K 4 determines cancer cell phenotype by controlling the plasma membrane-associated proteome

How the phenotype of Medulloblastoma (MB) tumor cells adapts in response to growth factor cues is poorly understood. We systematically determined alterations in the plasma membrane (PM)-associated proteome in growth factor-activated MB cells. We found that ligand-induced activation of c-MET receptor tyrosine kinase triggers specific internalization of c-MET and of membrane-associated and transmembrane proteins including nucleoside and ion transporters. In contrast, c-MET activation caused increased PM association of the PVR/CD155 adhesion and immunomodulatory receptor, promoting MB cell motility and tumor cell growth in the cerebellar tissue. Both increased and decreased PM association of a number of these proteins including PVR/CD155 is regulated by the Ser/Thr MAP4K4. We further identified Endophilin A proteins as potential regulators of this process downstream of MAP4K4 to contribute to HGF-induced invasion control. Together, our findings indicate a novel link between MAP4K4 overexpression in MB and the maintenance of a cellular phenotype associated with growth and invasiveness.


Introduction
The plasma membrane (PM) proteome is a dynamic interface of mammalian cells that regulates cellular interactions with the extracellular environment. This physical barrier regulates the transfer of chemical and mechanical information between extracellular and intracellular spaces. The net outcome of the PM barrier functions affects the cellular phenotype and the composition and biophysical structure of the extracellular microenvironment. The PM proteome primarily consists of receptors, cell adhesion molecules, enzymes and ion channels 1 , which define strength and duration of signal transmission and how the cell interacts with its environment. The diversity of proteins associated with the PM and their differential expression, localization or regulation in pathological conditions such as cancers or inflammation not only represents condition-specific biomarkers 2 , but also a druggable repertoire of therapeutic targets to interfere with the oncogenic phenotype.
The PM proteome of a cell is dynamically controlled through membrane turnover via endocytic uptake and exocytic fusion 3,4 . This membrane trafficking is a key mechanism that determines cellular responses to complex environments and enables the uptake of nutrients or extrusion of signal transmitters. In addition, endocytosis and the downstream vesicular trafficking inside the cells control adhesion turnover, transduction, and the duration of mitogenic or mobilizing signals. De-regulation of physiological trafficking in tumor cells may thus contribute to cell transformation 5 , and altered endocytic activity can broadly impact tumor cell growth, invasion and metastasis and viability 6,7 .
Medulloblastoma (MB) is a malignant embryonal neuroepithelial tumor of the cerebellum with a high propensity to disseminate and metastasize within the central nervous system 8 . Several omics studies described the genomic heterogeneity in MB, which is now classified into four subgroups with a total of twelve subtypes 9,10 . Despite this molecular understanding of MB, the current standard therapy still consists of surgical resection of the tumor and craniospinal irradiation followed by chemotherapy 11 .
These anti-cancer therapies can have a devastating impact on the developing brain of the patients and can cause severe long-term side effects such as cognitive or mobility impairments 12 . Thus, novel therapeutic options with phenotype-specific anti-tumor activity are needed to replace or complement current treatments and reduce therapy-associated side effects.
The Ser/Thr kinase MAP4K4 is overexpressed in MB and contributes to the pro-invasive phenotype of MB cells downstream of hepatocyte growth factor (HGF)-cellular mesenchymal-epithelial transition factor (c-MET) stimulation 13,14 . c-MET receptor tyrosine kinase (RTK) activation caused increased membrane dynamics, increased endocytic uptake and the activation and turnover of integrin ß1 adhesion receptor 13 . Integrin activation and internalization of integrins are necessary for cell migration and invasion 15,16 . Internalization of activated RTKs and integrins furthermore modulates pathway activation at endosomal compartments 17,18 , thereby adding a layer of regulatory impact of endocytic activity on tumor cell behavior. How the PM proteome of MB cells is regulated in response to RTK activation and potential therapeutic targets in the dynamic PM compartment are not known.
Endophilins are members of the BAR (Bin/amphiphysin/Rvs) domain superfamily that play an essential role in cellular processes like synaptic vesicle endocytosis, receptor trafficking, apoptosis and mitochondrial network dynamics 19,20 . Dysregulation of these processes are associated with cancer or neurodegenerative diseases 19,21 . Endophilins were originally considered a peripheral component of clathrin-mediated endocytosis 22,23 . A novel clathrin-independent endocytic pathway based on endophilin A subfamily activity, termed fast endophilin mediated endocytosis (FEME) was described recently 21 , and several studies also implicated endophilins in cell motility control [24][25][26] . The endophilin A1 is almost exclusively expressed in brain tissue 27 Herein we used a spatial proteomics approach 28 and determined the PM-associated proteome of resting and RTK-activated MB cells. We furthermore addressed whether the pro-oncogenic activity of MAP4K4 could be coupled to RTK-induced, dynamic alterations in the PM proteome. This analysis allowed the identification of PM-associated proteins subjected to c-MET-induced spatial re-distribution and the identification of MAP4K4 as a driver of this process.

HGF-c-MET signaling promotes PM proteome reorganization in medulloblastoma cells
To determine changes in PM and PM-proximal association of proteins in response to growth factor stimulation, we established a PM proteome analysis approach for MB cells. We used membraneimpermeable Sulfo-NHS-SS-biotin labeling of intact sgControl (sgCTL, gC) and sgMAP4K4 (MAP4K4 knockout, gM4) MB tumor cells, followed by streptavidin protein capture and high-resolution mass spectrometry analysis (Fig. 1A). To determine whether PM association of protein expression in MB cells is regulated by growth factor receptor activation, we compared starved and HGF-stimulated cells.
Proteins identified were filtered either using the transmembrane (TM) prediction algorithm TMHMM 29 or by subcellular localization annotation 30 . These stringent sorting criteria identified 952 TM proteins and 336 PM-associated proteins, respectively (Tables 1 & 2). In sgCTL cells, HGF stimulation significantly decreased the abundance of a number of TM (Fig. 1B, Table 3) or PM-annotated proteins (Fig. 1C, Table 4). This indicated that either internalization of these proteins in response to c-MET activation is increased, or recycling decreased. We previously found that internalization of activated c-MET depends on MAP4K4 13 . Therefore, we tested whether MAP4K4 is required for reduced PM association in cells,  Table 3) or PM-associated proteins (Fig. 1C, Table 4) after c-MET activation, suggesting a regulatory function of MAP4K4 in PM localization of these proteins. A paradigmatic example is c-MET itself, which is internalized in a MAP4K4-dependent manner in DAOY cells 14 and detected at lower levels in the PM proteome in response to HGF ligation (Fig. 1D). MAP4K4 is necessary for this process as depletion of MAP4K4 significantly increases c-MET abundance in HGFstimulated cells (Fig. 1D). None of the five additional RTKs detected by mass-spectrometry displayed similar behavior, suggesting that RTK internalization either requires the precedent activation of the receptor kinase domain or is specific for c-MET. We concluded from this experiment that c-MET activation triggers the turnover of PM-associated proteins and that MAP4K4 is indispensable for the turnover of a number of those proteins. Gene ontology analysis of TM-associated proteins revealed MAP4K4 dependencies in several pathways controlling endo-and exocytotic activities of cells as well as immune cell recognition or solute exchange (Fig. 1E). It also revealed that MAP4K4 depletion in HGF-stimulated cells increased PM association of proteins with reported implication in chemotherapy sensitivity/resistance (SLC29A1, SLC4A7, LRRC8) 31,32 . Conversely, MAP4K4 depletion decreased PM association of proteins involved in tumor immune evasion (CD155, CD276, HLA-G) [33][34][35] , in solute influx regulation (CLIC1, SLCs, ABCCs) 36,37 and cell migration (CD155, CD276) 33,38 (Fig. 1F).
Taken together, our data suggested that MAP4K4 is involved in mediating HGF-c-MET induced turnover of PM-associated proteins. However, MAP4K4 also enables the maintenance of some transporters and receptors such as CLIC1, CD155, and CD276 at the PM after MET activation (Fig.   1F). In conclusion, we found that MET activation by HGF triggers reorganization of the PM-associated proteome in MB cells and that MAP4K4 selectively controls this turnover.

Depletion of MAP4K4 moderately decreases sensitivity to Lomustine
Membrane transporters and channels are proteins involved in the transfer of ions, nutrients and drugs across the PM. Dysregulation of influx or efflux mechanisms can lead to both chemotherapy sensitivity or resistance in cancer cells 39,40 . We detected altered PM association of transporters including SLC29A1, SLC4A7, ABCC1, ABCB1 and LRRC8A (Fig. S2A), where contribution to the cytotoxic drug influx/efflux has been proposed. SLC29A1 for example mediates the uptake of Lomustine LRRC8A is a bi-directional transporter and mediates influx/efflux of drugs such as Cisplatin and Temozolomide 32,43 . In our experimental settings, HGF stimulation reduced PM association of SLC29A1, SLC4A7 and LRRC8A in sgCTL cells (Fig. S2A). Reduced PM association depends on MAP4K4 as plasma membrane association of these transporters is increased in sgMAP4K4 cells stimulated with HGF ( Fig. S2A). MAP4K4 depletion also causes a significant increase in PM-association of ABCB1 and ABCC1 in HGF-stimulated cells (Fig. S2A). These differences in PM association are not the consequences of altered regulation of gene expression (Fig. S1B). Thus, these results suggested a potential implication of HGF stimulation and MAP4K4 expression in the influx/efflux of cytotoxic drugs.
To test this possibility, we generated dose-response curves for Lomustine 44 and Etoposide 45 in DAOY cells. Under normal growth conditions (10%FBS), depletion of MAP4K4 did not significantly impact the sensitivity of DAOY cells to Etoposide compared to control cells (IC50 value of 3.2 µM and 2.8 µM respectively) but slightly increased resistance to Lomustine (IC50 value of 46.2 µM and 39.5 µM respectively) (Fig. S2B). To investigate the effect of HGF stimulation on drug response, cells were first maintained in low serum media for 24h (1% FBS) and then treated for 48h with HGF in combination with Lomustine or Etoposide. Under these conditions, depletion of MAP4K4 moderately reduced Lomustine sensitivity (Fig. S2C). We observed no impact on the response to Etoposide treatment. HGF stimulation did not affect the sensitivity of the cells to the two drugs, independent of the MAP4K4 expression level. Taken together, these results suggest a moderately sensitizing effect of MAP4K4 on tumor cell sensitivity to Lomustine.

MAP4K4 controls ion homeostasis and modulates PM localization of potassium channels
To better understand which protein functions are particularly affected by HGF stimulation and/or MAP4K4 depletion, we annotated the filtered proteins using Protein Atlas 46,47 . We found that 223 of 952 transmembrane proteins (Fig. S3A) and 84 of 336 plasma membrane proteins (Fig. S3B) are annotated as transporters. These trans-membrane proteins mediate the transfer of ions, small molecules, and macromolecules across membranes 48 , thus suggesting the potential implication of the HGF-c-MET-MAP4K4 axis in MB ion homeostasis.
Specifically, HGF-dependent alterations in PM association of ion channels, transporters and exchangers (Fig. S3C), including CLIC1 (Clchannel), CLCN7 (Clchannel), SLC12A7 (K + /Cltransporter), ATP1B3 (Na+/K+ exchanger) and TMEM165 (H + /Ca2 + /Mn2 + exchanger) depends on MAP4K4 expression ( Fig. 2A). Depletion of MAP4K4 in HGF-stimulated cells either increases or decreases their PM association compared to HGF-treated control cells ( Fig. 2A and S3C). Additionally, pathway enrichment analysis performed on the significantly regulated transmembrane predicted proteins between HGF-stimulated sgMAP4K4 KO cells and control cells revealed an enrichment of TM proteins belonging to the gene ontology for potassium ion homeostasis (Fig. 2B). These proteins include several solute carriers (SLCs) and ATPases. HGF stimulation of sgCTL cells reduced the abundance of these proteins at the PM (Fig. 2C). Depletion of MAP4K4 maintained PM association of most of these SLCs and ATPases in HGF-stimulated cells or resulted even in increased PM association compared to unstimulated cells (Fig. 2C).
To access the functional significance of the HGF-and MAP4K4-dependent modulation of PM localization of ion channels on MB cell ion homeostasis, we performed whole-cell patch clamp measurement of the electrical properties of DAOY cells with or without MAP4K4 expression. In starved conditions, MAP4K4 depletion lead to an increased whole-cell current compared to control cells (Fig.   2D). Quantification of the resting membrane potential (Vmem) indicated that HGF stimulation forced the hyperpolarization of MAP4K4-depleted cells compared to control cells (Fig. 2E, filled circles). This last finding is in accordance with the hypothesis that disseminating cancer cells present a more depolarized Vmem and that their hyperpolarization leads to loss of their metastatic potential 49 .
One channel with a significantly altered PM association is the chloride intracellular channel protein 1 (CLIC1). CLIC1 contributes to MB proliferation in vitro, and CLIC1 depletion in MB tumor cell lines significantly reduces tumor cell growth in vivo 37 . Gene expression analysis of several MB patient sample cohorts shows CLIC1 overexpression in the tumor samples compared to control tissues (Fig.   2F). Importantly, this elevated CLIC1 expression is associated with worse outcomes in SHH-alpha and Gr4 MB tumors (Fig. 2G). In control cells, PM association of CLIC1 is moderately but not significantly increased by HGF stimulation. In contrast, HGF stimulation of MAP4K4 depleted cells caused a significant drop of CLIC1 PM association (Fig 2A), suggesting that MAP4K4 is either necessary for transporting CLIC1 to the plasma membrane or for CLIC1 maintenance at the PM in growth factorstimulated cells.
Taken together, these data indicated the implication of MAP4K4 in the control of electrophysiological parameters of MB cells. However, the relevance on cell migration or tumor progression remains to be investigated.

HGF-controlled CD155 expression promotes medulloblastoma cell invasion and motility
We found differential PM association of the adhesion and immune-modulatory proteins CD155 and CD276 (Fig. 1F). CD276 impacts colorectal cancer cell migration through the Jak2/Stat3/MMP-9 signaling pathway 38 . CD155 interacts with integrins or RTKs and increases FAK, Ras, or RAP1 downstream signaling, leading to increased cell motile capabilities 50 . CD155, also referred to as poliovirus receptor (PVR), is a member of the nectin-like family of proteins. CD155 is involved in cell adhesion, migration, proliferation, or immune regulation 33 , and it is often upregulated in tumors compared to healthy tissues 50 . This renders CD155 particularly interesting in the context of tumor progression, where it could contribute to immune evasion and dissemination.
Our data indicated that the PM association of CD155 is moderately increased in HGFstimulated cells. This increase in PM association requires MAP4K4, as depletion of MAP4K4 in HGFstimulated cells causes a significant reduction of CD155 PM association (Fig. 3A). As we did not measure a significant reduction in CD155 mRNA expression in sgMAP4K4 cells, reduced PM association of CD155 in sgMAP4K4 cells stimulated with HGF is not the result of transcriptional repression. Furthermore, HGF stimulation increased mRNA expression of CD155 both in sgCTL and sgMAP4K4 cells by 10 -20% (Fig. 3B). Thus, the significant decrease in CD155 PM association in HGF-stimulated sgMAP4K4 cells is not the consequence of transcriptional repression. To validate MAP4K4 implication in PM association of CD155, we determined CD155 surface expression levels by flow cytometry analysis. HGF stimulation increased surface localization of CD155 in sgCTL cells by 20% (Fig. 3C). Depletion of MAP4K4 not only prevented increased CD155 surface expression after HGF-stimulation but even caused a 10 -20% reduction of surface-expressed CD155 relative to unstimulated controls cells (Fig. 3C). This suggested that HGF triggers increased surface expression of CD155 and that MAP4K4 is either required for the transfer or the maintenance of CD155 at the PM under conditions that trigger membrane dynamics and endocytic turnover.
We hypothesized that reduced PM association of CD155 and CD276 in sgMAP4K4 cells could be causative for the reduced MB cell dissemination observed with these cells 13,14 . To test this hypothesis, we measured the impact of reduced expression of CD155 or CD276 on HGF-induced collagen I invasion. We found that depletion of CD155, but not of CD276, prevented HGF-induced collagen I invasion (Fig. 3D). To corroborate the implication of CD155 in migration control with another cell line, we tested the effect of CD155 depletion on collagen I invasion in ONS-76 cells. These cells invade collagen I independently of exogenous growth factor stimulation. Depletion of CD155 significantly reduced collagen I invasion of ONS-76 cells (Fig. 3E). The reduced invasion could be a consequence of reduced speed of migration. To test this possibility, we performed single-cell motility assays with ONS-76 cells seeded in collagen I coated plates. Cell movements were recorded by timelapse video microscopy for five hours. CD155 depletion reduced the average speed of ONS-76 cells by 50% ( Fig 3F). Phenotypically, cells with decreased CD155 expression displayed increased circularity, increased area and increased width to length ratio ( Fig 3F, movies 1 and 2).
Taken together, these data identified the CD155 cell adhesion molecules as a promotor of cell motility and invasiveness in SHH MB cell models.

CD155 is necessary for ex vivo cerebellar tissue infiltration
To assess whether CD155 depletion prevented cerebellar tissue infiltration of MB cells, we implanted ONS-76-LA-EGFP tumor spheroids onto cerebellar slices and determined growth and invasion after tumor-cerebellar slice co-culture 51 . After 48 hours of co-culture, ONS-76-LA-EGFP cells transfected with siCTL displayed sheet-like infiltration of the surrounding brain tissue (Fig. 3G). In contrast, spheroids of cells transfected with siCD155 did not display an invasive behavior and remained clustered at the site of initial spheroid positioning. As an indirect measurand for tumor growth and invasion, we quantified the volume of the tumor cell clusters. We observed a significant (~2-fold) reduction of the cluster volume in siCD155 transfected compared to siCTL cells (Fig. 3G, S4). Some cells in the siCD155 condition still display an invasive phenotype, possibly due to variations in transfection efficiency at single-cell level. We also evaluated tumor cell proliferation in the tissueembedded tumor cell clusters using EdU incorporation. We found that depletion of CD155 caused a marked decrease in the number of proliferative cells in the ex vivo model (Fig. 3G, S4).
Together, these findings suggest that the cell adhesion receptor CD155 is critical for effective tissue invasion and growth in SHH MB cells.

Expression of MAP4K4 and of the endocytosis effector EndoA1 is increased in SHH MB
MAP4K4 is overexpressed in MB patient tumor samples 52 and gene expression profiling of 763 primary MB samples 53 revealed that MAP4K4 expression correlates positively with genes involved in endocytosis control (Fig. S5A). Interestingly, the strongest positive correlate with MAP4K4 expression among the endocytosis-regulating genes is SH3GL2 (EndoA1, r = 0.501), while expression of both CLTB (r = -0.228) and CTLC (r = -0.228) correlate negatively with MAP4K4 (Fig. S5B). The highest positive correlation between SH3GL2 and MAP4K4 expression is observed in SHH MB tumors, where also CLTC and SH3GL2 expression correlate negatively (r=-0.144) (Fig. S5C). These correlation analyses indicated that the decrease in cell migration observed in MAP4K4-depleted cells 52 could be mechanistically linked to a clathrin-independent mechanism of endocytosis. Increased levels of MAP4K4, SH3GL2 and SH3GL3 -both at protein and RNA levels -but not of SH3GL1, were also observed in MB and ganglioglioma samples of a recent genomic and proteomic analysis of a cohort of 218 pediatric brain tumors 54 (Fig. S5D). These findings revealed high expression of endophilins in MB compared to other central nervous system tumors and indicated the potential importance of fastendophilin-mediated endocytosis (FEME 21 ) in this tumor type.

EndoA family proteins are involved in HGF-induced MB cell migration
To explore endocytic pathways (Fig. S6A) involved in HGF-induced migration control more broadly, we determined investigated which endocytic pathways are necessary for maintaining the HGFinduced invasive phenotype. We used 25 siRNAs targeting 16 key components of 8 different endocytic pathway (Fig. S6B) to identify relevant components. Downregulation efficacy of the siRNAs was confirmed by qRT-PCR (Fig. S6C). The Boyden Chamber transwell migration assay was used to quantify siRNA impact on HGF induced transwell motility (Fig. 4A). Consistent with previous studies 13, 52 , HGF stimulation increased the number of transmigrated cells (~ 2.7-fold increase). Depletion of dynamins 1 and 2 (DNM1 and DNM2) abrogated HGF-induced migration (Fig. 4A), confirming a general role of endocytosis in migration control. Interestingly, depletion of Endophilin A proteins completely abrogated (SH3GL2, EndoA1) or drastically reduced (SH3GL1, EndoA2 or SH3GL3, EndoA3) HGFinduced cell migration (Fig. 4A). In contrast, depletion of clathrin or caveolin, two essential components of endocytic pathways for GF signaling 55 , did not prevent or reduce HGF-induced DAOY cell migration.
Similarly, depletion of two regulators of micropinocytosis, CTBP1 and RAC1 56 , did also not block HGFinduced migration. Both non-transfected and siCTL transfected cells showed similar levels of proliferation up to 72 hours, while depletion of SH3GL1/2/3 and DNM1 exhibited even moderately increased levels of proliferation (Fig. S6D). These data exclude a proliferation defect as underlying cause of the smaller number of transmigrated cells counted in the siSH3GL1/2/3-or siDNM1transfected cells and indicate the implication of EndoA1 and DNM1 in migration control.

Endophilin A1 is required for 3D collagen I matrix invasion
We next tested whether depletion of EndoA1, DNM1, or clathrin heavy chain (CLTC) also affects 3D collagen I invasiveness using the spheroid invasion assay (SIA) 57,58 (Fig. 4B, S6E). HGF caused a robust increase in the number of cells disseminating into the collagen I matrix and increased the distance of invasion (~ 6.8-fold increase compared to untreated control). Depletion of EndoA1 or DNM1 significantly reduced HGF-induced collagen I invasion, whereas depletion of CLTC had no effect, indicating that EndoA1 functions in clathrin-independent endocytosis. Consistent with a conserved role of EndoA1 in migration control, depletion of EndoA1 also reduced (~ 1.6-fold decrease) collagen I invasion of ONS-76 cells (Fig. 4C). Depletion of MAP4K4 did not affect expression of EndoA1 (Fig.   S6F). To determine whether a compensatory mechanism mediated by EndoA2 or EndoA3 24 may be active in our cell model, we depleted all Endophilin-A proteins simultaneously (TKD). We found that endophilin TKD abrogates HGF-induced cell migration in DAOY cells moderately more effective than siEndoA1 alone (Fig. 4D). We concluded that EndoA2 or EndoA3 contribute to migration control, similar to what was observed in other cell models 24 , and that EndoA1 is the most relevant Endophilin in our cell model. In conclusion, we found that EndoA1 mediates a migratory phenotype in MB cells, possibly through a mechanism involving clathrin-independent endocytosis.

Endophilin-A1 depletion reduces tissue invasion
We next assessed the role of EndoA1 in tumor growth and invasion control in the tissue context with tumor-cerebellar slice co-culture experiments using ONS-76-LA-EGFP cells. Depletion of EndoA1 markedly reduced cerebellum tissue infiltration and prevented the formation of streams of invading cells, which are highly characteristic for this type of MB tumor cell (Fig. 4E, panel a). Individual EndoA1depleted tumor cells still displayed invasion at the border of the spheres. However, the morphology of these cells appeared more rounded, less mesenchymal compared to control cells (Fig. 4, panel b).
Although morphology of the collectivity of the invading cells varied between samples, the considerably reduced invasiveness of EndoA1-depleted cells was apparent, while proliferation did not seem to be significantly affected in vitro (Fig. S6D) and ex vivo (Figs 4E, S6G). Taken together, these data indicate that EndoA1 contributes to the invasive behavior of the tumor cells in the cerebellar tissue without affecting proliferation.

MAP4K4 controls subcellular localization of CIP4
Disturbance of vesicle trafficking is expected when endocytic processes are targeted. We therefore determined whether EndoA1-or MAP4K4-depleted cells display altered subcellular abundance and distribution of early and late endosome markers compared to control cells using anti-EEA1 and anti-LAMP1 antibodies, respectively (Fig. S7). Depletion of EndoA1 and MAP4K4 led to a 20% reduction of EEA1 puncta per cell (Fig. S7A), indicating that both EndoA1 and MAP4K4 enable formation of early endosomes. Depletion of MAP4K4 but not of EndoA1 also reduced the number of LAMP1 puncta per cell by 15% (Fig. S7B). Thus, in addition to contributing to early steps of endocytosis, To test whether MAP4K4 could be involved in PM localization of FEME components, we searched the proteins detected in the surface proteome (Tables 1 & 2) for CIN85/SH3KBP1, CIP4 and FBP17, three regulators of FEME priming 59 (Fig. 5A). This experiment suggested that HGF stimulation caused a moderately increased PM association of SH3KBP1/CIN85, CIP4, FNBP1/FBP17 and EndoA2 CIN85/SH3KBP1, CIP4 and FNBP1/FBP17 are considered critical mediators of FEME by priming the PM for endophilin A clustering 59 . PM enrichment of these proteins in a MAP4K4-dependent manner in HGF-stimulated cells could indicate a link between MAP4K4 and FEME priming. Interestingly, we also detected the potential interaction of SH3KBP1/CIN85 with MAP4K4 in an unrelated proteomic interaction analysis 60 . Due to their cytoplasmic localization, we further tested whether MAP4K4 depletion alters subcellular localization of CIP4. We found that depletion of EndoA1 and MAP4K4 increased cortical localization of CIP4 in circular patches near cellular protrusions (Fig. 5B).
These patches were not apparent in siEndoA1 and siMAP4K4-transfected cells (Fig. S7A). CIP4 is associated with microtubule-like structures 61 , and consistently we observed a filamentous pattern of CIP4 staining that does not co-localize with F-actin (Fig. 5C), a known target of MAP4K4 activity 13,52 .
These results indicate the implication of MAP4K4 in membrane recruitment, cortical localization and organization of key regulators of FEME, thereby possibly facilitating an effective FEME sequence 59 ( Fig. 5D).

Discussion
We found that HGF-induced c-MET activation caused significant alterations in the abundance of transmembrane and plasma membrane-associated proteins. The abundance of a majority of these proteins is reduced by HGF-c-MET activation, indicating either GF-dependent internalization or attenuated recycling of PM-associated proteins. We found that PM association of some of these proteins is MAP4K4 dependent. MAP4K4 control of PVR/CD155 PM association is particularly intriguing as we also found that PVR/CD155 is necessary for efficient tumor cell migration and brain tissue invasion. Our data furthermore point towards a role of fast-endophilin-mediated endocytosis in migration control through the EndoA protein family and provide a conceptual basis for novel strategies to target the oncogenic phenotype of MB and other solid tumors.
PM-associated proteins were enriched by biotin labeling prior to mass-spectrometry analysis.
This methodology is expected to label predominately extracellularly bound or anchored proteins, transmembrane proteins, and proteins associated with endocytic vesicles. To exclude non-PM proteins, we either used the TMHMM algorithm to predict transmembrane domains in proteins 29 or a data-based filtering approach, yielding two datasets. Former is highly efficient to detect surface proteins. However, it also predicts transmembrane domains of proteins associated with other membrane compartments such as the golgi or ER membranes. The data-based filtering approach selects proteins based on published subcellular localization (detailed annotations available in Table 2). This approach is not specifically enriching for proteins with transmembrane domains, and it may include non-PM proteins complexed with PM-associated proteins. The TMHMM and data-based annotation approach selected 21% and 8% of all proteins detected in the initial MS analyses, respectively. The TMHMM approach was highly efficient for determining MAP4K4 impact on PM-association of transporters and channels, which all contain a TM domain. The data-based annotation approach is highly stringent, and it allows for the comparative analysis of the variations in PM association in all the conditions tested using pathway enrichment analysis. Therefore, we used latter for most of the subsequent analyses.
MAP4K4 implication in HGF-induced internalization of membrane transport proteins and drug transporters 31 (SLC29A1, ABCC1, ABCB1, LRCC8) indicated a potential therapeutic benefit of targeting this function of MAP4K4 during chemotherapy, which remains one of the first line treatments for cancer patients. We hypothesized that indirect modulation of drug up-take or extrusion mechanisms could improve drug efficacy. MAP4K4 depletion moderately decreased the susceptibility of DAOY cells to Lomustine, but it had no effect on etoposide treatment efficacy. Although of potential interest, our data are inconclusive and additional chemotherapeutic drugs and cell lines should be tested to identify specific susceptibilities. Disturbed cell-surface localization of ion channels and transporters could furthermore modify the overall ion homeostasis of the cell. The changes in PM association of ion channels we observed is associated with subtle alterations in electrical currents. This is consistent with previous studies that demonstrated altered electrical conductance properties in tumors compared to healthy tissues 37,62,63 . We identified CLIC1, CLIC4, SLC29A1 or ATP1A3 among others as channels susceptible to HGF or MAP4K4 regulation. Increased current in MAP4K4-depleted cells indicates MAP4K4 repression of ion channel distribution or activity that could control ion homoeostasis of the tumor cells. The lowered potential in MAP4K4-depleted cells after HGF stimulation could be the consequence of a reduced influx of cations, increased efflux of cations or increased influx of anions.
Pathway enrichment analysis of PM associated channels pointed towards an altered potassium homeostasis. We observed increased PM-association of potassium channels and potassium transportassociated molecules including the P-type cation transporter ATP12A, the Ca 2+dependent channels ATP1A3 and KCNMA1 in MAP4K4-depleted cells after HGF stimulation. Thus, MAP4K4 could maintain optimal polarization in HGF-stimulated cells by controlling potassium efflux rates through regulating PM-association of K+ transporters. However, whether differential PM association of potassium channels and de-regulated potassium fluxes are at the origin of the electrical disbalance observed in MAP4K4-depleted cells and contribute to the MAP4K4-associated pro-migratory phenotype remains to remains to be determined.
We observed reduced PM-association of CD155 and CD276 in MAP4K4-depleted cells Depletion of CD155 in these cells triggered mesenchymal to epithelial transition and repressed migration and invasion in vitro and in vivo., suggesting that CD155 expression in MB cells could also contribute to migration control indirectly via transcriptional reprogramming. CD155 also represents an attractive therapeutic target due to its abilities to inhibit NK cell and CD8 + T-cell activity, thereby contributing to immune evasion of tumor cells 66,67 . Kinase-controlled modulation of CD155 surface expression as we found herein enable this process as rather the abundance -and not the absolute absence or presence -of CD155 at the surface of the cell is critical for its immunomodulatory capabilities 50 . Under physiological conditions, CD155 is expressed at low levels, where a balance between its immune activating and inhibitory functions maintains the normal function of immune cells.
Thus, the MAP4K4-dependent increased surface expression of CD155 on growth factor-activated cells could not only affect the migratory potential of the cells but also locally (in the tumor microenvironment) repress immune system activation.
Our findings point towards a role of fast-endophilin mediated endocytosis (FEME) through EndoA1 in the control of the invasive phenotype of these cells. EndoA1 and A3 are highly expressed in SHH MB compared to other brain tumors and expression correlates positively with MAP4K4 in this tumor type. This contrasts with Clathrin, which correlates negatively with MAP4K4 in SHH MB and with the lack of clear evidence that clathrin-mediated endocytosis is necessary for migration control in our cell models. The role of endophilins in cancer progression is complex and controversial, as endophilins can both promote and reduce cancer cell migration, depending on tumor type 24,59,68 . We previously found that 5-(N-ethyl-N-isopropyl)amiloride (EIPA), a micropinocytosis inhibitor 69 , blocks HGF-induced invasion 52 . However, depletion of RAC1 or CTBP1, two upstream activators of micropinocytosis, did not significantly impact HGF-induced motility. This suggests that the pronounced effect of EIPA could be due to the endocytosis-unrelated effects of this drug on ion transport, intracellular pH and cytoskeleton regulation, thus affecting FEME indirectly 24,59 . FEME is involved in the endocytosis of activated receptors at the leading edge of migrating cells 70 and fast and ultrafast endocytosis are important for chemotaxis 71 . FEME could be involved in the internalization of c-MET, thereby controlling signal transmission and persistence of this receptor 24 . Our study identified a possible link between SH3KBP1, CIP4 and FBP17. We recently identified a potential interaction of MAP4K4 with SH3KBP1using a Bio-ID based approach 60 , and in this study, we found that MAP4K4 regulates CIP4 PM-association and its accumulation in cortical patches. The early endosome marker EEA1 also localizes to these patches, and it is thus possible that one aspect of FEME processing towards early endosomes occurs preferentially there. Depletion of MAP4K4 prevented EEA1 recruitment to these patches, led to an overall decrease in EEA1-positive vesicles and caused accumulation of CIP4 in this location instead. These phenomena are quite exactly phenocopied by EndoA1 depletion. It is thus possible that MAP4K4 controls FEME by orchestrating a step between FEME priming and trafficking towards early endosomes. Our working model is that MAP4K4 orchestrates PM association of CIP4 from a non-PM-associated subcellular compartment to the "CIP4 patch" in close proximity of or associated with the PM, where also early endosomes enrich. Upon HGF stimulation, CIP4 translocates to the PM, where it is rapidly turned over through EndoA1-dependent FEME before it is recycled back to the PM. Without MAP4K4 function, CIP4 remains associated with the patches, hence overall amount of PM associated CIP4 increases. Some key questions remain: (i) Does the MAP4K4-dependent mechanism of action involve phosphorylation of one of the components involved in FEME, (ii) is MAP4K4 regulation of CIP4 involved in CIP4-dependent recruitment of Endophilin-A1 to the plasma membrane and (iii) does MAP4K4 control of FEME priming involve the activation of CDC42 downstream of GEFs and GAPs. Elucidation of these mechanisms will be essential to explore therapeutic targeting strategies for specifically repressing MAP4K4 controlled pro-migratory endocytosis.

Cell-surface labelling
DAOY cells were prepared in four T75 flasks to reach a 70% confluency (per condition and replicates).
Standard medium was then replaced with SFM. After 24 hours, cells were 85% confluent, and media were replaced with either pre-warmed SFM or pre-warmed SFM supplemented with 20 ng/mL HGF

Protein identification and label free protein quantification
The acquired raw MS data were processed by MaxQuant (version 1.6.2.3), followed by protein identification using the integrated Andromeda search engine 74 . Spectra were searched against a Swissprot Homo sapiens reference proteome (taxonomy 9606, version from 2016-12-09), concatenated to its reversed decoyed fasta database and common protein contaminants.
Carbamidomethylation of cysteine was set as fixed modification, while methionine oxidation and Nterminal protein acetylation were set as variable. Enzyme specificity was set to trypsin/P allowing a minimal peptide length of 7 amino acids and a maximum of two missed cleavages. MaxQuant Orbitrap default search settings were used. The maximum false discovery rate (FDR) was set to 0.01 for peptides and 0.05 for proteins. Label free quantification was enabled and a 2-minute window for match between runs was applied. In the MaxQuant experimental design template, each file is kept separate in the experimental design to obtain individual quantitative values. Protein fold changes were computed based on Intensity values reported in the proteinGroups.txt file. A set of functions implemented in the R package SRMService 75 was used to filter for proteins with two or more peptides allowing for a maximum of four missing values, and to normalize the data with a modified robust z-score transformation and to compute p-values using the t-test with pooled variance. If all protein measurements are missing in one of the conditions, a pseudo fold change was computed replacing the missing group average by the mean of 10% smallest protein intensities in that condition. For data visualization and normalization across all samples, all group comparison correspond has been processed by statistical analysis of all sgMAP4K4 samples (starved and stimulated) versus all sgCTL samples (starved and stimulated).

Chemotherapy sensitivity analysis
400 DAOY sgCTL or sgMAP4K4 were seeded per well in 384 well microplate and incubated 24 hours at 37°C. Media were either replaced with low-serum media (1% FBS) or with fresh standard culture media (10% FBS) and incubated for another 24 hours at 37°C. Cells were treated with either low-serum media or low-serum media supplemented with 20 ng/mL HGF (PeproTech, 100-39) or by standard culture media supplemented with Lomustine (Selleckchem, S1840, 5 µM to 100 µM) or Etoposide (Selleckchem, S1225, 1 nM to 10 µM). Cell proliferation was quantified after 48 h of incubation using cell proliferation reagent WST-1 (Roche, 11644807001) following the manufacturer instructions.
Sample absorbance was measured against background using a microplate reader (Biotek Instruments, Cytation 3) and statistically analyzed using Prism software (GraphPad).

Primary tumor gene expression analysis
Gene expression data were obtained from the R2 genomics and visualization platform KCl. Compensations and measurements were done within 30 seconds after obtaining whole-cell to avoid drastic changes in cytoplasmic ionic concentration.

Gene annotation and pathway enrichment analysis
Gene annotation and pathway enrichment analysis on the surface-proteomics results were performed using the Metascape webtool (https://metascape.org) developed by Zhou et al. 46 . Transmembrane proteins were annotated using TMHMM prediction algorithm 29 . We performed pathway enrichment using all pathway enrichment categories, standard settings and using all proteins identified by massspectrometry as background genes.  Table S1. After 6 hours, the media were changed, and cells were kept in culture overnight for downstream analyses.

Ex vivo Organotypic Cerebellum Slice Culture (OCSC)
Ex vivo Organotypic Cerebellum Slice Culture was carried out essentially as described previously 56 .
Wild type C57BL/6JRj mice were sacrificed at postnatal day 8-10. Cerebella were dissected and kept in ice-cold Geys balanced salt solution containing kynurenic acid (GBSSK) and then embedded in 2% low melting point agarose gel. 350 µm thick sections were cut using a vibratome (Leica, VT1200S) and transferred on inserts (Merck Millipore, PICM03050) for further in vitro culture. Slices were kept in culture and monitored for 15 days, and media were changed daily for the first week and once in two days thereafter. Spheroids of DAOY-LA-EGFP or ONS-76-LA-EGFP cells transfected with siCTL or CD155 were implanted and then grown in the slices for 48 hours. For HGF stimulation, the feeding medium was supplemented with 20 ng/mL HGF. After treatment, the co-cultured slices were fixed with 4% paraformaldehyde and washed three times with PBS. Inserts were incubated in standard cell culture trypsin EDTA and incubated at 37 • C in a humidified incubator for 23 min. After 3 washes, the slices were blocked in PBS containing 3% fetal calf serum, 3% bovine serum albumin (BSA), and 0.3% Triton×100 for 1 hour at room temperature. Primary anti-Calbindin (Abcam, ab108404, 1:1000) and anti-human nuclei (Merck, MAB4383, 1:250) were diluted in the blocking solution and incubated overnight on a shaker at 4 • C. Proliferative cells were detected using the Click-iT EdU cell proliferation kit following (Invitrogen, C10340). Unbound primary antibody was removed with 3 washes with PBS supplemented with 3% BSA at RT. Secondary antibodies (Table S3) were incubated for 3 h at RT. The inserts were flat mounted in glycergel mounting medium (Dako, C0563). Image acquisition was performed on a SP8 Leica confocal microscope (Leica Microsystems) and analyzed using Imaris (Oxford Instruments) and ImageJ (Fiji) software.

Immunoblotting (IB)
Immunoblot was carried out as described in 14 . DAOY cells were prepared in 6 well plates to reach 70%  (Table S3). GAPDH was used as internal loading control. HRP-linked secondary antibodies (1:2000) were used to detect the primary antibodies. Chemiluminescence detection was carried out using ChemiDoc Touch Gel imaging system (Bio-Rad). The integrated density of detected bands was quantified using Image Lab software (Bio-Rad).

Boyden chamber assay
Transwell with 5 µm pore size (Sarstedt, 83.3932.500) were coated with 0.07 μg/μl Pure Col Collagen I (Advanced BioMatrix, 5005) dissolved in 70% EtOH for a final concentration of 10 μg.cm -2 . 7'500 transfected cells (24 h after transfection) were resuspended in serum-free medium and seeded in the upper chamber of a 24 well plate containing the coated transwell. Medium in the lower chamber was either serum-free medium or serum-free medium supplemented with 20 ng/ml HGF (PeproTech, 100-39). Transwell migration was allowed for 18 hours, and cells were then fixed in 4% paraformaldehyde (Thermo Scientific, 28908) in PBS for 10 min. Fixed cells were stained with DAPI for 15 min. Remaining non-invading cells on the upper surface of the membrane were removed using a cotton swab. Images of DAPI-stained nuclei were acquired using an Axio Observer 2 plus fluorescence microscope (Zeiss) at 5x magnification. Nuclei were counted on the membrane areas using ImageJ software and plotted on Prism 9 software (GraphPad).

Cell proliferation assay
1'000 DAOY cells were seeded in 96 well plates (Greiner Bio-One, 655087). The following day, cells were transfected according to the protocol described above. Six hours after transfection, medium was replaced with serum-free medium. For timepoint 0, WST reagent was added to the corresponding well and incubated 30 min at 37°C. Absorbance at 440 nm was measured using a microplate reader (Biotek Instruments, Cytation 3). 20 ng/ml HGF was added to remaining timepoints and incubated for 24, 48, and 72 hours. At the end of each timepoint, WST reagent was added to corresponding well and absorbance measured. Each day media was replaced with fresh serum-free medium supplemented with HGF. Changes in absorbance over time were analyzed using Prism 9 software (GraphPad).

Immunofluorescence analysis
1'500 transfected cells (24 h after transfection) were seeded in 384 well plate (Greiner Bio-One, 781090) and incubated overnight in normal growth conditions. Cells were starved in low serum condition (1% FBS) for 24 hours and treated with 20 ng/ml HGF for 10-or 30-min. Cells were fixed directly with 4% PFA at 37°C for 20 min. Between all subsequent steps, cells were washed five times with PBS using the 50 TS washer (BioTek). Cells were permeabilized with 0.05% saponin for 20 min at RT. Blocking was performed during 1 hour at RT with 1% FBS. Primary antibody (Table S3) solution was added to sample and shaken at 100 rpm for 2 hours. Secondary antibody (Table S3) solutions were added to sample and shaken at 100 rpm for 2 hours. Nucleic acids were stained with Hoechst-33342 (Sigma Aldrich, B2883, 1:2000 in PBS) for 20 min at RT. Image acquisition was performed using the Operetta CLS High-Content Analysis System (PerkinElmer, HH16000000) at 40x magnification. 9 fields per well with 9 z-planes with 500 nm z-spacing were acquired per site. Maximum intensity projection was computed and used for subsequent analysis. For puncta quantification, EEA1 or LAMP1 signals were segmented using the Harmony software (PerkinElmer) and quantified. Total number of puncta was referred to the number of cells present per field.