Nature | Vol 577 | 30 January 2020 | 691VEGFC mRNA construct shows peak expression after 24 h, whereas the
expression of AAV-VEGF-C does not reach maximal levels for several
weeks^15. We evaluated the therapeutic efficacy of both modalities by
administering mice with each vector on different days, and found that
only prophylactic (day 60) treatment with AAV-VEGF-C resulted in
long-term survival (Extended Data Fig. 4f–h). By contrast, when mono-
therapy with the VEGFC mRNA construct was administered at days 0,
3 or 7, survival benefits were still observed—although none of these
treated mice survived in the face of rapid glioblastoma development
(Extended Data Fig. 4f ).
Similarly, anti-PD-1 and other checkpoint blockade therapies show
little benefit against glioblastoma when used as monotherapies in
preclinical models^16 ,^17 , and have not shown favourable responses in
clinical trials^18. However, strategies that promote T cell responses—
such as activation of dendritic cells^19 , or radiation^16 —potentiate anti-
PD-1 therapy against glioblastoma, suggesting that the hurdles in the
treatment of cancers of the CNS may be during T cell priming. Hypoth-
esizing that VEGF-C increases immune surveillance of the brain, we
reasoned that therapy using VEGF-C would have synergistic antitumour
effects with checkpoint inhibitor therapy. Whereas anti-PD-1 alone had
marginal effects on survival, a combination of treatment with VEGFC
mRNA and anti-PD-1 antibody resulted in regression of tumours and
survival benefits in mice bearing established GL261 tumours (Fig. 3a,
b), in a T-cell-dependent manner (Extended Data Fig. 5d). When mice
that survived in the long term were rechallenged with tumours in the
contralateral hemisphere, they exhibited resistance to the secondary
challenge (Extended Data Fig. 5b), and T cells transferred from the
draining lymph nodes and spleen of mice that rejected tumours con-
ferred protection against a primary challenge (Extended Data Fig. 5c).
A survival benefit was also observed in mice bearing tumours derived
from a more proliferative and invasive syngeneic glioblastoma cell
line (CT-2A) after treatment with a combination of VEGFC mRNA and
anti-PD-1 together with the T-cell-activating agent anti-4-1BB^20
(Extended Data Fig. 5e–g). In addition, other checkpoint inhibitors that
are known to be ineffective^16 ,^17 ,^21 were potentiated with VEGFC mRNA
(Extended Data Fig. 5h–j). Even when mice were treated after the tumour
mass had increased substantially (day 20), the combination therapy
showed survival benefits (Extended Data Fig. 5k, l).
To assess whether T cell priming against glioblastoma in the deep cer-
vical lymph nodes was increased after mice were treated with VEGF-C,
we used an endogenous tumour antigen that is present in many mouse
cancers^22. Endogenous retroviruses are integrated remnants of retro-
virus in the host genome that are silenced epigenetically, but which
often become aberrantly expressed in the dysregulated transcrip-
tional states that are found in cancers^23. We identified endogenous
retrovirus sequences based on the gene Emv2 that were overexpressed
in GL261 cells (Extended Data Fig. 6a, b). We therefore constructed
tetramers against emv2-env (an envelope protein coded by Emv2)^22
and found that after administration of GL261 cells in the flank of mice,
tetramer-positive CD8 T cells were enriched in the draining inguinal
lymph nodes (Extended Data Fig. 6d)—demonstrating that endogenous
tumour-specific T cell priming occurs in an antigen-dependent manner
in response to GL261.
Intracranial inoculation of mice with GL261-Luc resulted in small
tetramer-positive populations of T cells in the deep cervical lymph
nodes (1.54%) (Fig. 3c, d). However, treatment with the VEGFC mRNA
construct resulted in significant increases in the tumour-specific T cell
population in the deep cervical lymph nodes (3.65%) (Fig. 3c, d). In addi-
tion, increases in both the percentage (Fig. 3d) and the absolute num-
bers (Extended Data Fig. 5a) of tumour-specific T cells were detected in
the brain after treatment with the VEGFC mRNA construct. These data
reveal that intracranial tumours elicit minimal CD8 T cell responses in
the deep cervical lymph nodes, but that VEGFC mRNA enhances the
priming of CD8 T cells against tumours in the brain.
We then examined the specificity of VEGF-C in potentiating check-
point inhibitor therapy. First, to confirm the requirement of menin-
geal lymphatic vessels, a soluble version of VEGFR-3 (VEGFR-31–3-Ig)
was administered, using an AAV construct, to sequester VEGF-C; this
resulted in atrophy of lymphatic vasculature in the dura^4 ,^24 (Extended
Data Fig. 7a, b). Treatment with VEGFR-31–3-Ig abrogated the efficacy
of VEGFC mRNA and anti-PD-1 therapy (Extended Data Fig. 7c). VEGF-C
also uniquely provided therapeutic benefits in combination with
anti-PD-1 when compared to other recombinant proteins of the VEGF
family (VEGF-A, VEGF-B, VEGF-Cs and VEGF-D; Cs is Cys156Ser, a VEGFR-
3-selective agonist) (Extended Data Fig. 7d–f ). These experiments
demonstrate the specificity of the antitumour capacity of VEGF-C and
the requirement for the meningeal lymphatic vessels in providing thera-
peutic benefits against glioblastoma.
Tumour-intrinsic VEGF-C has also previously been reported to medi-
ate increased metastases in melanoma and breast cancer^9 ,^24 ,^25. To assess
whether this was a possibility in the CNS, cells expressing blue fluo-
rescence protein (BFP) were tracked to determine the distribution of
tumour cells in vivo. We collected the brain and deep cervical lymph
nodes to measure tumour cells (CD45−BFP+), as well as immune cells
that may have phagocytosed tumour cells (CD45+BFP+). No CD45−BFP+
cells were found in the deep cervical lymph nodes after treatment with
VEGF-C, suggesting that VEGF-C does not promote metastasis of glio-
blastoma to the lymph nodes (Extended Data Fig. 7g–k). In addition,
direct effects of VEGF-C on tumour cells are unlikely, as glioblastoma
cells showed no expression of VEGFR-3, and VEGF-C had no effect on
cell proliferation in culture (Extended Data Fig. 7l, m). Of note, both
VEGF-C delivery vectors (AAV and mRNA) resulted in an increase in the
number of CD45+BFP+ cells in the deep cervical lymph nodes (Extended
Data Fig. 7i)—consistent with reports that antigen drainage is increased
by exogenous VEGF-C^1 ,^2.
To examine how VEGF-C modifies the immune landscape of tumours
before anti-PD-1 therapy, we performed flow cytometry in the brain,
meninges and deep cervical lymph nodes of mice after treatment with
the VEGFC mRNA construct. Myeloid populations of cells showed mini-
mal changes, with no differences in their levels of activation (CD80) or
antigen-presentation capabilities (major histocompatibility complex
(MHC) II) (Extended Data Fig. 8a–g). The largest changes induced by
treatment with the VEGFC mRNA construct were in the number and(^3) Cortex GBM
45
67
89
1011
12
log
(TPM) 2
VEGFA
P < 0.0001
(^2) Cortex GBM
3
4
5
6
7
8
9
log
(TPM) 2
CD31
P < 0.0001
–3 Cortex GBM
–2–1
01
23
45
67
log
(TPM) 2
VEGFC
P = 0.0003
–40–20 20406080
–50
50
100
150
ΔVEGFC
ΔCD3E
PR = 0.0261 = 0.45
–40–20 20406080
–1,000
–500
500
1,000
ΔVEGFC
PR = 0.0001 = 0.71
–40–20 20406080
–20
–10
10
20
30
ΔVEGFC
ΔCD8B
P R = 0.53= 0.0072
abc
de f
ΔCD4
Fig. 2 | Human glioblastoma is deprived of lymphangiogenic signals at
steady state, and VEGF-C levels correlate with T cell inf iltration with
anti-PD-1 therapy. RNA sequencing (RNA-seq) data of tumour tissue (TCGA,
study accession phs000178.v10.p8) and healthy brain cortex (GTEX, study
accession phs000424.v7.p2). a–c, Expression profiles of V EG FA (b), CD31
(angiogenic) (a) and VEGFC (lymphangiogenic) (c) genes in samples of healthy
cortex versus glioblastoma (cortex, n = 133 samples; glioblastoma, n = 147
samples). d–f, Correlation (from RNA-seq) of changes in the expression of
VEGFC and of the T cell markers CD3E (d), CD4 (e) and CD8B (f) after PD-1
therapy (data from GSE121810; n = 24). Data are mean ± s.d. P values were
calculated by two-tailed unpaired Student’s t-test or Pearson’s correlation.