VEGFC negatively regulates the growth and aggressiveness of medulloblastoma cells

Medulloblastoma (MB), the most common brain pediatric tumor, is a pathology composed of four molecular subgroups. Despite a multimodal treatment, 30% of the patients eventually relapse, with the fatal appearance of metastases within 5 years. The major actors of metastatic dissemination are the lymphatic vessel growth factor, VEGFC, and its receptors/co-receptors. Here, we show that VEGFC is inversely correlated to cell aggressiveness. Indeed, VEGFC decreases MB cell proliferation and migration, and their ability to form pseudo-vessel in vitro. Irradiation resistant-cells, which present high levels of VEGFC, lose the ability to migrate and to form vessel-like structures. Thus, irradiation reduces MB cell aggressiveness via a VEGFC-dependent process. Cells intrinsically or ectopically overexpressing VEGFC and irradiation-resistant cells form smaller experimental tumors in nude mice. Opposite to the common dogma, our results give strong arguments in favor of VEGFC as a negative regulator of MB growth. Manon Penco-Campillo, Yannick Comoglio et al. show that VEGFC decreases the proliferation and migration of medulloblastoma cells, as well as their ability to form pseudo vessels. Cells expressing high levels of VEGFC also form smaller tumors when subcutaneously injected into the flank of nude mice, thus highlighting a negative regulatory role for VEGFC on tumor growth.

M edulloblastoma (MB) is the most frequent malignant, pediatric cerebellum tumor. MB rarely occurs in adults 1 . MB is composed of several molecular subgroups. The actual number of MB subgroups is unknown, and it is likely that each subgroup is further divided into several subtypes. The current consensus describes four principal subgroups: Wingless (Wnt), Sonic Hedgehog (SHH), Group 3 and Group 4. Wnt and SHH are characterized by aberrant activation of the corresponding signaling pathways. Since much less is known about the remaining two subgroups and no specific signaling pathway seems to play a prominent part, the consensus was to retain generic names until further discoveries about their biology. Group 3 and Group 4 MBs present N-and c-myc overexpression, p53 inactivation and deleterious chromosomal abnormalities [2][3][4] .
The standard of care for MB associates surgical resection of the tumor, risk-adjusted photontherapy (high-energy X-rays) and chemotherapy. This treatment leads to up to 70% survival at five years following diagnosis. Most patients suffer long-term side effects 5,6 and relapse is fatal in all cases.
Tumor angiogenesis is related to poor prognosis, metastasis and tumor resistance to treatment 7 . However, metastasis also depends on lymphatic vessels, which are involved in draining tumor cells towards lymph nodes and beyond. Indeed, the current belief states that: (i) tumor cells produce Vascular Endothelial Growth Factor C (VEGFC), the main lymphatic endothelial cell growth factor. VEGFC induces sprouting of nearby lymphatic capillaries, intravasation of tumor cells into the neo-formed vessels, thus contributing to lymph node metastasis or even more distant tumor spreading [8][9][10] ; (ii) tumor cells colonize lymph nodes as pre-metastatic lymphovascular niches 11 ; (iii) tumor cells eventually saturate lymphatic vessels and lymph nodes, collateral lymphatic vessels with alternative lymph nodes then bypass the sentinel lymph node and participate in tumor distant metastasis 11,12 . A lymphatic transport system has been documented in the dura mater of mammalian brain [13][14][15][16] , which allows central nervous system perfusion by the cerebrospinal fluid, drainage of the interstitial fluid towards deep cervical lymph nodes and transport of immune cells to the peripheral lymphatics 17 . Specific markers of lymphatic endothelial cells (LYVE1, VEGFR3, NRP2, PROX1, PDPN) might thus be used as prognostic markers of the severity or adverse evolution of MB.
Antiangiogenic therapies promote VEGFC-dependent lymphangiogenesis in clear cell renal cell carcinomas 18 and radiotherapy induces a lymphangiogenic response in head and neck squamous cell carcinomas 19 . Docetaxel chemotherapy elicits VEGFR3-dependent lymphangiogenesis in breast cancer cells, thus potentiating breast tumor growth and metastasis 20 . Hence, lymphangiogenesis might constitute a common thread linking tumor aggressiveness to the reference treatment.
In the current paper, we meant to demonstrate-for the first time to our knowledge in pediatric brain tumors-the correlation between lymphatic marker expression, lymphatic vessels and MB aggressiveness; (i) in vitro, in naive or irradiated cells; (ii) in vivo, in immunodeficient xenografted mice.
We reveal that, opposite to the current dogma, VEGFC represses MB cell proliferation and migration in vitro and presents antitumoral effects in vivo.

Results
Correlation of lymphatic marker expression with MB subgroups. We analyzed the basal expression of markers of lymphatic vessel development 21,22 as a function of MB subgroups-in the most aggressive subgroups (SHH, Group 4, Group 3)-in the Cavalli 23 database of the R2: Genomics Analysis and Visualization Platform (http://r2.amc.nl). Early markers of lymphatic vessel development (i.e., LYVE1, VEGFR3, PROX1) are also lymphatic endothelial cell (LEC) specification markers and VEGFC, a late marker in the development of lymphatic vessels, is rather a marker of lymphatic network expansion 22 . LYVE1, VEGFR3 and PROX1 were significantly more expressed in the SHH group (moderate to intermediate prognosis) than in more aggressive MB subgroups (groups 3 and 4): p < 0.001. Conversely, the SHH tumors presented a lower expression of VEGFC: p < 0.05 vs Group 3; p < 0.01 vs Group 4 (Fig. 1a). Neuropilin-2 (NRP2) is a co-receptor of VEGFR3: it enhances the VEGFC response via VEGFR3 24,25 . However, NRP2 can also function in VEGFC signaling independently of its role as a co-receptor 26,27 and promote tumor lymphangiogenesis and metastasis. Here, in the case of MB, NRP2 mRNA levels present the same profile as VEGFC mRNA levels (Fig. 1a). This suggests that VEGFC signaling relies on NRP2 rather than and independently of VEGFR3 in MB.
To confirm this result, we analyzed the expression of VEGFC and of its receptors in the WNT subgroup and compared it to the other three subgroups ( Supplementary Fig. S1a). Surprisingly, the WNT subgroup patients presented significantly higher expression of VEGFC, VEGFR2 and NRP2, when compared to the other groups, unlike VEGFR3 or CD146, which encodes a VEGFR2 coreceptor 28 . This suggests that, in MB, VEGFC signaling is mostly VEGFR2 and NRP2-directed, and that, especially in WNT, Group 4 and Group 3, VEGFR3 may not be the main receptor involved in VEGFC signaling. Supplementary Fig. S1b presents the overall survival of WNT patients as a function of VEGFC. A low level of VEGFC tended to lead to worse outcome in the least aggressive tumors. In contrast, in the more aggressive SHH subgroup, Group 4 or Group 3 ( Supplementary Fig. S1c-e), high VEGFC tended to correlate with a worse prognosis. This result underlines the dual function of VEGFC (beneficial vs. deleterious) in MB patients.
In an independent cohort of more than 250 patients 29 , high expression of LYVE1 was correlated to a longer survival in SHH group; p = 0.002 (Fig. 1b, left panel), while it was linked to shorter survival in Group 3; p = 0.017 (Fig. 1b, right panel). In Group 4, for which the clinical prognosis is intermediate, high expression of LYVE1 also tended to be favorable (Fig. 1b, middle panel).
We analyzed the R2 data considering the metastatic status of patients in each subgroup ( Supplementary Fig. S2). A high amount of VEGFC mRNA was related to poor prognosis only in metastatic patients ( Supplementary Fig. S2a). High levels of VEGFC mRNA were strongly associated with shorter survival metastatic patients (M1) from the SHH subgroup (Supp. Fig. S2b). Surprisingly, neither Group 4, nor Group 3 patients showed a similar result, although the trend was the same, thus emphasizing the importance of the tumor microenvironment. High expression of LYVE1 was a marker of longer survival in M0 patients (p = 2.5 × 10 −3 ). This result was reversed in the M1 group of patients, in which high expression of LYVE1 was rather a marker of short survival ( Supplementary Fig. S2c).
Finally, we studied the expression of podoplanin (PDPN), a biomarker of malignant disorders 30 , as a function of the metastatic status of MB patients. Tumors derived from M0 patients were labeled all over the epithelial cells, in a diffuse and regular manner (Fig. 1c). Contrarily, tumors derived from metastatic (M1) patients presented localized, lumensurrounding labeling (Fig. 1c), suggesting that PDPN was expressed in lymphatic endothelial cells, only in the most aggressive MB.
These results suggest that lymphangiogenesis is associated with a poor or a favorable outcome depending on the MB genetic subgroup and metastatic status.
Expression of lymphangiogenesis markers in MB cell lines. We mainly used two MB cell lines in our in vitro study. Daoy cells are representative of the SHH subgroup and have a well-defined epithelial morphology ( Supplementary Fig. S3a). HD-MB03 cells are derived from a Group 3 tumor. HD-MB03 are semi-adherent: some of the cells develop in clusters 31 , (Supplementary Fig. S3b). This difference in morphology prompted us to study the status of our models as for epithelial vs. mesenchymal phenotype. In both cell types ( Supplementary Fig. S3c), we analyzed the basal expression of three genes involved in epithelial-tomesenchymal transition (EMT). Cadherin-1 (CDH1) encodes E-cadherin and Cadherin-2 (CDH2) encodes N-cadherin; both are cell-cell adhesion proteins. CLDN1 encodes Claudin-1, a major constituent of the tight junction complexes that regulates the permeability of epithelia. When detected, the epithelial marker CDH1 was low and variable. Thus, we measured CLDN1 expression to confirm the epithelial phenotype of the cells. Daoy cells expressed 11 ± 4 times more CDH2 mRNA, but 202 ± 25 times more CLDN1 mRNA, than HD-MB03 cells (n = 3; p < 0.01 and p < 0.001, respectively). The ratio is in favor of an epithelial phenotype for Daoy cells.
Because of this cluster phenotype of HD-MB03 cells, we evaluated the expression of CD133, the main stem cell marker. HD-MB03 cells were CD133-positive while, in Daoy cells, expression was below the detection threshold ( Supplementary  Fig. S3d).
We defined the in vitro behavior of these two cell lines, focusing on aggressiveness characteristics, in connection with their content in lymphangiogenesis markers. Daoy and HD-MB03 cell proliferation was consistent with their assumed To ascertain whether the slower proliferation characteristics were a general behavior of the SHH-derived tumor cells, we measured ONS-76 cell proliferation in the same conditions as above and compared it with Daoy cell proliferation ( Supplementary  Fig. S4a). ONS-76 cell doubling time was 0.8522 day (95% CI: 0.5875 to 1.213). It was close to the HD-MB03 doubling time and significantly different from Daoy cell proliferation (p < 0.05), thus showing that MB cell models present the same heterogeneity as the pathology they originate from and that any conclusion derived from these in vitro models may be drawn with caution. Surprisingly, the highly aggressive HD-MB03 cells barely migrated in Boyden chambers, while Daoy cells had a migratory phenotype (Fig. 2b).
Since tumor growth and migration/invasion have been closely associated with VEGFC-dependent lymphangiogenesis [33][34][35] , we hypothesized that the above MB phenotypes were linked to the VEGFC/VEGFC-receptor axis. Daoy cell VEGFC mRNA level was 44.3 ± 4.3-fold that of HD-MB03 cells (p < 0.001, Fig. 2c). Moreover, VEGFC secreted protein level was 7427.0 ± 683.2 pg/ ml/10 6 cells/48 h (n = 11) in Daoy cells, while it was barely 133.5 ± 40.2 pg/ml/10 6 cells/48 h (n = 8) in HD-MB03 cells (p < 0.001, Fig. 2d). Both cell types produced VEGFC protein, as shown by immunocytochemistry experiments (Fig. 2e). This result and our ELISA experiments (Fig. 2d) showed that Daoy cells secrete VEGFC, while the cytokine gets trapped in HD-MB03 cells. Although in ONS-76 cells, VEGFC mRNA level was 10 times lower than the Daoy level ( Supplementary Fig. S4b), both cell lines secreted similar amounts of VEGFC (Supplementary Fig. S4c). This result indicates that VEGFC production is a common feature of Daoy and ONS-76 cells, two models of SHH MB. Moreover, it confirms that there is not a strict correlation between mRNA and protein levels, as previously described in other tumor models 36,37 .
We measured the mRNA levels of the most important lymphangiogenesis genes in MB cells. VEGFR2, VEGFR3 and NRP2, the receptors and co-receptor of VEGFC, and PROX1, the main transcription factor involved in lymphangiogenesis, were highly expressed in Daoy cells when compared to HD-MB03 cells (Fig. 2c, Supplementary Fig. S4b). In ONS-76 cells, only VEGFR2 and VEGFR3 present a high expression ( Supplementary Fig. S4b), suggesting that VEGFC signaling does not use the NRP2 pathway in this cell line.
In cells representative of the Group 3 MB, the situation was more complex. HD-MB03 and D458Med cells presented low levels of VEGFC mRNA and of all the receptors and co-receptors ( Fig. 2c, Supplementary Fig. S4b). However, D458Med showed a high production of VEGFC ( Supplementary Fig. S4c), when compared to HD-MB03 and to D341Med, a Group 3 MB model, known for being devoid of VEGFC 38 . This result suggests that autocrine VEGFC signaling does not occur in these two models, despite their synthesis (HD-MB03) and even secretion (D458Med) of VEGFC.
Thus, VEGFC autocrine signaling is a subgroup-dependent phenomenon in MB.
Thus, VEGFC and lymphatic marker levels were higher in the least aggressive cells. Unexpectedly, high VEGFC lowered two MB aggressiveness hallmarks: proliferation and migration.
We questioned the signaling pathway involved in the VEGFCdependent proliferation. ERK pathway was activated in Daoy VEGFC KO cells, while its activity was reduced in VEGFC++ HD-MB03 cells (Fig. 3e, Supplementary Fig. S3g), thus demonstrating the implication of this pathway in MB cell proliferation.
Directional cell migration was increased in Daoy VEGFC KO cells (Fig. 3f, g) and migration was reduced in the VEGFCoverexpressing-HD-MB03 cells (Fig. 3h, i, n = 5; p < 0.001). Based on the vasculogenic mimicry concept 39 , we assessed the ability of our cell lines to form vessel-like structures as a function of their VEGFC content. This concept was named "lymphomimicry". Daoy and ONS-76 cells (SHH subgroup) were both able to form pseudo-vessels ( Supplementary Fig. 4d, e). Daoy: 42.3 ± 2.3 vs ONS-76: 109.4 ± 4.5 pseudo-vessels (n = 4). HD-MB03 cells (Group 3), which do not secrete VEGFC, were also able to form vessel-like structures (424.6 ± 18.8 (n = 4); Supplementary  Fig. S4d, e). Conversely, D341Med and D458Med were unable to organize as pseudo-vessels, thus suggesting that the tumor cells of these two types take another metastatic route than the classic blood or lymphatic pathways.
VEGFC knocking-out significantly reduced the ability of Daoy cells to form vessel-like structures (Fig. 3j, k; n = 3; p < 0.001). However, HD-MB03 cells also lost their ability to organize into tubes when overexpressing VEGFC (Fig. 3l, m; n = 3; p < 0.001), thus suggesting that two mechanisms of tube formation are at stake in the two different cell lines. We conclude that high VEGFC level is a marker of low aggressiveness in MB and that this low aggressiveness model involves an autocrine or paracrine regulation.
Role of VEGFC in MB cell induced aggressive phenotype. In order to mimic relapses following radiotherapy treatment of MB patients, we generated irradiation-resistant cells and examined their VEGFC levels. For two independent populations of Daoy cells, VEGFC expression was not altered at the mRNA level ( Fig. 4a) but was significantly decreased in both populations at the protein level (Fig. 4b). However, VEGFC level stayed high, in the range of the control value. Conversely, HD-MB03 cells displayed both VEGFC mRNA and protein significant increases in both populations of irradiation-resistant cells (Fig. 4a, b). This treatment-related increase in VEGFC mRNA has previously been documented in other tumor models 18,19 . We suspect a similar upregulation process to occur in HD-MB03 cells. Daoy cell proliferation was not affected in resistant cells when compared to control cells (Fig. 4c). Conversely, both populations of irradiation-resistant HD-MB03 cells proliferated slower than control cells (Fig. 4c).
Both populations of Daoy resistant cells migrated less in Boyden chambers than their control counterparts (Fig. 4d). Irradiationresistant HD-MB03 cells displayed no migratory phenotype, as already shown for naive HD-MB03 cells (Fig. 2). Focusing on the lymphomimicry behavior of the cells, both Daoy-and HD-MB03resistant cells lost their ability to form tube-like structures when compared to naive cells (Fig. 4e, f). Thus, the increase in VEGFC following irradiation (Fig. 4a) was enough in HD-MB03 cells to trigger lower different features of cell aggressiveness in vitro. The high level of VEGFC in Daoy cells, whether irradiated or not, triggered the same inhibitory effect.
We measured the expression of VEGFC receptor mRNA in irradiation-resistant cells. VEGFR3 was at a low level and thus not accurately detected in any of our MB cells. In Daoy cells, VEGFC receptor mRNA did not undergo any change under irradiation, except a modest but significant (p < 0.05) reduction of VEGFR2 expression in only one out of two populations of irradiated cells (Fig. 4g). On the contrary, in HD-MB03 cells, CD146 mRNA was drastically increased (10 to 20-fold; p < 0.001), in the same proportion as VEGFC (Fig. 4h), suggesting that the irradiation-induced inhibitory effect of VEGFC is relayed, in this cell line, by the CD146 receptor in vitro. At the mRNA level, CD146 basal expression in Daoy cells was more than 500-fold the HD-MB03 level ( Fig. 4i; p < 0.001). Strikingly, these results did not correlate with membrane expression of CD146. Indeed (Fig. 4j, l), more than 80% of Daoy cells displayed a high membrane labeling by CD146, while only 40% of Daoy irradiation-resistant cells (R1 and R2 populations) were slightly labeled by CD146 (n = 4; p < 0.001). Opposite to Daoy cells, very few (5-10%) HD-MB03 cells presented a faint membrane CD146 level, which was not modified in either irradiation-resistant population (Fig. 4k, l). This demonstrates again that two different aggressiveness mechanisms are involved in the two MB cell models.
Mesenchymal-to-epithelial transition of irradiated MB cells. Since cell proliferation or migration, as well as vasculogenic mimicry have been associated to the epithelial-to-mesenchymal transition (EMT) phenomenon 40,41 , we analyzed the expression of EMT genes in irradiation-resistant cells. Irradiation effect was modest if any, in Daoy cell line (Fig. 5a), since only one out of two populations of irradiation resistant Daoy cells displayed an increase in CDH1 and CLDN1 mRNAs, counterbalanced by an increase in CDH2. More neatly, both populations of HD-MB03 cells presented a significant increase in CLDN1 (n = 3; p < 0.001) coupled to a decrease in CDH2 mRNA (Fig. 5b). The protein levels were consistent with the mRNA analysis ( Fig. 5c-h). Indeed, CDH2 expression was barely modified in Daoy resistant cells, while it was significantly decreased in HD-MB03 resistant cells, although expressed at very low level (Fig. 5c, d; Supplementary Fig. S4f). Immunofluorescent labeling of CDH1 and CLDN1 demonstrated a rather unchanged phenotype in Daoy cells (Fig. 5e, f) while both CDH1 and CLDN1 were induced in HD-MB03 cells (Fig. 5g, h). Hence, HD-MB03 cells adopt a more epithelial phenotype, after chronic irradiation, thus arguing against an increase in cell mobilization after irradiation.
VEGFC effect in in vivo tumor growth. Experimental tumors were generated by subcutaneous injection of MB cells into the flank of nude mice. This experiment was without effect on the mice health and behavior (Supplementary Fig. S5a). HD-MB03derived tumor incidence was high: 100% of the mice were bearing  Fig. S5b). Daoy tumor bearing mice were 90% at most (only 60% for the control group; Supplementary  Fig. S5c). HD-MB03 control tumors reached 1 cm 3 after 24 days, while 75 days were necessary to the biggest Daoy VEGFC KO tumors to reach the same volume (Fig. 6a, b). Tumors generated from HD-MB03 cells were bigger and heavier at time of sacrifice than their VEGFC++ or irradiation-resistant counterparts (n = 8 mice/group, p < 0.001). There was no statistical difference between the other three HD-MB03 groups (Fig. 6a, Supplementary Fig. S5d). Daoy VEGFC KO tumors were significantly bigger than Daoy Ctl tumors at time of sacrifice: n = 6-9 mice per group, p < 0.01 (Fig. 6b, Supplementary Fig. S5e). In the HD-MB03 groups, all the tumors looked very angiogenic, whereas Daoy tumors where much less reddish, except for VEGFC KO (Fig. 6a, b). The in vivo tumor growth experiments were then correlated with the in vitro measurements of cell proliferation, thus confirming that the effects of VEGFC or of the absence of VEGFC may principally consist in an autocrine or paracrine effect on tumor cells.
To emphasize this conclusion, we measured gene expression in the harvested tumors. HD-MB03 tumors that contain high levels of VEGFC, also presented high levels of NRP2 (Fig. 6c). This suggests that the autocrine or paracrine positive effect of VEGFC is transmitted by this co-receptor, rather than by VEGFR3, which was not detected, nor by VEGFR2, which was not upregulated under these conditions. Surprisingly, in VEGFC KO Daoy tumors, downregulation of VEGFC resulted in a downregulation in NRP2, while VEGFR2 level was not modified (Fig. 6d). This infers that VEGFR2, but neither by NRP2 nor VEGFR3, relayed the negative control of VEGFC on tumor.
The mouse homolog of genes involved in lymphangiogenesis was upregulated in both HD-MB03 tumors generated from engineered or irradiated cells (Fig. 6e), suggesting that lymphangiogenesis may take place around HD-MB03 VEGFC-overexpressing tumors and irradiation-resistant tumors. Likewise, VEGFC KO -derived Daoy tumors showed a 3 to 4-fold increase in the same lymphangiogenesis genes suggesting that a lymphangiogenesis process is in progress, which is consistent with the absence of neo-formed lymphatic vessels (Fig. 6f).
Lymphangiogenic markers correlate with tumor aggressiveness. We analyzed the correlation between the lymphangiogenic marker PDPN and tumor aggressiveness in the above-mentioned experimental tumors (Fig. 7a, b).
HD-MB03 tumors (Fig. 7a) were devoid of PDPN labeling (n = 8 mice). In the HD-MB03 VEGFC++ -injected pool of mice (Fig. 7a), only 1 out of 8 mice (12.5%) showed a strong PDPN labeling, with focal and isolated cell labeling, consistent with the high level of VEGFC in the cells.
All but one (n = 7) of the subcutaneous Daoy cell-derived tumors were too small (Fig. 6b) to be used in histochemistry experiments. The last one was strongly labeled (isolated cells and focal labeling) by PDPN, consistent with high levels of lymphangiogenesis markers in Daoy cells. VEGFC KO -injected mice produced bigger tumors (Fig. 6b). 4 out of 6 tumors were labeled by PDPN, with focal labeling.

Discussion
We provide evidence supporting that the VEGFC/VEGFC receptor axes and associated lymphangiogenesis exert a beneficial effect in pediatric medulloblastoma, unlike the admitted dogma. This analysis must be taken as a pre-clinical study, using very few in vitro models of medulloblastoma, when compared to the large number of available cell lines 42 . We thus keep in mind that our conclusions should be broadened with caution.
We determined that VEGFC and several genes involved in lymphangiogenesis are differentially expressed in the subgroups of MB patients. The genes that are necessary for lymphatic specialization (LYVE1, VEGFR3, PROX1) 22 , displayed higher expression in the least aggressive subgroups, opposite to the dogma stating that lymphangiogenesis is an essential feature of several types of aggressive tumors [43][44][45] . Both VEGFR3 46 and PROX1 47,48 mRNAs are highly expressed in the developing cerebellum (https://www.proteinatlas.org/ENSG00000037280-FLT4/ tissue/cerebellum), respectively in the Purkinje cells and the external granule layer, and both mRNAs regulate the neuronal development at early postnatal stages. The high level of VEGFR3 and PROX1 mRNAs might thus be related not only to lymphatic, but also to neuronal development and interestingly, specific neural progenitors from the CNS have recently been shown to promote growth and metastasis of tumors of different origin 49 .
VEGFC, a late gene in the lymphangiogenesis process, essential to the lymphatic system expansion 22 , presented lower expression in the patients from SHH subgroup when compared to more aggressive subgroups. Patient database analysis and immunohistochemistry experiments demonstrated that lymphatic gene expression (VEGFC, LYVE1, PDPN) is a function of the metastasis status. Thus, lymphangiogenic genes do not play the same role in all the MB subgroups. These genes participate in shaping the tumor cell characteristics. Especially, in SHH subgroup, VEGFC expression is correlated with aggressiveness, while LYVE1, PDPN and PROX1 expression is inversely correlated with aggressiveness. Such discrepancies between genes and MB subgroups prompted us to study the role of lymphangiogenic genes, especially VEGFC, in pediatric MB cell aggressiveness.
Counter-intuitively, lymphangiogenic mRNAs (VEGFC, VEGFR3, NRP2, PROX1) were overexpressed in cells from the SHH subgroup (Daoy, ONS-76) when compared to cells from Group 3 (HD-MB03, D458Med, D341Med). Hence, the related genes are not pejorative per se and tumor lymphangiogenesis has beneficial effects in some instances. We demonstrate that cell irradiation, which promotes VEGFC and lymphangiogenesis genes, concomitantly tends to increase the epithelial phenotype of Daoy and HD-MB03 cells, thus reducing the ability of cells to disseminate 50 . Hence, as already shown in thyroid carcinoma for PROX1 33 , we propose that lymphangiogenesis genes have a double effect in MB. In a cancer seldom metastatic such as SHH MB, lymphangiogenic genes exert anti-tumoral effects: lymphangiogenesis paves the way for tumor cell destruction by the immune system. In a more aggressive tumor, such as Group 3 MB, in which cells grow rapidly, the higher number of cells rapidly overwhelms the immune system. Neo-lymphangiogenesis and rerouting of lymphatic vessels (synthesis of collateral lymphatic vessels) thus occurs upon vessel occlusion due to high tumor cell density 9,11 , hence participating in tumor cell propagation towards distant sites. Consistent with these hypotheses; (i) SHH cells and tumors grow slowly, thus keeping low the number of cells to get rid of by the immune system; (ii) SHH cells form pseudo-vessels in basal conditions, suggesting that transdifferentiation into lymphatic vessels is possible and give way to immune cells; (iii) knocking-out VEGFC gene in SHH cells results in increased aggressiveness. The opposite observations were made concerning Group 3 MB cells: although containing low amounts of lymphangiogenesis genes, they proliferate quickly and give rise to fast-growing tumors. They are able to form pseudo-vessels in basal conditions but overexpression of VEGFC reduces cell aggressiveness. We conclude that, unexpectedly, lymphangiogenesis genes, especially VEGFC, negatively regulate MB tumor cell aggressiveness.
Within the markers of tumor "aggressiveness", we used the new concept of lymphangiogenic mimicry or "lymphomimicry" 51 . For the first time to our knowledge in MB, we showed that cells of three tumor models (Daoy, ONS-76 and HD-MB03) are able to organize into pseudo-vessels on matrigel. However, while Daoy-and probably ONS-76-vessel formation is highly dependent upon VEGFC, HD-MB03 cells, which do not express VEGFC, nevertheless organize into a VEGFC-free type of pseudovessels. We hypothesize that Daoy and ONS-76 cells form pseudo-lymphatic vessels able to hybridize with real lymphatic vessels. HD-MB03 cells transdifferentiate into pseudo-blood vessels, hence hybridizing with blood vessels and generating a different path for cell escape and metastasis. This hypothesis was partially validated by our in vivo experiments, showing that Daoy-derived tumors stay very white, with few visible blood vessels, while HD-MB03-derived tumors are extremely reddish (angiogenic, and without any PDPN labeling).
MB are usually described as angiogenic tumors 38,52 . However, as mentioned above, it is important to consider MB as a group of several pathologies inasmuch as the sub-grouping is a source of heterogeneity. Tumors derived from SHH cells show little apparent vascularization, which is part of the explanation for the small size (low infusion of O 2 and nutriments toward the tumor) and low aggressiveness of the tumors. If any, metastasis only occurs via the lymphatic system. Conversely, tumors from Group 3 are highly vascularized. Although the vasculature is abundant, it is of poor quality 53 , thus preventing adequate doses of treatment to be provided to the tumor.
Secretion of VEGFC, expression of its receptors on tumor cells and the resulting autocrine/paracrine effect of this cytokine, are major determinants of MB tumor size. It is difficult to be categorical whether VEGFC effect is autocrine or paracrine in our models, which are multiple and for which the cells bear several types of receptors, each of them acting differently. However, tumors are not homogeneous, and most authors describe the socalled "autocrine" effect of VEGFC as the effect generated locally, on a specific type of cells-especially tumor cells-bearing the receptors, by the VEGFC secreted by this very type of cells. Paracrine effect is described as an effect occurring when VEGFC is transported by vessels (lymphatics) towards more distant targets [54][55][56] . To ascertain the autocrine effect of VEGFC, internalization of tagged-VEGFC receptors might be implemented. However, it is probable that this effect might be compensated by the other types of receptors (NRP2, CD146).
In two different models of MB, high amounts of VEGFC/ VEGFC receptors correlate with lower tumor size and lower migration of tumor cells. Our hypothesis is that the initial effect of VEGFC is to reduce MB cell proliferation/migration, thus keeping the tumor in a state where it can more easily be attacked and destroyed by the immune cells. Later on, when this system gets saturated, or for a fast-growing tumor such as Group 3 tumors, overwhelming the system leads to synthesis of more lymphatic vessels and to metastasis.
Thus, an efficient MB treatment must take into account: (i) the differences between subgroups and (ii) the time of administration.
Daoy cells have an epithelial phenotype 31 , while HD-MB03 cells are more mesenchymal. Moreover, HD-MB03 are CD133 + cells 57 , which is associated with increased metastasis and poor outcome of patients 57,58 . In vivo, HD-MB03 cells gave rise to rapid-growing tumors. This high rate of tumor growth was reduced by VEGFC expression by the cells. The same result was achieved in vitro, by exogenous addition of VEGFC to the HD-MB03 cells. This observation was surprising. It is possible to consider VEGFC as a treatment for Group 3 MB, under certain conditions. This provocative suggestion needs to be reinforced by more in vivo experiments in immunocompetent mice, but this is the first step approaching a potential treatment for the aggressive Group 3 MB.
Our results seem inconsistent with gene expression in patients. This demonstrates the complexity of a whole tumor, the role of the tumor microenvironment, blood, lymphatic and immune systems and of phenomena such as hypoxia, which are not totally understood yet [59][60][61] . The effect of tumor microenvironment has previously been demonstrated in the lab. Indeed, VEGFC knockout has opposite effects on tumor growth in immunodeficient or immunocompetent mice 37 .
In conclusion, unexpectedly, the common thread between the different MB subgroups is the downregulation of tumor cell aggressiveness by VEGFC. We infer that this phenomenon is involved in the early regulation of tumor development by the immune system, both by maintaining tumors at a small size and by generating the tumor lymphatic vessels conveying these small tumors towards lymph nodes, where they are degraded. These observations pave the path for the development of new therapeutic strategies based on combined treatment between VEGFC and immune checkpoint inhibitors, in MB patients. VEGFC KO clones: The VEGFC gene was knocked-out in wild type Daoy cells (WT-Daoy) by the CRISPR-Cas9 technique 62 . Briefly, a human VEGFC target oligonucleotide (5′-GAGTCATGAGTTCATCTACAC-3′) was cloned into the pX330-U6-Chimeric_BB-CBh-hSpCas9 vector (gift of Dr. Feng Zhang; Addgene plasmid # 42230). Two VEGFC KO clones were obtained by PEI transfection (Tebu Bio, Le-Perray-en-Yvelines, FRANCE) of the resulting vector into Daoy cells and further selection on 5 µg/ml puromycin (InvivoGen, Toulouse, France), for 10-15 days. Control cells (Ctl) were obtained by transfection of WT-Daoy cells by an empty pX330 vector and puromycin selection. The mutations leading to VEGFC invalidation were revealed, for each clone, by genomic DNA sequencing, using the following primers: Sense, 5′-TTGTGTTAGGGAACGGAGCAT-3′; Antisense, 5′-AGAACCAGGCTGGCAACTTC-3′. Clone 1 was homozygous while clone 2 was heterozygous, but in both cases, no VEGFC protein was translated (Supp. Table S2). Effectiveness of VEGFC invalidation was confirmed by ELISA assay of VEGFC production.
X-ray resistant cells: Highly confluent Daoy and HD-MB03 cells (two populations each) were X-ray irradiated every week for 10 weeks with no cell subculturing, using a Faxitron cabinet X-ray irradiator (160kV-6.3 mA; Edimex, Le Plessis-Grammoire, FRANCE). An 8-Gy dose was delivered at each irradiation. After 10 weeks, naive and resistant cells were irradiated once at 8 Gy and cell viability was assessed 5 days later using the ADAM cell counter (MBI, Dorval, CANADA). Viability was close to 80% for both Daoy and HD-MB03 resistant cells, while naive Daoy cells survived at 60% after one such irradiation and naive HD-MB03 at only 40% (Supp. Fig. S6).
All cells were cultured at 37°C in a humidified atmosphere with 5% CO 2 . For all experiments, cell lines were maintained for no more than 2 months.
A list of all the antibodies used in this study is provided in Supplementary  Table S3.
Chemicals. All standard chemicals were purchased from Sigma-Aldrich, unless otherwise stated.
Cell proliferation. 1500 Daoy or ONS-76 cells or 7500 HD-MB03 cells were seeded in six-well plates in triplicates and cells were counted every day for 8 days, using a Coulter counter (Villepinte, FRANCE). The relative number of cells (vs day 1) was assessed daily. Cell growth was fitted to an exponential growth equation: with Y = Cell number at day X; Y0 = Cell number at day 0 (graphically calculated); X = Day following cell plating (cells plated at X0 = 0); k = Exponential growth constant rate. Doubling times were compared.
Enzyme-linked immunosorbent assay (ELISA). Cells were seeded in 12-well plates in triplicates (50, Subcutaneous xenografts. This study was conducted in compliance with the National Charter on the ethics of animal experimentation. Our experiments were approved by the "Comité National Institutionnel d'Éthique pour l'Animal de Laboratoire (CIEPAL)" (reference: NCE/2017-383). 10 6 Daoy cells or 0.5 × 10 6 HD-MB03 cells were injected, in medium containing 50% Corning® matrigel® matrix (VWR), subcutaneously, into the flank of 6-week-old NMRI-Foxn1nu/ Foxn1nu female mice (Janvier Labs, Le Genest-Saint-Isle, France). Tumor volume was measured every other day with a caliper and calculated as follows: where L = tumor length; l = tumor width (in a 2D space, tangential to the mouse skin). The experiment was stopped when the tumors of one experimental group reached a volume of 1000 mm 3 .
Immunohistochemistry and immunofluorescence. Each tumor was fixed in 10% formalin and paraffin embedded. Morphologic examination was performed upon Hematoxylin and Eosin stained sections (3-4 μm).
Five micrometers thick frozen serial sections were fixed 30 min in 4% PFA before immunostaining procedure. Prior to primary antibody application, tissue sections were blocked 60 min in phosphate-buffered saline (PBS) containing 3% bovine serum albumin, 5% horse serum, 5% goat serum and 1% bovine serum albumin (Sigma-Aldrich). Incubation with primary antibodies was carried out overnight at +4°C. Negative controls were left with blocking solution, without primary antibody overnight at +4°C. Incubation with fluorichrome-conjugated (AlexaFluor 488 and AlexaFluor 594) secondary antibodies, which were specific to each primary antibody, was performed for 60 min at room temperature in dark. DAPI was used to visualize nuclei.
Patient sample immunohistochemistry. In collaboration with the Nice (Dr. Fanny Burel-Vandenbos) and Marseille (Dr. Nicolas André) hospitals, we analyzed sections from formalin-fixed and paraffin-embedded MB samples for lymphatic marker labeling, as described elsewhere 19 . Briefly, the samples were incubated at room temperature with monoclonal, primary mouse anti-human PDPN and CD31 antibodies and biotinylated secondary antibodies. Labeling was detected with the diaminobenzidine substrate against a hematoxylin counterstain. An accredited clinical pathologist (Dr. Burel-Vandenbos) evaluated marker expression.
Technical resources. Dr. Steven C. Clifford (Newcastle-upon-Tyne, UK) provided us with a databank of 250 samples of MB, with transcriptomic and clinical data, including overall and progression-free survival 29 . We also compiled and analyzed MB data from the "R2: Genomics Analysis and Visualization Platform" (http://r2.amc.nl) for lymphangiogenesis gene expression.
Statistics and reproducibility. Results were expressed in the text as mean ± SEM of at least three independent experiments. Unless otherwise stated, statistical analyses were performed using one-way or two-way ANOVA tests with Dunnett's multiple comparison tests. For two independent groups, Mann-Whitney analyses were performed. Results were considered significant when p-value < 0.05.
For in vitro experiments, 3 to 5 samples were used to assess a biological effect. Most authors consider that replicating three times the same experiment, taking care that cell culture conditions do not change is enough to accurately describe a biological phenomenon. We increased the number of samples for qPCR experiments, where variability is higher. For in vivo experiments, each group of animals is composed of 8-10 individuals. Our experiments did not require blind tests.
No data were excluded.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The datasets generated during and/or analyzed during the current study are available in the Figshare repository, https://figshare.com/articles/dataset/CommsBio_19-1661A-Raw_data_xlsx/12881351.