P-selectin Targeting Stimulates Caveolin-1-Mediated Endothelial Transcytosis in Medulloblastoma


 Medulloblastoma (MB) is the most common malignant pediatric brain tumor with ~30% mediated by Sonic hedgehog (SHH) signaling. While SHH effector Smoothened inhibition is promising, therapeutic index is reduced by high systemic doses required for sufficient intracranial drug concentrations. Treatment-related toxicities associated with radiation therapy result in lifelong morbidity, and SHH pathway inhibition in children notably results in permanent bone defects. Here, we found that nanoparticle-mediated P-selectin targeting on tumor vasculature induces caveolin-1-dependent transcellular passage across the blood-brain barrier. A vismodegib nanocarrier that targets P-selectin exhibited striking efficacy and markedly reduced both bone-related toxicities and drug exposure to healthy brain. These findings demonstrate a potent strategy and novel mechanism for targeted intracranial pharmacodelivery that overcomes the restrictive endothelial barrier to achieve enhanced tumor-selective penetration and has therapeutic implications for other diseases within the central nervous system.


Summary: Medulloblastoma (MB) is the most common malignant pediatric brain tumor with ~30% 24 mediated by Sonic hedgehog (SHH) signaling. While SHH effector Smoothened inhibition is promising, 25 therapeutic index is reduced by high systemic doses required for sufficient intracranial drug concentrations. 26
Treatment-related toxicities associated with radiation therapy result in lifelong morbidity, and SHH pathway 27 inhibition in children notably results in permanent bone defects. Here, we found that nanoparticle-mediated 28 P-selectin targeting on tumor vasculature induces caveolin-1-dependent transcellular passage across the 29 blood-brain barrier. A vismodegib nanocarrier that targets P-selectin exhibited striking efficacy and 30 markedly reduced both bone-related toxicities and drug exposure to healthy brain. These findings 31 demonstrate a potent strategy and novel mechanism for targeted intracranial pharmacodelivery that 32 overcomes the restrictive endothelial barrier to achieve enhanced tumor-selective penetration and has 33 therapeutic implications for other diseases within the central nervous system. 34 this is likely a result of the high doses required for therapeutic efficacy 7,8 . 48 49 Given the many limitations for the passage of small molecules across the BBB, nanoparticles have been explored as 50 a vehicle to improve delivery into brain tissues 9 . To date, much of this work has focused on strategies that enhance 51 passive mechanisms of transport for drug loaded nanoparticles across the BBB. For instance, in diseases that result 52 in a compromised BBB, such as glioblastoma, nanostructures have been observed to extravasate through the leaky 53 vasculature to accumulate at tumor sites 10 . Also, researchers have implemented analogous approaches to improve 54 drug delivery past an intact BBB, by developing strategies that first disrupt this barrier 11,12,13 . However, by permitting 55 unregulated passage across the BBB, such approaches not only abrogate the homeostatic functions of the BBB, but 56 potentially expose the brain to harmful toxins and pathogens. Alternative approaches for diseases such as SHH 57 subgroup medulloblastoma that retain BBB integrity have utilized non-targeting nano-carriers to extend systemic 58 circulation of small molecule drugs with only partial improvement of on-target toxicity profiles at high-doses 14 . 59 Importantly, recent work has suggested that passive entry of nanoparticles into solid tumors through gaps between 60 endothelial cells represents a minor mechanism of entry and that up to 97% of transport is through an active process 61 through endothelial cells 15 . However, the molecular mechanism of this active transcellular transport across 62 endothelial barriers has not yet been elucidated, and little is known of whether this transendothelial nanoparticle 63 transport occurs at the BBB. 64 mediated transcytosis. Using a genetic mouse model of SHH-MB with an intact BBB, we found that P-selectin 71 targeting results in active transport in tumor endothelium to enable delivery of fucoidan-based nanoparticles 72 selectively into the tumor microenvironment, which is enhanced by RT. Fucoidan nanoparticles encapsulating the 73 Smoothened inhibitor vismodegib (FiVis) exhibited significant effector inhibition at low drug doses, striking anti-74 tumor efficacy, and attenuated on-target bone-related toxicities. These findings demonstrate a novel approach for 75 improving the therapeutic index of vismodegib for SHH-MB and present a potent adjuvant strategy for delivering 76 drugs to treat brain tumors in combination with standard radiotherapy. Furthermore, we report an active mechanism 77 of transendothelial transport that can be exploited to improve drug delivery across activated brain endothelial cells 78 in conditions with an intact BBB. 79 80

Results 81
We first characterized the brain vasculature in a genetically engineered mouse (GEM) Ptf1a cre/+ ;Ptch1 fl/fl SHH-MB 82 model to investigate the integrity of the BBB. To assess the permeability of the BBB in mice at advanced stages of 83 SHH-MB, symptomatic mice (14 weeks or older) were injected intravenously with TMR-dextran. We observed 84 minimal extravasation of TMR-dextran into parenchymal brain tumor tissue (Extended Data Fig. 1). We conclude 85 that the BBB of these mice appears to remain intact well into advanced tumor stages, and thus parallels the physiology 86 of human patients with SHH-MB 6 . 87 88 We next examined P-selectin expression in the SHH-MB tumor microenvironment and the effects of low-dose X-89 ray irradiation (XRT) 18 . We found that P-selectin is expressed in SHH-MB tumor vasculature in the absence of 90 radiation (Fig. 1a) and that this expression could be further enhanced following a single 2 Gy dose of XRT (Fig. 1c). 91 Notably, the elevation of P-selectin expression following irradiation was confined to tumor regions and not apparent 92 in adjacent, normal brain tissue (Fig. 1c). In order to assess the potential to mitigate RT-related toxicity, we sought 93 to identify the minimal required dose of irradiation that could still achieve significant induction of endothelial P-94 selectin expression. We found that P-selectin expression could still be robustly induced following a single dose of 95 0.25 Gy XRT (Fig. 1b). P-selectin expression was observed to reach significantly elevated levels at approximately 6 96 hours after XRT, and these levels persisted for at least 24 hours afterwards (Fig. 1e, Extended Data Fig. 2). To 97 confirm the clinical relevance of endothelial P-selectin expression as a potential target molecule, we examined human 98 SHH-MB tumor tissue surgically resected from pediatric patients. IHC analysis similarly showed P-selectin 99 expression in tumor-adjacent vasculature (Fig. 1f). To target an SHH pathway modulator to P-selectin, we 100 synthesized nanoparticles encapsulating the Smoothened inhibitor vismodegib. We used a nanoprecipitation process, 101 incorporating the polysaccharide fucoidan, to target P-selectin, and a near infra-red dye (IRDye783), to facilitate 102 imaging 16 . The resulting fucoidan-vismodegib (FiVis) nanoparticles exhibited an average size of 80 nm +/-10 103 nm and were relatively homogeneous as determined by atomic force microscopy and dynamic light scattering (DLS) 104 ( Fig. 2 a, b). The FiVis nanoparticles retained this uniform size with no observable aggregation for up to 7 days when 105 stored at 4°C in deionized water (Extended Data Fig. 3a). FiVis nanoparticles were also tested for their stability in 106 serum and exhibited effective drug encapsulation for up to 12 hours as determined by quantification of free drug 107 released into the serum mixture, which would be indicative of nanoparticle disassembly (Extended Data Fig. 3b). We next sought to assess the issue of on-target vismodegib-related systemic and bone toxicity that has been observed 139 in preclinical studies in juvenile mice as well as in children with SHH-MB treated with therapeutic doses of 140 vismodegib 8,21,22 . In order to confer a survival benefit comparable to the FiVis nanoparticle treatment cohort, free 141 We next investigated nanoparticle localization to SHH-MB tumors as a function of P-selectin expression. In non-irradiated mice, treatment with FiVis nanoparticles resulted in a slight increase in targeting to the cerebellar tumor compared to adjacent normal forebrain tissue. However, in mice that were pre-treated with irradiation (1 Gy), we observed pronounced tumor accumulation of FiVis in contrast to non-tumor-bearing WT mice where it remained in an intravascular location (Fig. 2c). We also synthesized control dextran sulfate-based vismodegib nanoparticles (DexVis) that do not exhibit affinity to P-selectin (Extended Data Fig. 4) 19 . These nanoparticles, exhibiting similar size and drug encapsulation characteristics to FiVis, did not demonstrate a comparable tumor localization effect, even following irradiation (Fig. 2c). In contrast, pre-treatment with 0.25 Gy XRT resulted in upwards of a 3-fold increase in FiVis nanoparticle localization at tumor sites as compared to normal adjacent forebrain tissue (Fig. 2d). In addition, localization of FiVis nanoparticles to tumors was abrogated in P-selectin knockout SHH-MB mice (SELP null SHH-MB, Ptf1acre/+; Ptch1fl/fl; Selp-/-). Quantification of the vismodegib cargo by liquid chromatography-mass spectrometry (LC-MS) found similar accumulation within cerebellar tumor regions while relatively sparing normal adjacent forebrain regions (Fig. 2e).
We proceeded to test the functional response and efficacy of FiVis nanoparticle treatment in SHH-MB mice. We used qRT-PCR to measure the expression of Gli1, a downstream effector of SHH pathway activation. When combined with 1 Gy XRT, FiVis treatment resulted in Gli1 target inhibition in a dose-dependent manner (~80% inhibition with FiVis 10 mg/kg and ~90% inhibition with FiVis 20 mg/kg) (Fig. 3a). However, treatment with control DexVis nanoparticles in combination with XRT did not result in increased Gli1 inhibition compared to XRT alone ( Fig. 3b). Furthermore, the Gli1 inhibition conferred by a combination of XRT and FiVis nanoparticle treatment was sustained even with a very low 0.25 Gy XRT dose, which alone does not affect Gli1 levels (Fig. 3c). Of note, comparable levels of Gli1 inhibition with free drug in SHH-MB mouse models typically requires treatment with vismodegib at doses upwards of 50 mg/kg 20 . Using this Gli1 inhibition data to inform nanoparticle drug dosing in a survival study, we tested the efficacy of low dose XRT and FiVis for treating mice with advanced stage SHH-MB. Treatment of mice with 10 mg/kg FiVis significantly prolonged survival on its own, however when combined with 0.25 Gy XRT, FiVis treatment at 10 mg/kg further extended survival by more than two-fold (Fig. 3d).
Importantly, we did not see enhanced mouse survival following treatment with a combination of low dose 0.25 Gy XRT and free vismodegib at 10 mg/kg, suggesting that the efficacy observed with P-selectin targeted FiVis nanoparticles is likely not mediated by any potential BBB leakiness induced by this low level of ionizing radiation. vismodegib had to be given daily at a considerably elevated 50 mg/kg dose (Extended Data Fig. 9). However, short-142 term treatment of P10 mice with these high doses of vismodegib caused visible growth stunting (Fig. 4a). Notably, 143 these growth defects were abrogated in young mice treated with FiVis at 10 mg/kg, at which significant Gli1 inhibition 144 and survival benefit were observed in advanced stage SHH-MB mice. To assess general toxicity, mice were weighed 145 over the duration of treatment. We increased the free vismodegib dose to 100 mg/kg to better assess physiologic 146 changes at levels shown to promote significant intracranial drug levels in preclinical studies 20 . While mice treated 147 with 100 mg/kg free vismodegib exhibited growth restriction, the weights of FiVis treated mice paralleled those of 148 vehicle-treated control mice (Fig. 4b). Closer examination of bone tissue revealed significant abnormalities in femur 149 lengths (Fig. 4c) and trabecular bone number using micro-CT (Fig. 4d) in P10 mice treated with free vismodegib 150 doses effective in SHH-MB that were not observed in mice treated with efficacious FiVis dosing. 151 152 We next investigated the mechanism of material transport across the BBB. We observed from immunohistological 153 analysis of brains obtained from the above SHH-MB treatment studies that Cav1 expression was selectively 154 upregulated in endothelial cells within SHH-MB tumor regions and not within normal brain regions when SHH-MB 155 mice were treated with FiVis ( Fig. 5, a, b) but not in SHH-MB mice treated with low dose XRT alone (Fig. 5c). This 156 finding suggests that P-selectin engagement with fucoidan-based nanoparticles on endothelial cells might stimulate 157 a caveolae-mediated transendothelial transport mechanism. To further investigate transcytosis mechanisms, we 158 evaluated the uptake of FiVis nanoparticles in bEnd.3 murine brain endothelial cells. We found that 0.25 Gy XRT 159 significantly enhanced P-selectin expression on bEnd.3 cells in vitro similar to our in vivo SHH-MB GEM model 160 studies (Extended Data Fig. 6a). Notably, this XRT dose also resulted in enhanced uptake of FiVis nanoparticles into 161 bEnd.3 cells (Extended Data Fig. 6b). We next interrogated endocytosis pathways that may potentially mediate this 162 nanoparticle uptake. For pharmacological inhibition of either caveolin-dependent endocytosis or clathrin-dependent 163 endocytosis, we treated bEnd.3 cells with methyl-ß-cyclodextrin (CD) or chlorpromazine (CPZ), respectively 23 . 164 While treatment with CPZ did not affect nanoparticle uptake in bEnd.3 cells, groups treated with CD showed 165 significantly reduced FiVis uptake in a dose-dependent manner (Fig. 5d, Extended Data Fig. 6b). This result 166 suggested FiVis entry into murine brain endothelial cells was mediated by caveolin-dependent endocytosis. To 167 further investigate caveolin-mediated transcytosis, we assessed nanoparticle uptake using caveolin-1 knockout 168 (Cav1KO) bEnd.3 cells. We observed that FiVis uptake into Cav1KO cells was marginal, as compared to wildtype 169 (WT) cells (Fig. 5e). The WT bEnd.3 cells demonstrated elevated levels of nanoparticle uptake with increasing doses 170 of FiVis. We next sought to determine whether Cav1 also contributes to the transcytosis of FiVis nanoparticles across 171 brain endothelial cells. Cav1KO and WT bEnd.3 cells were cultured on porous transwell inserts to directly assess 172 transcellular transport. Confluent monolayers of the cells were evaluated for integrity by transendothelial electrical 173 resistance (TEER) and paracellular permeability (Extended Data Fig. 7c, d). FiVis nanoparticles introduced to the 174 upper chamber of the insert were quantified by fluorescence in the bottom chamber over the course of a 4-hour 175 incubation. The passage of FiVis across the cells and into the bottom chamber was significantly restricted in Cav1 176 6 passage across bEnd.3 cells into the bottom chamber and found to be equivalent in size to those used at the onset of 178 the assay, as measured by dynamic light scattering (Fig. 5g), consistent with a transcytosis mechanism of transport. 179

180
To investigate the role of Cav1 in material blood-brain barrier transport in vivo, we bred the Cav1 null allele onto 181 our SHH-MB GEM model to generate homozygous Cav1 null SHH-MB mice (Ptf1a cre/+ ; Ptch1 fl/fl ; Cav1 -/-). These 182 Cav1 null SHH-MB tumor mice showed similar latency and penetrance as Cav1 wild-type SHH-MB mice with 183 resultant tumors having similar histological features and levels of P-selectin expression (Extended Data Fig. 8 a, b). 184 We administered FiVis to these Cav1 KO SHH-MB mice following low dose XRT and observed that nanoparticles 185 localized to tumor vasculature presumably due to intact endothelial P-selectin targeting but did not appear to 186 extravasate into the brain tumor parenchyma (Fig. 5h). Furthermore, the SHH pathway target inhibition was 187 considerably diminished in Cav1 null SHH-MB mice, as evident by the abrogation of Gli1 target inhibition following 188 treatment with FiVis nanoparticles, as compared to that in Cav1 WT SHH-MB mice (Fig. 5i). These results support 189 a role for Cav1 in this active nanoparticle transport process across the BBB. A proposed transport model for fucoidan-190 based nanoparticle drug delivery across the blood-brain barrier is shown in Fig. 5j. 191 192 Discussion 193 The highly specialized endothelial cells that comprise the BBB act as a physiological barrier that regulates the entry 194 of molecules into the central nervous system. While this is favorable for maintaining homeostasis under normal 195 conditions, it is an impediment to effective drug delivery for the treatment of diseases that manifest in parenchymal 196 brain tissue. We report that one way to overcome this pervasive challenge is by stimulating the active transport 197 machinery of brain endothelial cells. Here we show that fucoidan-based nanoparticle targeting of P-selectin on tumor 198 vasculature facilitates caveolin-1 dependent transendothelial transport and enhances drug delivery across an intact 199 blood-brain barrier. Using a combination of nanoengineering and genetic approaches, we provide evidence for 200 fucoidan-based nanoparticle targeting to P-selectin expressing brain endothelial cells in vitro and to P-selectin 201 expressing SHH-MB endothelium in vivo. In addition, using both pharmacological and genetic approaches in vitro 202 and in vivo, we show that the cellular entry and transendothelial BBB transport of fucoidan nanoparticles to SHH-203 MB tumor tissue is caveolin-1 dependent and less likely through passive passage through interendothelial spaces as 204 suggested in recent studies 15 . This starkly contrasts previous efforts that have aimed to improve passive delivery 205 of drugs into brain tissue through disruption of tight junctions in the BBB that consequently might impact the 206 structural integrity of the neurovascular unit. 207 208 Our study presents novel mechanistic insights for nanoengineering a specific receptor-ligand interaction to facilitate 209 targeted and controlled delivery of nanoparticle-encapsulated therapeutic agents across the blood-brain barrier. 210 Previous work by others supports a role for P-selectin mediated engagement and caveolin-1 interaction. Notably, 211 activated endothelial cells have been shown to have significant colocalization of P-selectin and caveolin-1, a primary 212 constituent of caveolae 24 . Our data supports the notion that the tendency for P-selectin to partition into caveolae may 213 effectively prime P-selectin bound FiVis nanoparticles for uptake by caveolae-mediated endocytosis and subsequent 214 transcytosis across the BBB. Although beyond the scope of this current study, it is currently not known whether 215 engagement of fucoidan nanoparticles with P-selectin activates a signaling pathway to mediate caveolin-1 dependent 216 cellular entry of nanoparticles or if it is mediated through more direct mechanical forces. Along these lines, previous 217 work has shown that PSGL-1 binding to P-selectin on endothelial cells has shown involvement of intracellular PI3K 218 and Src kinases within leukocytes, however little is known about involved signaling pathways within endothelial 219 cells following P-selectin engagement 25 . Furthermore, recent work has also shown that CD44 and/or spectrin 220 cytoskeletal networks on endothelial cells can restrict selectin mobility to maintain apical density and clustering thus 221 allowing for leukocyte rolling through PSGL-1 24 . Interestingly, activated endothelial cells following tumor necrosis 222 factor-alpha stimulation also resulted in apical distribution and restricted mobility of caveolae with subsequent 223 colocalization of caveolae with P-selectin. Future studies will be important to distinguish whether fucoidan 224 nanoparticle engagement with P-selectin on activated endothelial cells results in involvement of either endogenous 225 signaling pathways or alterations in CD44 and/or spectrin cytoskeletal networks for subsequent cellular entry. 226

227
Although we demonstrate efficacy with fucoidan-encapsulated vismodegib as proof of principle for this P-selectin 228 nanotargeting strategy in a SHH-MB GEM model, there are several other potential clinical applications for this 229 approach. Many hydrophobic drugs including standard chemotherapeutics (such as vincristine, doxorubicin, 230 paclitaxel, etc.) and molecularly targeted small molecules (including trametinib, tazemetostat, etc.) have been 231 successfully incorporated into fucoidan-based nanoparticles 16,17 . Along these lines, we have recently reported a 232 computational algorithm (Q-SNAP) that predicts the feasibility of encapsulating drugs into fucoidan-based 233 nanoparticles for tailored applications 26 . Further clinical applicability of this P-selectin targeting approach likely 234 extends beyond medulloblastoma to other intracranial tumors including glioma 10,17 . Combination therapies to address 235 tumor heterogeneity and treatment resistance mechanisms have often resulted in additive dose-limiting toxicities. 236 The increased therapeutic indices observed with our P-selectin nanotargeting approach may allow tumor cell 237 autonomous and non-autonomous combination drug treatment strategies to facilitate tolerability in patients. In 238 addition, several central nervous system disorders including multiple sclerosis 27,28 , ischemic stroke 29 , and focal 239 epilepsy 30 have been shown to upregulate restricted endothelial P-selectin expression at sites of disease exacerbation 240 where leukocyte trafficking plays a role in disease pathogenesis. These settings may provide additional opportunities 241 for the targeted delivery of therapeutic agents specifically to sites of intracranial disease to enhance efficacy while 242 minimizing neurotoxicity and systemic toxicities. In conclusion, we anticipate that the continued investigation and 243 development of methods that harness and improve the active uptake of nanoparticles across the blood-brain barrier 244 and potentially other endothelial barriers will be instrumental for improving the efficacy of nanomedicines.   FiVis only, or (c) XRT only. (d) Uptake of FiVis nanoparticles into bEnd.3 cells following pre-treatment with 376 pharmacological inhibitors of endocytosis pathways as measured by flow cytometry. CPZ = Chlorpromazine 377 (inhibitor of clathrin-mediated endocytosis); MβCD = methyl-β-cyclodextrin (inhibitor of caveolae-dependent 378 endocytosis). Uptake was compared to a group of cells administered nanoparticles but no inhibitor. Data are means 379 ± SEM.***P < 0.001, ****P < 0.001 (one-way ANOVA); ns, not significant. (e) Uptake of FiVis nanoparticles at 380 indicated doses in Cav1 wildtype (WT) bEnd.3 cells compared to homozygous Cav1 knockout (KO) bEnd.3 cells.

Preparation of Vismodegib Nanoparticles 399
Fucoidan-encapsulated vismodegib nanoparticles (FiVis) were prepared by a nanoprecipitation method adopted from previous work by our group 16,17 . In a microcentrifuge tube, the aqueous phase was prepared by combining the following solutions: 400 μl of fucoidan (15 mg/mL), 50 μl of IR-783 (2 mg/mL), 50 μl of IR-820 (2 mg/mL), and 100 μl of 0.01mM sodium bicarbonate. While gently vortexing this mixture, 50 μl vismodegib (20mg/ml, dissolved in DMSO) were added dropwise. Upon complete addition of the organic phase, vortexing was ceased and the  The concentration of vismodegib in the nanoparticle suspension was quantified using HPLC. Prior to analysis, 411 nanoparticles were diluted 1:10 in DI water. An aliquot of this dilution was then mixed with acetonitrile at a ratio of 412 1:4 in order to extract vismodegib from the nanoparticles. Samples were then analyzed on an Agilent 1260 Infinity 413 II HPLC system with an InfinityLab Poroshell 120 EC-C18, 4.6 x 75 mm, 2.7 µm, analytical LC column. The mobile 414 phase was comprised of acetonitrile and/or DI water, each containing 0.1% TFA. Chromatographic separation was 415 achieved by gradient elution with acetonitrile (0% to 95%) at a flow rate of 1 ml/min. A single peak corresponding 416 to vismodegib was characteristically observed with absorbance at 260 nm and a retention time of 3.04 minutes. The 417 size and zeta potential of the nanoparticles were determined using dynamic and electrophoretic light scattering 418 measurements acquired with a Malvern Zetasizer Nano ZS. 419 420 Flow Cytometry 421 Murine brain endothelial (bEnd.3) cells were plated in a 12-well plate at a density of 150,000 cells/well in 1mL of media (DMEM, 10% FBS, 1% P/S). Once confluent, cells in treatment groups receiving ionizing radiation were exposed to 0.25 Gy XRT. After 1 hour, cells were collected, transferred to microcentrifuge tubes, and fixed on ice using 2% paraformaldehyde. containing both a membrane dye (CellMask Green, diluted 1:1000) and a nuclear dye (Hoescht 33342, diluted 447 1:10000). Following a 15-minute incubation at 37°C, cells were washed twice more with HBSS. Images were taken 448 using an Olympus IX51 fluorescence microscope equipped with XM10IR Olympus camera and an X-Cite Xenon 449 lamp. ImageJ software was used to process the data to create overlays of images taken from different channels. mouse was confirmed by PCR genotyping of tail biopsy using primers for Ptch1, Cre, Cav1, and Selp (see Table S3 458 for primer sequences). Both sexes were used for all studies. Animals were housed on a 12-hour light/dark cycle and 459 were given access to food and water ad libitum. The number of tumors and/or mice analyzed are provided in the 460 Results section and/or in the figure legends. 461 462

Assessment of Mouse SHH-MB Blood-Brain Barrier Integrity 463
A bolus of 100 μL of a 10 mg/mL 70kDa Dextran-TMR solution (Life Technologies) was injected via the tail vein 464 into tumor-bearing SHH-MB mice at advanced symptomatic stages, as previously described 6 . Brains were then 465 removed two hours later without perfusion, fixed overnight in 4% paraformaldehyde, embedded in OCT, and then 466 (5.0, 6.25, 7.50, 8.75 mM), or regular media. After 8 hours, cells were washed with PBS and treated with nanoparticles (1:100 dilution of FiVis in complete DMEM media). Cells were incubated with nanoparticles for 30 minutes at 37°C.
Afterwards, cells were washed twice with PBS and resuspended in freshly prepared FACS buffer containing propidium iodide as a viability stain. Data was collected on a BD LSR II flow cytometer, using the APC-Cy7 channel (excitation with 633 nm red laser, detection with 780/60 nm bandpass filter) to detect fluorescent signal from the IR-dyes within the nanoparticles. As previously, data was analyzed using FCS Express Software.
sections prepared at a thickness of 12μm. Immunofluorescent staining on tissue sections was performed using 467 antibodies against CD31 and P-selectin with appropriate secondary antibodies, counterstained with DAPI to visualize 468 nuclei, and then coverslipped using Fluoro-Gel mounting medium as described below. Detection of TMR-Dextran 469 in context of P-selectin and CD31 immunostaining was imaged using a fluorescent microscope (Zeiss Axioobserver) 470 and TIFF images post-processed using Adobe Photoshop CS6. Immunohistochemical staining of murine SHH MB tissues was performed at the Molecular Cytology Core Facility 483 of Memorial Sloan Kettering Cancer Center. Brain tissues were harvested from SHH-MB mice and fixed in 4% PFA 484 overnight. Fixed tissues were embedded in OCT and frozen sections were prepared at a thickness of 12μm. Heat 485 antigen retrieval (95 °C for 20 minutes) was performed with citric acid buffer (pH=6), and sections were blocked for 486 30 minutes with 10% normal rabbit serum in PBS. Sections were incubated with primary antibodies (CD31, P-487 selectin, Cav1) overnight at 4 °C and secondary antibodies for 1 hour at room temperature (see Table S1 for antibody 488 details). Slides were counterstained with Hoechst 33258 dye (Invitrogen) and coverslipped with Fluoro-Gel mounting 489 medium (Electron Microscopy Sciences). Immunohistochemical detection of human SHH-MB tissue was performed 490 at the Weill Cornell Medicine Center for Translational Pathology. De-identified human SHH-MB tissue was 491 molecularly characterized using genome-wide methylation classification approaches as previously described 492 described 35 . Tumor tissue was formalin fixed, paraffin embedded, and prepared as 5 μm tissue sections. 493 Immunophenotyping was performed on a Leica Bond III system using the modified protocol J. Sections were pre-494 treated using heat-mediated antigen retrieval with sodium citrate buffer (pH=6, epitope retrieval solution 1) for 30 495 minutes. Sections were then incubated with P-selectin antibody (Lifespan Biosciences; LS-B3656) for 60 minutes at 496 room temperature and detected using an alkaline phosphatase conjugated compact polymer system. Fast Red was 497 used as the chromagen. The section was then counterstained with hematoxylin and mounted with micromount. All 498 images were taken with a bright-field and fluorescence microscope (Zeiss Axio Observer) or digital Panoramic Slide 499 Scanner (3D Histech, Budapest Hungary). TIFF images (with no compression) were post-processed using Adobe 500 Photoshop CS6. 501 Immunoblotting and Quantification 502 SHH-MB tissue was dissected from tumor bearing mice following 2 Gy irradiation, homogenized through tissue 503 sonication in tissue lysis buffer (50 mM Tris-HCl, 120 mM NaCL, 5 mM EDTA, 0.5% NP-40, 100 mM NaF, 2 mM 504 Na3VO4, 10 mM Na4P2O7) supplemented with protease inhibitor (Sigma P8849). Protein concentration was 505 determined using the Pierce BCA protein assay kit (ThermoFisher), and proteins were separated by NuPAGE Novex 506 10% Bis-Tris precast gels (ThermoFisher) and then electrophoretically transferred to PVDF membranes. Membranes 507 were incubated with Odyssey Blocking Buffer (LI-COR Biosciences) and then incubated with primary antibodies 508 for P-selectin, p53, and GAPDH and the appropriate secondary antibodies and analyzed using near-infrared imaging 509 using the LI-COR Odyssey CLx Imaging System. The antibodies are described in Table S2. 510 511 qRT-PCR 512 Total mRNA was isolated from dissected mouse SHH-MB brain regions using TRI Reagent (Molecular Research 513 Center, Inc). Reverse transcription was performed with the iScript cDNA synthesis kit (Bio-Rad), and qpCR was 514 performed using SsoAdvanced Universal SYBR(R) Green Supermix (Bio-Rad) according to the manufacturer's 515 instructions. Fold-changes in expression were calculated using the ΔΔCT method. The Gapdh gene was used to 516 normalize the results. The primer sequences used are described in Table S3. 517 518 Nanoparticle Fluorescence Imaging and Quantification 519 Nanoparticle localization in brains of SHH-MB mice were analyzed ex vivo using a LI-COR Odyssey CLx Imaging 520 System. Brains were scanned at a depth of 1 mm from the dorsal surface at 42 m resolution in the 800 nm channel 521 to detect the fluorescence emission from the IRDye783 incorporated into the FiVis nanoparticles. Images were 522 quantified using Image Studio Software, Version 5.2.5 (LI-COR Biosciences). In brief, forebrain and cerebellar areas 523 for quantification were demarcated and the relative signal per demarcated area was defined as the sum of the pixel 524 density per demarcated area. The relative signal per demarcated area was normalized to pixel density signal from 525 forebrain regions of untreated mice. For consistency, total cerebellar regions were demarcated, which included SHH 526 MB tumor regions, and compared to demarcated forebrain regions. 527 528 Vismodegib Bone Toxicity Studies 529 Juvenile C57BL/6 wild-type mice (age P10) were administered either FiVis (10 mg/kg or 20 mg/kg) or free 530 vismodegib (100 mg/kg) twice daily for a total of 4 doses and compared to vehicle control mice at 6 weeks age. 531 Skeletal effects of each treatment were assessed by measurement of femur lengths using calipers. micro-CT ( CT) 532 analysis was conducted using a Scanco Medical CT 35 system at the Citigroup Biomedical Imaging Core as 533 previously described described 36 . Briefly, an isotropic voxel size of 7 m was used to image the distal femur. Scans 534 were conducted in 70% ethanol and used an X-ray potential of 55 kVp, an X-ray intensity of 0.145 mA, and an 535 integration time of 600 ms. CT analysis was performed by an investigator blinded to the treatment of the animals 536 under analysis. All endpoint CT analysis was carried out on 6-week-old mice. 537 538 Nanoparticle Transwell Assay 539 Mouse brain endothelial cells (bEnd.3) were seeded on the upper surface of the membrane in polyester transwell 540 inserts (0.4 μM pore size, 1 x 10 8 per cm 2 pore density, 8.4 mm diameter) at a density of 1 x 10 5 cells per well. Media 541 was changed every other day and cells were cultured for 5-7 days, until a confluent monolayer formed. Before 542 initiating the transport studies, transendothelial electrical resistance (TEER) across cell layers was measured until 543 they reached a TEER of 30 Ω x cm 2 . Once bEnd.3 cells monolayers reached this threshold TEER, endothelial 544 paracellular barrier function was evaluated by measuring the permeability of cells to 70-kDa tetramethylrhodamine-545 dextran (TMR-Dextran). The concentration of TMR-Dextran was determined by measuring fluorescence (excitation 546 at 555 nm and emission at 580 nm) using a TECAN plate reader. After confirming restriction to paracellular transport, 547 transport studies with FiVis nanoparticles were carried out by adding 200 µL of FiVis nanoparticles (20 µg/ml) to 548 the upper chamber of the insert. At various timepoints thereafter, the entire basal well volume was removed and 549 assayed for nanoparticle concentration by measurement of fluorescence (excitation at 790 nm and emission at 815 550 nm) and for nanoparticle size using dynamic light scattering. 551 552

Mass Spectrometry for Quantification of Vismodegib in Brain Tissue 553
Analysis of vismodegib concentrations in brain tissue were performed by Integrated Analytical Solutions (IAS). As 554 with efficacy studies, mice were irradiated with 0.25 Gy XRT. Two hours later, mice were injected intraperitoneally 555 with nanoparticles containing 10 mg/kg vismodegib. After 4 hours, mice were sacrificed, and brain tissue was 556 sectioned into two regions: forebrain (normal tissue) and cerebellum (tumor tissue). The tissue of each region was 557 weighed and snap-frozen prior to LC-MS analysis by IAS. A standard curve that was generated by adding known 558 amounts of vismodegib to homogenized brain tissue from non-treated mice. The concentrations of vismodegib in 559 forebrain and cerebellar tumor tissue samples were then calculated to determine mass of vismodegib (ng) per gram 560 of tissue. 561 562

Statistical Analyses 563
All data are shown as the mean ± s.d. or mean ± s.e.m., unless otherwise indicated. For comparison between two 564 groups, unpaired, two-tailed Student's t tests were used. Analysis of variance (ANOVA) followed by a post hoc test 565 for multiple comparisons (Dunnett's) was used for comparison of groups of 3 or more. For Kaplan-Meier survival 566 analysis, log-rank (Mantel-Cox) test was used. GraphPad Prism Version 9.1.0 software was used for statistical 567 analysis. P < 0.05 was considered statistically significant and additional indicators of statistical significance are 568 provided accordingly in the text or in figure legends. Statistical analysis was performed on measurements taken from 569 distinct samples. control cells (left) were cultured using standard media. Irradiated groups received 0.25 Gy XRT and were 624 cultured using standard media or media containing endocytosis inhibitors. CPZ = Chlorpromazine (inhibitor 625 of clathrin-mediated endocytosis); CD= methyl-β-cyclodextrin (inhibitor of caveolae dependent endocytosis).