Negative regulation of urokinase receptor activity by a GPI-specific phospholipase C in breast cancer cells

The urokinase receptor (uPAR) is a glycosylphosphatidylinositol (GPI)-anchored protein that promotes tissue remodeling, tumor cell adhesion, migration and invasion. uPAR mediates degradation of the extracellular matrix through protease recruitment and enhances cell adhesion, migration and signaling through vitronectin binding and interactions with integrins. Full-length uPAR is released from the cell surface, but the mechanism and significance of uPAR shedding remain obscure. Here we identify transmembrane glycerophosphodiesterase GDE3 as a GPI-specific phospholipase C that cleaves and releases uPAR with consequent loss of function, whereas its homologue GDE2 fails to attack uPAR. GDE3 overexpression depletes uPAR from distinct basolateral membrane domains in breast cancer cells, resulting in a less transformed phenotype, it slows tumor growth in a xenograft model and correlates with prolonged survival in patients. Our results establish GDE3 as a negative regulator of the uPAR signaling network and, furthermore, highlight GPI-anchor hydrolysis as a cell-intrinsic mechanism to alter cell behavior.


Introduction
The urokinase-type plasminogen activator receptor (uPAR) is a central player in a complex signaling network implicated in a variety of remodeling processes, both physiological and pathological, ranging from embryo implantation to wound healing and tumor progression (Boonstra et al., 2011;Ferraris and Sidenius, 2013;Smith and Marshall, 2010). uPAR is a glycosylphosphatidylinositol (GPI)-anchored protein and hence lacks intrinsic signaling capacity. Instead, uPAR acts by binding two major ligands, namely the protease urokinase plasminogen activator (uPA) and the extracellular matrix (ECM) protein vitronectin (Ferraris and Sidenius, 2013;Madsen et al., 2007;Smith and Marshall, 2010). Through uPA binding, uPAR localizes plasmin generation to the cell surface and thereby promotes pericellular proteolysis and ECM degradation (Ferraris and Sidenius, 2013;Smith and Marshall, 2010). In addition, through vitronectin binding and functional interactions with integrins and growth factor receptors, uPAR activates intracellular signaling pathways leading to cytoskeletal reorganization, enhanced cell adhesion and motility and other features of tissue remodeling and cell transformation (Ferraris et al., 2014;Madsen et al., 2007;Smith and Marshall, 2010). As such, uPAR is a master regulator of extracellular proteolysis, cell motility and invasion. uPAR expression is elevated during inflammation and in many human cancers, where it often correlates with poor prognosis, supporting the view that tumor cells hijack the uPAR signaling system to enhance malignancy (Boonstra et al., 2011;Ferraris and Sidenius, 2013;Smith and Marshall, 2010). Increased uPAR expression in solid tumors and the corresponding activated stroma is being evaluated by PET-imaging for patient stratification (Persson et al., 2015).
It has long been known that full-length uPAR is released from the plasma membrane resulting in a soluble form (suPAR) (Pedersen et al., 1993;Ploug et al., 1992), which is detectable in body fluids and considered a marker of disease severity in cancer and other life-threatening disorders (Haupt et al., 2012;Hayek et al., 2016;Shariat et al., 2007;Sidenius et al., 2000;Stephens et al., 1999). Circulating suPAR is derived from activated immune and inflammatory cells (Ferraris and Sidenius, 2013;Smith and Marshall, 2010), and also from circulating tumor cells (Mustjoki et al., 2000).
Locally produced suPAR might function as a ligand scavenger to confer negative feedback on uPAR (Smith and Marshall, 2010). In addition, both uPAR and suPAR can undergo proteolytic fragmentation by uPA and other proteases, possibly leading to new signaling activities (Montuori and Ragno, 2009). Yet, despite decades of research, the mechanism of uPAR release and its physiological implications have been elusive. A GPI-specific phospholipase D (GPI-PLD) (Scallon et al., 1991) has often been assumed to mediate the shedding of GPI-anchored proteins, but this unique PLD does not function on native membranes (Mann et al., 2004).
A possible clue to the mechanism of uPAR release comes from recent studies showing that a member of the glycerophosphodiester phosphodiesterase (GDPD/GDE) family (Corda et al., 2014), termed GDE2, promotes neuronal differentiation by cleaving select GPI-anchored proteins, notably a Notch ligand regulator and heparan sulfate proteoglycans (glypicans) (Matas-Rico et al., 2016;Matas-Rico et al., 2017;Park et al., 2013). GDE2, along with GDE3 and GDE6, belongs to a GDE eLife digest Every process in the body, from how cells divide to how they move around, is tightly regulated. For example, cells only migrate when they receive the correct signals from their environment. These signals are recognised by receptor proteins that sit on the cell surface and connect the outside signal with the cell's response. However, in cancer cells, these processes are out of control, which is why cancer cells can grow very quickly or spread to many different parts of the body.
One important receptor protein is the urokinase receptor, which helps to reorganize the tissue, for example, when wounds heal, but also enables cancer cells to grow and spread. A special feature of urokinase receptor is the way it is connected to the cell surface, namely through a molecule that acts as an anchor, called the GPI anchor. The urokinase receptor and some other GPI-anchored proteins can be released from their anchor. However, until now it was not clear why and how the urokinase receptor is released from cells, or how losing the receptor affects the cell. Now, van Veen, Matas-Rico et al. studied breast cancer cells, and discovered that an enzyme called GDE3 cuts the urokinase receptor off its GPI anchor to release the receptor from the cells. However, when breast cancer cells shed the urokinase receptor, they also lost the receptor from the cell surface in specific areas. As a result, the receptor could not work anymore. When breast cancer cells were experimentally modified to produce high levels of GDE3, the cancer cells became less mobile and aggressive.
Van Veen, Matas-Rico et al. then implanted 'normal' breast cancer cells, and breast cancer cells with extra GDE3 into mice, and observed that the tumors of mice with additional GDE3 grew less quickly. Moreover, breast cancer patients with high levels of GDE3 tend to live longer than patients with low levels of GDE3. These results suggest that the enzyme GDE3 can suppress tumor growth.
These findings uncover a new way how cells can alter their behavior, namely by cleaving GPI anchors at the cell surface. Future experiments will need to address how GDE3 itself is controlled, and if it releases other GPI-anchored proteins from cells. Once we know how to increase GDE3 activity in tumor cells, the new knowledge could one day lead to therapies to help patients with cancer.
Here we identify GDE3 as the first mammalian GPI-specific phospholipase C (GPI-PLC) that cleaves and sheds uPAR with consequent loss of uPAR activities in both HEK293 and breast cancer cells.

Results
GDE3, but not GDE2, sheds uPAR from HEK293 cells We set out to determine whether uPAR can be released by any of the three related GDE family members, GDE2, GDE3 and GDE6. When expressed at relatively low levels in HEK293 cells, human GDE2 and GDE3 (HA-tagged) localized to distinct microdomains at the plasma membrane, possibly representing clustered lipid rafts where GPI-anchored proteins normally reside (Maeda and Kinoshita, 2011), as well as to filopodia-like extensions (Matas-Rico et al., 2016) (Figure 1-figure supplement 1A,B). By contrast, human GDE6 was mainly detected in intracellular compartments and therefore was not further tested ( Figure 1-figure supplement 1A).
To assess GDE activity, we generated stable uPAR-expressing HEK293 cells (HEK-uPAR cells), expressed GDE2 and GDE3 and examined the appearance of suPAR in the medium, using bacterial phospholipase C (PI-PLC) as a positive control (Matas-Rico et al., 2016). Strikingly, uPAR was readily released into the medium by GDE3 and PI-PLC, but not by GDE2 ( Figure 1B). GDE3 competed with exogenous PI-PLC to deplete uPAR, since PI-PLC was much less efficient in GDE3-overexpressing than in control HEK-uPAR cells (Figure 1-figure supplement 1C).

GDE2 versus GDE3: homology modeling
The selectivity of GDE3 versus GDE2 towards uPAR cleavage is striking. In an attempt to understand the structural basis of this selectivity, we constructed homology-based models of the globular a/b barrel GDPD domains, using I-TASSER (Yang et al., 2015). We reasoned that the catalytic domains must recognize not only the PI lipid moiety at the membrane-water interface, but also the attached protein. While GDE2 and GDE3 have a similar putative GPI-binding groove leading to the active site, they show striking differences in their surface charge distribution, particularly at the putative substrate interaction surface ( Figure 1F). It therefore seems likely that surface properties are a major determinant of selective substrate recognition by GDEs, a notion that should be validated by future structural studies. When compared to uPAR-deficient cells, HEK-uPAR cells showed markedly increased cell adhesion, loss of intercellular contacts and enhanced spreading with prominent lamellipodia formation on vitronectin, but not on fibronectin ( Figure 3A), typical features of a Rac-driven motile phenotype, in agreement with previous studies (Kjøller and Hall, 2001;Madsen et al., 2007). Cell spreading coincided with activation of focal adhesion kinase (FAK), indicative of integrin activation ( Figure 3B). Strikingly, expression of GDE3 largely abolished the uPAR-induced phenotypes and cellular responses ( Figure 3A-F). Catalytically dead GDE3(H229A) had no effect, neither had overexpressed GDE2 ( Figure 3F and results not shown). Thus, by releasing uPAR from the plasma membrane, GDE3 suppresses the vitronectin-dependent activities of uPAR. Treatment of diverse cell types with suPAR-enriched conditioned media from HEK293 cells did not evoke detectable cellular responses, supporting the notion that suPAR is biologically inactive, at least under cell culture conditions.

GDE3 locally depletes uPAR and suppresses its activities in breast cancer cells
We next assessed the impact of GDE3 on endogenous uPAR activity in MDA-MB-231 triple-negative breast cancer cells. These cells express relatively high levels of uPAR ( Figure 4A) and its ligand uPA  Figure 4C). To determine how GDE3 expression affects localized uPAR levels at the basolateral plasma membrane, we used confocal and dual-color super-resolution microscopy in TIRF mode. Strikingly, basolateral membrane microdomains containing wild-type GDE3 showed little or no colocalization with endogenous uPAR. In marked contrast, catalytically dead GDE3(H229A) clearly colocalized with uPAR in those membrane domains ( Figure 4D). Quantification of co-localization was performed on multiple cells ( Figure 4D, right panel). These results strongly suggest that active GDE3 depletes uPAR levels from distinct domains at the basolateral membrane. Consistent with this, CRISPR-based knockout of GDE3 resulted in increased basolateral uPAR levels when the cells were plated on vitronectin ( Figure  To confirm that GDE3 acts through uPAR attack, we expressed non-cleavable uPAR-TM to compete out endogenous uPAR, and found that the GDE3-induced reduction of cell spreading on vitronectin was largely inhibited ( Figure 5B). Furthermore, shRNA-mediated knockdown uPAR ( Figure 5D) gave rise to the same phenotype as GDE3 overexpression, namely reduced cell adhesion, spreading and lamellipodia formation on vitronectin ( Figure 5E,F,G).  GDE3 overexpression attenuates the transformed phenotype, slows tumor growth in vivo and correlates with improved survival in breast cancer In long-term assays, wild-type MDA-MB-231 cells showed marked scattering, indicative of increased cell motility with loss of intercellular contacts ( Figure 6A). Again, GDE3 overexpression mimicked uPAR depletion either in greatly reducing both cell motility and clonogenic potential, using either shRNA-mediated knockdown ( Figure 6A Having shown that GDE3 suppresses the non-proteolytic activities of uPAR, we next examined how GDE3 affects uPAR-driven proteolytic matrix degradation by MDA-MB-231 cells. Also in this cell system, GDE3 overexpression mimicked uPAR silicencing in inhibiting the degradation of a gelatin matrix (mixed with vitronectin) in the presence of serum ( Figure 6C,D) (Figure 6-figure supplement 1). On the basis of these results, we conclude that GDE3 attenuates the transformed phenotype of uPAR-positive breast cancer cells through loss of functional uPAR.
When injected into the mammary fat pads of female nude mice, GDE3-overexpressing MDA-MB-231 cells showed diminished tumor growth over time, when compared to empty vector-expressing cells ( Figure 7A), consistent with the cell-based data. However, the full implications of GDE3 expression on uPAR-dependent tumor growth remain to explored in further detail. Finally, in patients, high expression of GDPD2 was found to correlate with prolonged relapse-free survival in breast cancer, particularly in triple-negative (basal-like) subtype patients (N = 618) ( Figure 7B). No such correlation was found for GDE2 (encoded by GDPD5; not shown). This suggests that GDE3/GDPD5 may serve as a marker of clinical outcome in breast cancer.

Discussion
GPI-anchoring is a compIex post-translational modification that anchors select proteins in the outer leaflet of the plasma membrane. Despite decades of research, the biological significance of GPI anchors has long remained a mystery (Kinoshita and Fujita, 2016;Paulick and Bertozzi, 2008). Some GPI-anchored proteins are released from their anchor and detected in body fluids, implying involvement of one or more endogenous GPI-specific hydrolases. Recent studies have advanced the field by showing that cleavage and shedding of certain GPI-anchored proteins is mediated by a cellintrinsic transmembrane glycerophosphodiesterase, termed GDE2 (or GDPD5), thereby promoting neuronal differentiation through multiple signaling pathways (Matas-Rico et al., 2016;Matas-Rico et al., 2017;Park et al., 2013).
In this study, we focused on the shedding of GPI-anchored uPAR because of its regulatory role in multiple cellular and (patho)physiological activities, while soluble uPAR is considered a biomarker of various human pathologies. Here we report that GDE3 functions as a long-sought GPI-specific PLC that releases uPAR from its anchor. By contrast, its homologue GDE2 failed to release uPAR. As a consequence of GDE3 action, uPAR loses its vitronectin-dependent and matrix-degrading activities, when assayed in HEK293-uPAR and triple-negative breast cancer cells that express both uPAR and uPA. Importantly, loss of uPAR expression by GDE3 was found to be restricted to certain microdomains at the basolateral plasma membrane, where signal transduction is likely to take place. Thus, by acting as a GPI-specific PLC towards uPAR, GDE3 is a negative regulator of the uPAR signaling network ( Figure 7C) that includes uPAR's proteolytic and non-proteolytic activities. Consistent with this, GDE3 overexpression in uPA/uPAR-positive MDA-MB-231 breast cancer cells slowed tumor progression in a xenograft mouse model. Although statistically significant, the inhibitory effect of GDE3 overexpression on tumor growth was not dramatic, which should not come as a surprise since MDA-MBA-31 cells express the strongly oncogenic mutant K-RAS protein, which tends to override the regulation of numerous signaling pathways. Yet, this finding adds to the relevance of GPI- specific phospholipases in slowing tumor progression. Furthermore, high GDE3 expression was found to correlate with increased survival probability in triple-negative breast cancer patients. Interestingly, our previous work revealed a similar association between overexpression of GDE2 and positive clinical outcome in neuroblastoma patients, which appears attributable to GDE2-induced glypican shedding (Matas-Rico et al., 2016). The present patient survival analysis should be interpreted with caution, however, since involvement of uPAR release remains to be formally proven. Furthermore, we cannot rule out that GDE3 may cleave additional GPI-anchored substrates whose functional loss could contribute to positive clinical outcome.
The present results predict that, depending on its expression levels, GDE3 may downregulate normal uPAR-dependent remodeling processes. Indeed, upregulated GDE3 accelerates osteoblast differentiation (Corda et al., 2009;Yanaka et al., 2003) in a manner resembling the uPAR knockout phenotype (Furlan et al., 2007). Furthermore, a striking >200 fold upregulation of GDE3/GDPD2 is observed during blastocyst formation (Munch et al., 2016), implicating GDE3 in the invasion of preimplantation embryos, a process in which the uPA/uPAR signaling network has been implicated (Multhaupt et al., 1994;Pierleoni et al., 1998). Although correlative, these results support the view that GDE3 is upregulated to downregulate uPAR activity in vivo. The present findings also suggest that circulating full-length suPAR should be regarded as a marker of GDE3 activity, not necessarily reflecting uPAR expression levels.
It will now be important to determine how GDE3 expression and activity are regulated and, furthermore, to explore the substrate selectivity of the respective GDEs in further detail. Homology modeling revealed striking differences in electrostatic surface properties of GDE2 versus GDE3, suggesting that protein-protein interactions may determine substrate recognition by these GDE family members. Specific GPI-anchor modifications (Kinoshita and Fujita, 2016;Paulick and Bertozzi, 2008) could also determine the sensitivity of GPI-anchored proteins to GDE attack.
Finally, when regarded in a broader context, the present and previous findings (Matas-Rico et al., 2016;Matas-Rico et al., 2017;Park et al., 2013) support the view that vertebrate GDEs, notably GDE2 and GDE3, have evolved to modulate key signaling pathways and alter cell behavior through selective GPI-anchor cleavage.

Transfections and RNA i -mediated knockdown
Cells stably expressing uPAR-GFP or GDE3-mCherry were generated using retroviral transduction and subsequent selection with puromycin. Transient transfections were done using the calcium phosphate protocol or XtremeGene 9 agent (Roche). Stable uPAR knockdown in MDA-MB-231 cells was achieved using shRNAs in a lentiviral pLKO vector; five shRNAs from three RC human shRNA library were tested: TRCN0000052637, TRCN0000052636, TRCN0000052634, TRCN0000052633 and TRCN0000052635. The latter two were used for experiments; sequences: CCGGCCCATGAATCAA TGTC TGGTACTCGAGTACCAGACATTGATTCATGGGTTTTTG and CGGGCTTGAAGA TCAC-CAGCCTTACTCGAGTAAGGCTGGTGATCTTCAAGCTTTTTG, respectively. For virus production, HEK293T cells were transiently transfected using calcium phosphate, and virus particles were collected 48 hr thereafter. uPAR knockdown cells were selected in medium containing 2 mg/ml puromycin.

Liquid chromatography-mass spectrometry (LC-MS)
To determine the inositol phosphate content of cleaved uPAR, suPAR was immuno-precipitated from HEK293 cell conditioned medium using anti-GFP beads (ChromoTek). To remove inositol phosphate from suPAR, the beads were treated with 0.1M acetate buffer (pH 3.5) and subsequently with 0.5M NaNO2 or 0.5M NaCl (Control) for 3 hr as previously described (Mehlert and Ferguson, 2009). Inositol phosphate-containing samples were preprocessed by adding methanol to a final concentration of 70% and shaken at 1000 RPM at room temperature for 10 min. Following centrifugation (20,400 x g at 4˚C for 10 min), the supernatant was evaporated to dryness in a Speedvac at room temperature. The dried extracts were reconstituted in 50 mM ammonium acetate (pH 8.0), centrifuged (20,400 x g at 4˚C for 10 min) and transferred to autosampler vials. Liquid chromatography (LC) was performed using a Dionex Ultimate 3000 RSLCnano system (Thermo Fisher Scientific). A volume of 5 ml was injected on a Zorbax HILIC PLUS column (150 Â 0.5 mm, 3.0 mm particles) maintained at 30˚C. Elution was performed using a gradient: (0-5 min, 20% B; 5-45 min 20-100% B; Cell adhesion and spreading 48-wells plates were coated overnight at 4˚C with fibronectin (10 mg/ml) or vitronectin (5 mg/ml), or left untreated. Thereafter, plates were blocked for 2 hr at 37˚C using 0.5% BSA in PBS. Cells were washed and harvested in serum-free DMEM supplemented with 0.1% BSA. Equal numbers of cells were seeded and allowed to adhere for 1 hr. Non-adherent cells were washed away using PBS.
Attached cells were fixed with 4% paraformaldehyde (PFA) for 10 min, followed by washing and staining with Crystal violet (5 mg/ml in 2% ethanol) for 10 min. After extensive washing, cells were dried and lysed in 2% SDS for 30 min. Quantification was done by measuring absorbance at 570 nm using a plate reader. For cell-matrix contact area and lamellipodia measurements, coverslips were coated overnight with fibronectin or vitronectin and washed twice with PBS. Cells were trypsinized, washed and resuspended in DMEM and left to adhere and spread for 4 hr. After fixation (4% PFA) and F-actin staining with phalloidin, images were taken using confocal microscopy. Cell and lamellipodia area was quantified using an ImageJ macro.

Cell scattering and colony formation
Cell scattering was determined as described (LeBeau et al., 2013). In brief, single MDA-MB-231 cells were allowed to grow out as colonies, and the area covered by the scattered colonies (colony size) was measured at 6 days after plating. For measuring colony outgrowth, 500 cells were plated and colonies were counted after 14 days.

Matrix degradation
Coverslips were coated with OG-labelled gelatin (InVitrogen) supplemented with vitronectin (5 mg/ ml). About 100.000 cells per coverslip were seeded in DMEM with 10% FCS. After 20 hr, cells were washed, fixed with 4% PFM and stained with phalloidin-Alexa647 (InVitrogen). Gelatin degradation was determined from confocal images of >15 randomly automatically chosen fields of view per coverslip (testing at least two coverslips/condition on two separate days: total four coverslips per condition). The images were randomized and the area of degradation was normalized to the total area of cells or to the number of cells.

Western blotting
For Western blotting, cells were washed with cold PBS, lysed in RIPA buffer supplemented with protease inhibitors and spun down. Protein concentration was measured using a BCA protein assay kit (Pierce) and LDS sample buffer (NuPAGE, Invitrogen) was added to the lysate or directly to the medium. Equal amounts were loaded on SDS-PAGE pre-cast gradient gels (4-12% Nu-Page Bis-Tris, Invitrogen) followed by transfer to nitrocellulose membrane. Non-specific protein binding was blocked by 5% skimmed milk in TBST; primary antibodies were incubated overnight at 4˚C in TBST with 2% skimmed milk. Secondary antibodies conjugated to horseradish peroxidase (DAKO, Glostrup, Denmark) were incubated for 1 hr at room temperature; proteins were detected using ECL Western blot reagent.
Triton X-114 phase separation HEK293 cells transiently transfected with GDE2, GDE3 or empty plasmid, were plated on PEI-coated 6-well plates. After 24 hr complete medium was replaced with 1 ml serum free medium, 24 hr thereafter the conditioned medium was collected. 2% pre-condensed Triton X-114 was added to ice-cold conditioned medium and phases were separated as previously described (Doering et al., 2001). The top aqueous phase and the bottom detergent phase were analyzed by SDS-PAGE and Western blotting.

Microscopy
Cells cultured on 24 mm, #1,5 coverslips were washed and fixed with 4% PFA, permeabilized with 0.1% Triton X-100 and blocked with 2% BSA for 1 hr. Incubation with primary antibodies was done for 1 hr followed by incubation with Alexa-conjugated antibodies or Phalloidin for 45 min at room temperature. For confocal microscopy, cells were washed with PBS, mounted with Immnuno-MountTM (Thermo Scientific) and visualized on a LEICA TCS-SP5 confocal microscopy (63 x objective). Super-resolution imaging was done using a SR-GSD Leica microscope equipped with an oxygen scavenging system, as previously described (Matas-Rico et al., 2016). In short, 15000 frames were taken in TIRF mode, at 10 ms exposure time. After post image analysis, movies were analyzed and corrected using the ImageJ plugin Thunderstorm (http://imagej.nih.gov/ij/) followed by correction with an ImageJ macro using the plugin Image Stabilizer. For Total Internal Reflection (TIRF) microscopy, HEK293 cells stably expressing UPAR-GFP and transiently transfected with GDE3-mCherry were imaged using a Leica AM TIRF MC microscope with a HCX PL APO 63x, 1.47 NA oil immersion lens. Excitation was at 488 and 561 nm and detection of fluorescence emission was by a GR filter cube (Leica). Before each experiment, automatic laser alignment was carried out and TIRF penetration depth was set to 200 nm. Data were acquired at 500 ms frame rate. Basolateral uPAR patches were visualized using confocal microscopy and the area was quantified using an ImageJ macro that measured the uPAR patches based on fluorescence intensity.

Flow cytometry
HEK293 cells stably expressing uPAR-GFP were left untreated, treated with PI-PLC or transiently transfected with GDE3-mCherry. MDA-MB-231 cells were stably transfected with GDE3-mCherry. Cells were trypsinized, blocked in 2%BSA and stained with rabbit anti-GFP primary antibody or mouse anti-uPAR (13F6) antibody followed by AlexaFluor-647 coupled anti-rabbit secondary antibody. Cells were analyzed using a BD LSR Fortessa flow cytometer.

Mouse xenograft assay
Equal groups (n = 16) of eight-week-old female NMRI nude mice were injected subcutaneously into the mammary fat pads. Prior to injection, GDE3-expressing or control MDA-MB-231 cells (5 Â 105) were suspended in an equal volume of cold phosphate buffered saline and diluted 1:1 with Matrigel. The tumor growth was monitored 3 times a week for 9 weeks by caliper measurements. The tumor volumes were calculated with the formula: 0,5 x length x width:2 and statistical analysis of the tumor growth was done using a multiple t-test corrected for multiple comparison. A P value < 0.05 was considered statistically different. The mouse experiments were approved by the Animal Ethics Committee of the Netherlands Cancer Institute (protocol number: 30100 2015 407 appendix 1 WP 6061).

Statistical analysis
For all single comparisons, a two-tailed unpaired Student's t-test was used; for multiple comparisons, an ordinary ANOVA with Tukey's test was used. A P value < 0.05 was considered statistically significant. Error bars shown in the bar diagrams were calculated as the standard error of the mean (SEM); whiskers in the box plots depict 95% confidence intervals.
Yang J, Yan R, Roy A, Xu D, Poisson J, Zhang Y. 2015. The I-TASSER Suite: protein structure and function prediction.