. (A) Antigen-specific T cells as % of circulating CD8+ T cells versus serum VEGF-C. Left: Absolute values for each patient (dotted line indicates mean VEGF-C). Right: Comparison of T cell numbers in patients with low (<mean) versus high (>mean) VEGF-C. (B) IFN-α expression and (C) polyfunctionality in terms of IFN-α, TNF-α (tumor necrosis factor–α), IL-2 (interleukin-2), and CD107 expression in tetramer+ CD8+ T cells. (D) PFS of human melanoma patients (n = 76) enrolled in a phase 2 clinical study (NCT01927419) receiving combined αPD-1 and αCTLA-4 checkpoint blockade. Patients were stratified into three groups (high, mid, low) according to serum VEGF-C, VEGF-D, and VEGF-A concentrations measured before immunotherapy. Groups were compared using a nonparametric Spearman’s test for correlations, two-tailed Student’s t test for dot plots (*P < 0.05), and log-rank (Mantel-Cox) test for survival curves.

We next measured serum VEGF-C concentrations in a larger clinical trial of human metastatic melanoma patients who underwent combined anti–CTLA-4 (ipilimumab) and anti–PD-1 (nivolumab) therapy to determine correlations with progression-free survival (PFS). Strikingly, PFS correlated with high levels of serum VEGF-C but not VEGF-A or VEGF-D, each stratified into mid, low, and high according to means ± 0.4 SD (Fig. 5D). Together, these data demonstrate that serum VEGF-C concentration before immunotherapy not only predicts the magnitude and quality of immune responses raised by a cancer vaccine but also stratifies long-term patient responses to combined checkpoint blockade and further strengthens the case for investigating the use of serum VEGF-C as a predictive biomarker for immunotherapy candidates.

Lymphangiogenic potentiation of immunotherapy is dependent on CCR7 signaling and local activation of naïve T cells

Naïve T cells can be recruited and primed within primary tumors (3537), and we asked whether the increased accumulation of naïve tumor-infiltrating lymphocytes (TILs) within lymphangiogenic B16 tumors was responsible for the increased susceptibility to immunotherapy. We hypothesized that naïve CCR7+ T cells within lymphangiogenic tumors could be locally activated after immunotherapy-induced release of tumor antigen and innate immune activation, thereby increasing antigen spreading as well. Accordingly, we found that, 3 days after ATT, Iso-treated B16-OVA/VC tumors contained higher numbers of activated as well as naïve endogenous CD8+ T cells together with more transferred OT-I cells as compared to αR3-treated tumors (Fig. 6A). This was not the case in the dLNs, where the observed immune cell subsets accumulated to a similar extent, independent of VEGFR-3 signaling (Fig. 6B). To test whether the lymphangiogenic potentiation of ATT was dependent on VEGF-C–induced recruitment of naïve T cells into the tumor, CCR7-blocking antibodies were administered during tumor development until a few days before immunotherapy to avoid confounding effects. We found that after the initial response to ATT, B16-OVA/VC tumors of αCCR7-treated mice responded similarly as αR3-treated mice, namely, with a faster initial response to ATT but later relapse and rapid progression compared to Iso-treated mice (Fig. 6C). This suggested that the recruitment of naïve T cells to the tumor before immunotherapy, which depends on CCR7 signaling, was required for the VEGF-C enhancement of immunotherapy.

Fig. 6. Increased efficacy of immunotherapy in lymphangiogenic B16 melanomas depends on CCR7 signaling before therapy and local activation and expansion of TILs after therapy.

(A and B) B16-OVA/VC tumor–bearing mice treated with control IgG (Iso) or anti–VEGFR-3 (αR3)–blocking antibodies were euthanized 3 days after ATT, and tumor single-cell suspensions were analyzed by flow cytometry (n = 5). Quantification of overall naïve CD8+ (CD45+CD8+CD44CD62L+), effector CD8+ (CD45+CD8+CD44+CD62L), and OT-I (CD45+CD8+CD45.1+) T cells (A) in the tumor and (B) in the dLNs. (C) Tumor growth and survival curves of B16-OVA/VC tumor–bearing mice treated with anti-CCR7 (αCCR7), control IgG (Iso), or αR3 antibodies combined with ATT on day 9. CCR7 blockade was performed only before ATT (days 0, 3, and 6) (data pooled from two or more independent experiments, n ≥ 15 total). (D) Tumor growth curves of B16-OVA/VC tumor–bearing mice treated with control IgG (Iso) or αR3 antibodies received daily injections of the small molecular S1P inhibitor FTY720 starting on the same day as ATT was performed (day 9) (n ≥ 5). Statistics show differences between Iso + FTY720 and αR3 + FTY720 by one-way ANOVA. (E) Representative flow cytometry plots and (F) quantification of circulating CD4+ and CD8+ T cells (after B220 exclusion) in blood 26 days after tumor inoculation. *P < 0.05, **P < 0.01, ***P < 0.001 by two-tailed Student’s t test or one-way ANOVA and log-rank (Mantel-Cox) test for survival curves.

To assess whether these naïve T cells were activated within the tumor, we next performed ATT while blocking lymphocyte egress from LNs using the inhibitor FTY720. The enhanced efficacy of ATT in B16-OVA/VEGF-C tumors was unaffected by FTY720 (Fig. 6D), although it severely depleted circulating numbers of lymphocytes (Fig. 6, E and F), indicating that lymphangiogenic potentiation was independent of T cells activated in the LN after immunotherapy. Because VEGF-C–mediated potentiation of immunotherapy in B16 melanomas was dependent on CCR7-mediated attraction of naïve T cells to the tumor and independent of effector T cell recruitment after immunotherapy, it is likely that the local activation and expansion of recruited naïve T cells within the tumor microenvironment is the key mechanism underlying lymphangiogenic potentiation.

Mice that reject lymphangiogenic B16 melanomas after immunotherapy show epitope spreading and protection to rechallenge

If lymphangiogenic tumors attract more naïve T cells that can become activated locally upon tumor cell killing initiated by immunotherapy, then one would expect the antigenic repertoire of tumor-reactive T cells to broaden beyond the targeted antigen (that is, OVA in these experiments) as more of these recruited cells expand—a process called antigen or epitope spreading (3840). Accordingly, we found that, 2 weeks after OVA vaccination (day 23 after tumor inoculation), mice bearing lymphangiogenic tumors (Iso) had increased numbers of endogenous effector CD4+ and CD8+ T cells as well as the OVA-specific CD8+ T cells in blood as compared to either αR3-treated or non–tumor-bearing mice vaccinated with OVA + CpG (Fig. 7A).

Fig. 7. Mice rejecting primary lymphangiogenic B16 melanomas in response to immunotherapy show epitope spreading and long-term protection.

(A to D) B16-OVA/VC tumor–bearing mice that rejected the primary tumor [primary intradermal (1° i.d.) challenge] after therapeutic vaccination received a metastatic rechallenge with intravenous injections of 2 × 105 B16-WT or B16-OVA/VC cells [secondary intravenous (2° i.v.) challenge] at least 10 days after complete regression. Mice that received either no treatment (naïve) or vaccination only (Vax only) served as controls. (A) Flow cytometry analysis of circulating effector CD4+ (CD45+B220CD4+CD44+CD62L), effector CD8+ (CD45+B220CD8+CD44+CD62L), and tumor antigen–specific CD8+ (CD45+B220CD8+SIINFEKL-pentamer+) T cells 23 days after 1° i.d. challenge but before 2° i.v. rechallenge. (B) Representative images of lung metastases. (C) Quantification of metastatic nodules per lung of mice. (D) Circulating tumor antigen–specific CD8+ T cell responses 9 days after the 2° i.v. challenge (data pooled from two independent experiments, n ≥ 5 total). *P < 0.05, **P < 0.01, ***P < 0.001 by one-way ANOVA.

Because the tumors had completely regressed in most of the Iso-treated mice that had been therapeutically vaccinated with OVA-CpG (Fig. 2E), we rechallenged those mice at least 14 days after complete regression, with an intravenous injection of either B16-OVA/VC or B16-WT cells. Because the tumor-bearing mice that had been αR3-treated did not survive, we used non–tumor-bearing mice vaccinated with OVA + CpG and naïve mice as positive and negative controls, respectively. In naïve mice, the B16-OVA/VC cells drove more extensive lung metastasis than the B16-WT (Fig. 7, B and C), consistent with our expectations. Also as expected, the OVA-vaccinated mice were partially protected against metastasis from B16-OVA/VC tumors but not from B16-WT tumors. However, the mice that had previously rejected B16-OVA/VC tumors were almost completely protected against colonization of both B16-WT and B16-OVA/VC, despite having been vaccinated originally only against OVA (Fig. 7, B and C). Moreover, higher numbers of circulating OVA-specific CD8+ T cells were detected in these mice compared to the OVA-vaccinated only (Fig. 7D). Together, these data suggest that antigen-specific immunotherapy in lymphangiogenic B16 tumors not only potentiates ongoing antitumor immune responses but also induces secondary immune responses against a variety of endogenous tumor antigens, conferring long-term memory and protection against pulmonary metastasis.


Collectively, these data reveal a new and unexpected role for tumor-associated lymphangiogenesis in enhancing the efficacy of systemic immunotherapy. Although lymphangiogenic tumors are characterized by hallmarks of immunosuppression before immunotherapy, they were far more sensitive to systemic immunotherapy as compared to those where VEGFR-3 signaling was blocked. Our data suggest that lymphangiogenic potentiation of immunotherapy depends on the recruitment and local activation of CCR7+ cells, particularly naïve T cells and DCs, in melanoma tumors. We therefore hypothesize that upon immunotherapy-induced cytotoxicity, the release of antigens and danger signals promotes local T cell activation, thereby leading to antigen spreading and long-lasting memory. Furthermore, because T cell infiltration has been shown to correlate with patient response to immunotherapy (24, 2933, 41), our findings have translational implications in suggesting that serum VEGF-C may serve as a predictive biomarker for responsiveness to immunotherapy. We demonstrated two independent clinical immunotherapy trials in metastatic melanoma patients, one using peptide vaccination and another using combined PD-1 and CTLA-4 checkpoint blockade that prospectively measured serum VEGF-C correlates with the magnitude and quality of antitumor immune responses and PFS, respectively. This makes the case for VEGF-C– or tumor-associated lymphangiogenesis to be a key determinant of a patient’s “cancer immunogram” (42).

Our findings are surprising on the one hand because VEGF-C expression in human tumors is strongly correlated with LN metastases and poor prognosis (13, 43). However, we have also previously demonstrated (i) that local lymphatics are important for initiating and establishing the inflammatory tumor microenvironment via communication with the dLN (8), (ii) that VEGF-C up-regulates CCL21 as well as suppressive factors in the tumor microenvironment (further supported in the current work with human genomic data) (11, 18), and (iii) that CCL21 expression in mouse melanoma drives LN-like changes in the tumor stroma that support the recruitment and education of immune cells (20). In the absence of immunotherapy, locally recruited T cells either remained naïve or were rendered anergic or tolerant within the highly suppressive microenvironment (11, 20). Our current observations demonstrate that these cells can be activated or reactivated upon immunotherapy (or initiation of tumor cell killing) to form a more powerful defense against the tumor than if they had not been present at all. They also confirm relevance in human melanoma patients, in terms of VEGF-C correlating with both T cell infiltration and response to immunotherapy, despite the well-established correlation of VEGF-C with worse prognosis in the absence of immunotherapy.

Limitations of this study include uncertainty in the following: (i) the potential roles of other immune cell subsets that may be altered in VEGF-C–expressing tumors, including natural killer T cells, γδ T cells, and especially various DC subsets that we found to be enriched; (ii) the extent to which numerous other cytokines that are altered in lymphangiogenic tumors contribute to immunotherapy potentiation; (iii) cell sources of VEGF-C in human tumors; and (iv) indirect effects of naïve T cell recruitment on the immune microenvironment [for example, competition with Treg cells for nutrients (44) or homeostatic cytokines such as IL-7 (45)]. These will require further investigation.

Although tumor lymphangiogenesis has always been correlated to increased metastasis and poor patient prognosis, this work reveals its flip side, bringing into focus a more comprehensive understanding of how it shapes the immune microenvironment. We now appreciate the numerous mechanisms of immunosuppression that a T cell–inflamed tumor develops to survive, including lymphangiogenesis. On the other hand, lymphangiogenesis promotes immune recognition and supports the recruitment and local priming of naïve T cells. In an untreated developing tumor, the suppressive environment prevents the functional activation of these cells, but when the scales are tipped toward activating factors dominating over suppressive ones, as is the case with immunotherapy or possibly other means that drive immunogenic cell death, these T cells become robust participants in antitumor immunity. In this light, tumor-associated lymphatics can play on both teams: that of the tumor and that of host immunity. Figuring out how we can harness the latter is an exciting challenge for immunotherapy.


Study design

Experiments were designed to correlate infiltrating immune cells, chemokine levels, and response to various immunotherapy approaches with lymphangiogenesis and VEGFR-3 signaling in several mouse models. They included randomization across different cages and researcher blinding during caliper measurements. To determine relevance of findings to human melanoma patients, we first analyzed TCGA data sets for VEGFC, VEGFD, and VEGFA compared to gene signatures of immune infiltration. In two different clinical trials for immunotherapy of melanoma patients, serum VEGF-C was measured and correlated to immune status and patient survival. Details of the clinical trial designs are included below. Sample numbers and numbers of replicates performed for each experiment are included in the figure legends. All primary data, used to compile all figures, are given in table S1.

Vaccination trial in metastatic melanoma patients

Serum VEGF-C concentrations from stage III and IV melanoma patients that had been enrolled in a prospective phase 1 study evaluating an antitumor peptide vaccine (, NCT00112229) were analyzed (34). Patients were enrolled upon written informed consent. Briefly, patients had received monthly subcutaneous injections of a vaccine composed of CpG 7909 (PF-3512676) oligonucleotides and Melan-A/MART-1 peptide, emulsified in Montanide ISA-51. Melan-A–specific CD8+ T cell frequency and function were measured in blood by flow cytometry at the time point of peak response (after an average of eight injections). Serum VEGF-C was assessed using a commercial ELISA kit (DVEC00; R&D Systems).

Checkpoint blockade trial in metastatic melanoma patients

Serum was collected from treatment-naïve patients with unresectable, stage III/IV metastatic melanoma in a randomized, double-blind, placebo-controlled, multicenter, two-arm, phase 2 trial, BMS (Bristol-Myers Squibb) CheckMate-069 (CA209-069, NCT01927419). All patients provided written informed consent for the use of biological materials including serum analysis. The study compared ipilimumab (n = 47), given 3 mg/kg every 3 weeks for four doses, to combined (n = 89), ipilimumab (3 mg/kg) + nivolumab (1 mg/kg), given every 3 weeks for four cycles, followed by nivolumab alone (3 mg/kg) every 2 weeks until disease progression or unacceptable toxicity (46). Only patients in the ipilimumab + nivolumab arm were included here (n = 76), excluding 13 patients who had either very short follow-up (<10 days) or no clear evaluation of progression after 10 months. Pretreatment serum samples were analyzed (700 μl per sample) using the 440 Human Biomarker testing service (RayBiotech). PFS was defined according to the RECIST (Response Evaluation Criteria in Solid Tumors) 1.1 criteria (47). Patients were grouped according to the serum concentrations of VEGF-C, VEGF-D, and VEGF-A into mid, low, and high according to whether it was below, within, or above means ± 0.4 SD.

Statistical analysis

Statistics were performed using Prism (v5.0d; GraphPad). Differences between two experimental groups were determined by two-tailed Student’s t test and multiple groups by one-way ANOVA with Tukey’s post-test. Survival curves were assessed with a log-rank (Mantel-Cox) test. TCGA gene data correlations were tested using nonparametric Spearman’s test. All bar graphs show means ± SEM. *P < 0.05, **P < 0.01, ***P < 0.001.


Materials and Methods

Fig. S1. VEGF-C/VEGFR-3 signaling does not affect the growth of B16 melanomas, and only potentiates immunotherapy in lymphangiogenic tumors, including those that do not express OVA.

Fig. S2. In the naturally lymphangiogenic BrafV600E/Pten−/− mouse model, VEGFR-3 blockade reduces intratumoral lymphatics and total CCL21 in the tumor microenvironment.

Fig. S3. Expression of VEGFC, but not FIGF (VEGFD) or VEGFA, strongly correlates with hallmarks of inflammation within metastatic sites of human melanoma.

Table S1. Primary data.


  1. Acknowledgments: We thank B. Pytowski and Eli Lilly for the mF4-31C1 antibody, the BMS for sharing clinical data, P. Corthésy-Henrioud and Y. B. Saida for technical assistance, T. Gajewski for the BrafV600E/Pten−/− mice, K. Alitalo for the K14-VEGFR-3-Ig mice, and the Swiss Federal Institute of Technology Lausanne (EPFL) Flow Cytometry and Bioimaging Core Facilities. Funding: This study was supported by the Swiss National Science Foundation (CR23I2_143754 and 31003A_153471), the European Research Council (AdG-323053), SwissTransMed (35/2013), and Fonds Pierre-François Vittone. Author contributions: M.F., M.A.S.B., L.P., and M.A.S. designed and analyzed experiments. M.F., M.A.S.B., L.P., L.J., A.W.L., S.H., M.R.-R., E.D.C., and C.T. performed experiments. M.A.S.B., N.B., D.E.S., K.H., O.M., and D.H. collected, characterized, and analyzed human melanoma specimens. E.C. performed bioinformatics analysis. M.F., M.A.S.B., and M.A.S. wrote the manuscript with input and revisions from L.P., L.J., A.W.L., D.E.S., K.H., and D.H. Competing interests: Anti–VEGFR-3–blocking antibodies (mF4-31C1) were from Eli Lilly under a material transfer agreement with the EPFL and the University of Chicago. M.A.S., M.F., and M.A.S.B. are inventors on a patent application (62/329,133) submitted by the University of Chicago that covers the potential diagnostic and therapeutic uses of VEGF-C for cancer immunotherapy. The authors declare that they have no other competing interests.
View Abstract

Navigate This Article