Hypofractionated FLASH-RT as an Effective Treatment against Glioblastoma that Reduces Neurocognitive Side Effects in Mice

Radiotherapy is a cornerstone of cancer treatment used in over 50% of patients with cancer in high-income countries (1, 2). However, its efficacy remains suboptimal in many radiation-resistant tumors such as glioblastoma, for which standard treatment consists of surgical resection followed by radiotherapy and concomitant temozolomide administration. Classical therapeutic protocols induce debilitating neurocognitive complications in a vast majority of patients, including impairments in learning and memory, attention, executive function, and variety of mood disorders (3–7), without efficiently eradicating the tumors. Therefore, any approach that could enhance normal tissue tolerance would dramatically improve the benefits of radiotherapy, permitting increased doses to the tumor bed to achieve enhanced control (8–10). This fact has prompted efforts to develop truly innovative radiotherapy approaches able to eradicate tumors while sparing the normal brain from radiation-induced toxicities.

In this context, we have been the first to conceptualize and implement a novel modality of irradiation delivered at ultra-high dose rate (instantaneous dose rate > 10⁶ Gy/s), named FLASH radiotherapy (FLASH-RT) using a low-energy electron (LEE) prototype linear accelerator (LINAC) eRT6/Oriatron (11, 12). We have recently shown that classical pathogenic patterns observed in normal tissues exposed to radiation delivered at conventional dose rates were not induced by single fractions of FLASH-RT (11–14), collective observations that we have since coined as the “FLASH effect.” In the brain, long-term cognitive sparing was shown to be associated in part, by a lower production of reactive oxygen species (ROS), data obtained after the delivery of a single 10 Gy pulse over 1.8 μs (12, 14). While the capability of FLASH-RT versus conventional dose-rate radiotherapy (CONV-RT) to spare the normal brain from radiation-induced toxicities has been convincingly demonstrated, comprehensive studies exploring the efficacy of fractionated FLASH-RT on brain tumors were still lacking, but necessary to critically evaluate efficacy under a more clinically relevant scenario.

In this study, the LEE eRT6/Oriatron was set at its maximal electron current and the dose was delivered in a single—or maximum two pulses, which has proven to be optimal to achieve the FLASH effect (Fig. 1; ref. 15). These conditions were used to irradiate murine H454 glioblastoma tumors following orthotopic implantation of cells in the brain of Nude (NU_(Ico)-Foxn1^nu) mice. The choice of this orthotopic murine glioblastoma model, derived from genetically modified GFAP-HaRas^V12;GFAP-CRE;GFAP-LUC;Trp53^Flox/WT mice, was driven by its histopathologic resemblance to human glioblastoma, including a highly infiltrative phenotype, anaplasia, and polymorphism (16). Furthermore, hypofractionated treatments were designed for tolerance based on projected tumor growth and animals' response to anesthesia, such that reliable clinical outcomes could be reproducibly evaluated. In every circumstance (single dose or fractionated; whole brain or hemibrain), we found that FLASH-RT was equally efficient as CONV-RT in delaying tumor growth. Significantly, only animals that received FLASH-RT, either as 10 Gy single dose or hypofractionated regimens (2 × 7 Gy and 3 × 10 Gy), did not exhibit neurocognitive decline. These results show that neurocognitive sparing can be achieved with FLASH-fractionated regimens designed to approximate clinical treatment scenarios, without compromising tumor response. Moreover, these data again highlight a fundamental difference between the normal tissue and tumor response to FLASH-RT. While the “FLASH effect” remains to be elucidated at the mechanistic level, present data point to a unique opportunity to improve the radiotherapeutic management of brain cancer.

Animal experiments were approved by Swiss (VD2920 and 3241) Ethics Committee for Animal Experimentation and performed within institutional guidelines. Female Nude (NU_(Ico)-Foxn1^nu) mice (n = 8–14 per group) were purchased from Charles River Laboratories at the age of 8 weeks.

Irradiation was performed using a prototype 6 MeV electron beam LINAC of type Oriatron 6e (eRT6; PMB Alcen), available at Lausanne University Hospital (Lausanne, Switzerland) and described previously (17). The eRT6/Oriatron beam is horizontal and not equipped with three-dimensional imaging capabilities, but with portal films for positioning, checking, and dosimetry. Physical dosimetry has been extensively described and published to ensure reproducible and reliable biological studies (12, 17–20). This LINAC is able to produce a pulsed electron beam at a mean dose rate ranging from 0.1 Gy/s (i.e., comparable with conventional dose rates used in radiotherapy) up to 7.8 × 10⁶ Gy/s (at standard distance), corresponding to a dose, in each electron pulse, ranging from 0.01 up to 14 Gy. All FLASH irradiations were performed at an instantaneous dose rate above 1.8 × 10⁶ Gy/s (i.e., the intrapulse dose rate). The beam parameters used throughout this study are included in Table 1. The irradiation settings corresponding to the prescription dose for mouse irradiations were determined by surface dose measurements on a 30 × 30 cm²-solid water slab positioned behind a graphite applicator (13.0 × 13.0 × 2.5 cm³) with a 1.7-cm-diameter circular central aperture for whole brain irradiation (WBI) or semicircular aperture for hemibrain irradiation (HBI), as described previously (12, 20).

The H454 orthotopic murine glioblastoma model consists in injecting 500,000 H454 Luc+ murine glioblastoma cells (D. Hanahan, EPFL) in the right striatum of female Nude (NU_(Ico)-Foxn1^nu) mice with the coordinates: (AP: +1 mm; ML: +2 mm; DV: −3 mm). Cells were isolated from a GFAP-HaRas^V12;GFAP-CRE;GFAP-LUC;Trp53^Flox/WT genetically modified mouse model. Single dose or fractionated radiotherapy treatments (CONV or FLASH) started 3 days post-injection. Tumor development was assessed by bioluminescence (BLI) imaging the day of the first irradiation (normalized to 1) and weekly post-irradiation. Image acquisition was performed under isoflurane anesthesia using a Xenogen IVIS Lumina II (PerkinElmer, Inc.) and 10 minutes after an intraperitoneal injection of 15 mg/kg of luciferin and bioluminescence was quantified with Living Image Software (PerkinElmer, Inc.). The highly infiltrating features of H454 glioblastoma did not allow a postmortem correlation between semiquantitative BLI measurement and tumor volume. Tumor symptoms and survival were assessed.

The U87 orthotopic human glioblastoma model consists of injecting 50,000 U87 Luc+ human glioblastoma cells in the right striatum of Nude mice with the coordinates specified above. Tumor development was assessed by contrast-enhanced cone beam CT (CBCT) using iohexol contrast agent (200 μL of Accupaque 350 mg iodine per mL; GE Healthcare AG) right before imaging, providing for an accurate visualization of these bulkier tumors. Image acquisition was performed under isoflurane anesthesia using a small animal irradiator (X-rad 225Cx, Precision X-Ray) at 80 kV and 1.5 mA. Tumor volumes were measured from DICOM files with Osirix DICOM viewer (Pixmeo) and calculated with the formula of an ellipsoid volume: |V = {\frac{4}{3}} \times \pi \times L \times W \times H$|⁠.

All brain irradiations of tumor-bearing mice were performed under isoflurane anesthesia. For WBIs, the mouse was positioned directly behind the graphite applicator (in contact) with the head positioned in the 1.7-cm-diameter circular aperture in order to irradiate the whole encephalon region, while limiting the dose to the eyes, the mouth, and the rest of the body. For hemibrain irradiation, the head region was positioned in the 1.7-cm-diameter semicircular aperture to only expose the tumor-bearing brain hemisphere.

For WBIs, mice received 10 or 14 Gy single dose; daily fractionated doses of 4 × 3.5 Gy or 2 × 7 Gy; or 3 × 10 Gy spaced by 48 hours. For HBIs, a single dose of 25 Gy was delivered to the right hemisphere (see Table 1 for irradiation parameters). For all regimen, FLASH and CONV irradiation modalities were compared.

To determine the effects of the different regimens of conventional and FLASH dose-rate radiotherapy on cognitive function, mice were subjected to behavioral testing 1 month after the first radiotherapy treatment. Mice bearing H454 orthotopic glioblastoma tumors were administered the novel object recognition (NOR) task (14) to assess memory skills associated to the cortex function (21, 22). Discrimination index (DI) were calculated as |\left({{\frac{{{T_{novel}}}}{{{T_{total}}}}} - \ {\frac{{{T_{familiar}}}}{{{T_{total}}}}}}\right) \times 100$|⁠, where T_novel is the time spent exploring the novel object, T_familiar is the time spent exploring the familiar object, and T_total is the total time of exploration. Data analysis was conducted independently and blind and is presented as the average of all trials scored for each task. None of the treatment regimens resulted in observable skin toxicity or any other phenotypic differences that could have biased test scoring results.

Statistical analyses were carried out using GraphPad Prism (v8) software. P values were derived from the Mann–Whitney U test or log-rank test for survival studies. Results were expressed as mean values ± SD or mean values ± SEM, and all analyses considered a value of P ≤ 0.05 to be statistically significant.

As the number of groups investigating FLASH-RT steadily rise, the importance of carefully defining the critical beam parameters and other relevant conditions used to characterize the FLASH effect becomes increasingly important. In prior reports, these topics were discussed along with additional parameters such as instantaneous dose rate, duration of exposure and pulse repetition (among others), that now formally define the FLASH effect (15, 23). Here we have provided an updated version of the temporal dosimetry characteristics reported previously (15), that plot the duration of exposure versus the pulse dose rate from publications claiming to have produced—or not, the FLASH effect (Fig. 1). Importantly, we have collated and analyzed these data in detail to derive the optimized irradiation schedules selected for this study, and to help advance the field of FLASH-RT, we have provided a full characterization of all relevant FLASH beam parameters (Table 1). Dose-rate de-escalation studies using electrons have been conducted in the past (24) and more recently (12) and have identified a threshold of time of exposure and intrapulse dose rates that do not elicit the FLASH effect (colored cross Fig. 1).

Given the importance of carefully defining our beam parameters, the main focus of this study was not only to substantiate prior findings of normal tissue sparing in the FLASH irradiated brain but, to conclusively demonstrate that this new innovative irradiation modality was equally capable at controlling tumor growth compared with standard dose rate modalities. Second, and as alluded to above, we selected hypofractionated irradiation protocols to substantiate that both single and multifraction schedules were able to maintain the efficacy of tumor treatments while minimizing normal tissue toxicities.

To accomplish these objectives, we first used glioblastoma-bearing mice to investigate the antitumor efficacy of single doses of irradiation delivered with FLASH or CONV-RT. Nude mice implanted with H454 murine glioblastoma cells were given WBI using a single dose of 10 Gy, similarly to previously published normal tissue studies (12, 14), or 14 Gy with FLASH or CONV-RT, and tumor growth was quantified between nonirradiated controls and the irradiated groups as measured by bioluminescence over 4 weeks postirradiation. With 10 Gy, and for all time points, both conventional dose-rate irradiation (mean change in relative bioluminescence of 59.7 vs. 2436.0, P < 0.01 at 3-week time point) and FLASH-RT (mean fold change of 130.7 vs. 2436.0, P < 0.0001 at 3-week time point) showed a lower tumor burden compared with controls, whereas no difference in antitumor effects was observed between FLASH and conventional dose-rate irradiations (P = 0.90 at 3-week time point; Fig. 2A). Increasing the dose to 14 Gy single-dose WBI enhanced tumor growth delay with the two radiotherapy modalities (mean fold change of 12.7 for 14 Gy CONV and 9.4 for 14 Gy FLASH vs. 1625 controls; P < 0.001 at 3-week time point), but without reaching tumor control (Fig. 2B). Furthermore, CONV-RT and FLASH-RT resulted in comparable survival rates (Fig. 2D and E), but limited efficacy.

Simultaneously, neurocognitive consequences of irradiation were investigated using the same animals with a NOR task. To avoid the potentially confounding impact of tumor progression on neurologic function, assessment of cortex-dependent recognition memory by NOR was evaluated at 4 weeks posttumor initiation, when cognition was not affected by the tumor per se. Nontreated animals show no drop in DI (39.3 ± 6.0, mean DI ± SEM, n = 14) depicting an absence of tumor-induced cognitive impairment (Fig. 2G and H). However, a drastic and significant drop in the DI was observed for the conventional dose-rate 10 Gy irradiated group compared with controls (18.0 ± 5.1 vs. 39.3 ± 6.0; P = 0.021). Remarkably, tumor-bearing animals subjected to FLASH-RT exhibited no such decrements and were statistically similar to controls (39.9 ± 3.6 vs. 39.3 ± 6.0; P = 0.86;Fig. 2G). Nevertheless after 14 Gy WBI, both CONV and FLASH irradiated groups showed a significant decrease in the DI (17.2 ± 5.3 vs. 50.4 ± 7.1 and 20.4 ± 8.3 vs. 50.4 ± 7.1; P < 0.01 and P = 0.04, respectively; Fig. 2H). While the underlying causes responsible for such a sharp dose-response curve require further investigation, our recent results in tumor-free animals indicates that the vascular (25) and glial (26) pathologic responses to irradiation are probably right shifted (i.e., toward higher doses) at ultra-high dose rates.

Altogether, these data show that a single dose of 10 Gy FLASH-RT is as efficient as CONV-RT to delay the growth of orthotopic glioblastoma but preserves mice cognitive skills. Nevertheless, dose escalation to 14 Gy FLASH, even if more efficacious against tumor, cannot be delivered without impairing the mice cognitive function and is not sufficient to reach tumor control. Therefore, we investigated the effect of 14 Gy delivered in 2 fractions of 7 Gy, 24 hours apart (Fig. 2C). Whereas 2 × 7 Gy antitumor efficacy was similar to a single dose of 14 Gy, (P = 0.64 for CONV and P = 0.22 for FLASH), neurocognitive testing showed a significant decrease in DI for animals treated with 2 × 7 Gy CONV as compared with the controls (11.7 ± 7.1 vs. 33.5 ± 4.3; P = 0.01), whereas FLASH irradiated animals depicted excellent performance 4 weeks posttreatment (49.4 ± 8.5 vs. 11.7; P < 0.01; Fig. 2I). No significant increase in survival was induced by this treatment regimen (Fig. 3F).

Next, to get closer to a dose per fraction used in clinics, we studied the effect of a 4 × 3.5 Gy daily fractionated regimen on H454 glioma-bearing animals (Fig. 3A, C and E). For all time points, both conventional dose-rate irradiation and FLASH-RT showed a similar but limited growth tumor delay (mean change of 95.1 and 89.8; P < 0.001 vs. control at 3-week time point), without any improvement of overall and survival (Fig. 3A and C). No cognitive impairment was observed with this protocol (Fig. 3E). Animals irradiated with CONV-RT or FLASH-RT showed equivalent performance as the nonirradiated animals on this task (22.1 ± 4.5 and 19.3 ± 8.4 vs. 33.5 ± 4.3; P = 0.193 and P = 0.12), showing no radiation-induced cognitive alteration. Thus, at low dose per fraction, there is no particular benefit of FLASH-RT over conventional dose rate, despite the fact that that the antitumor effect of FLASH-RT is maintained.

As none of the previous treatment regimens reached complete tumor control, nor improvement of overall survival, we investigated the effect of hypofractionated regimen of FLASH-RT on both H454 gliomas and normal tissue. Animals were exposed to 3 × 10 Gy spaced by 48 hours fractionated CONV-RT or FLASH-RT, corresponding to a biologically effective dose (BED) of 60 Gy (using an α/β of 10 for the tumor), a value close to the BED used in clinical care (BED of 72 Gy at 30 × 2 Gy; Fig. 3B, D and F). At 4-weeks after treatment, animals show significant tumor growth delay, demonstrating the efficacy of this regimen (mean change of 2.2 for CONV-RT and 2.1 for FLASH-RT vs. 1946.2 for controls, P < 0.0001). Once again, no difference in efficacy was observed between CONV and FLASH-RT (P = 0.965) using the murine glioblastoma cell line H454 (Fig. 3B) and confirmed using the human glioblastoma cell line U87 followed-up by CBCT imaging (Supplementary Fig. S1). Moreover, using H454, a significant increase in overall and median survival was observed after both treatments, with animals living up to 90 days post-treatment (Fig. 3D) while no relapse was observed in FLASH treated U87 animals over a period of 7 weeks (Supplementary Fig. S1A). Animals treated with 3 × 10 Gy FLASH-RT showed no cognition deficits compared with the control animals (46.9 ± 4.9 vs. 47.2 ± 5.4; P > 0.999), whereas CONV-RT induced a significant drop in the recognition memory skills (26.6 ± 5.3 vs. 47.2 ± 5.4; P = 0.02; Fig. 3F). These results show the possibility to increase the dose per fraction delivered with FLASH-RT to further increase the tumor growth delay without compromising the neurocognition of irradiated animals, whereas CONV-RT drastically affects neurocognitive functions.

Finally, to mimic a conformal treatment, reach a complete tumor control and increase the overall survival of treated animals, we increased the dose delivered to the tumor site by irradiating only the tumor-bearing hemisphere, both improving the tumor targeting and normal tissue sparing. With this configuration, animals were exposed to 25 Gy HBI delivered with CONV or FLASH-RT (Fig. 4). Both CONV-RT and FLASH-RT showed a similar and significant improvement in tumor control compared with the nonirradiated animals, with a steady tumor burden up to 7 weeks postirradiation (2.3 and 2.7 mean change, respectively, vs. 1293 for controls, P < 0.0001; Fig. 4A). This improvement in tumor growth delay was coherent with an increase in overall and median survival in both treated groups, with animals free of tumors living up to 220 days post-irradiation (Fig. 4B). Nevertheless, with this irradiation configuration, both CONV or FLASH-RT groups showed comparable neurocognitive functions when compared with the nonirradiated animals (24.0 ± 8.9 and 21.1 ± 11.3, respectively, vs. 23.5 ± 11.6; P > 0.70), with large intragroup variations (Fig. 4C).

The present results show in a single-murine glioblastoma model that FLASH-RT delivered with LEE with an instantaneous dose rate above 1.8 × 10⁶ Gy/s is able to spare the normal brain from radiation-induced toxicities without compromising tumor cure. Despite technical limitations, including limited study time points due to tumor aggressiveness and immunocompromised mouse models, data also establish an initial framework for future clinical applications and highlights the fact that hypofractionated protocols will likely provide the best option for maximizing the FLASH effect. These results also provide some baseline physics parameters for the further exploration of normal tissue and tumor responses to FLASH irradiation. Moving forward, it will be critical to carefully validate the conditions required to observe the “FLASH effect” especially given that many research groups have entered the FLASH field, using various types of beams including protons, photons, and very high-energy electrons.

The definition and characterization of the optimal dose rate(s) able to produce the FLASH effect is an active topic of research in our group. While we first quoted the mean dose rate and reported thresholds for neurocognitive sparing above 100 Gy/s (12), we have since realized that this was an oversimplification. It should also be emphasized that in many of our past studies a single pulse was used, therefore, the mean dose rate was equal to the instantaneous dose rate. To help resolve any unnecessary confusion concerning FLASH-RT, we have recently documented the parametric characterization of the FLASH effect (Fig. 1; Table 1). In addition to the mean dose rate, additional parameters important for the FLASH effect include instantaneous dose rate, total duration of exposure, repetition rate (frequency), pulse dose, number of pulses, pulse width, total dose, and exposed volume. This comprehensive definition of FLASH-RT is required for scientific rigor, accuracy and reproducibility of not only current but, forthcoming datasets. These parameters are highly important to more completely understand the physical, physicochemical and biological processes involved in the FLASH effect. Such a formal characterization should also be viewed as flexible rather than fixed, as new findings and discoveries will undoubtedly evolve and reshape our current views of FLASH-RT. Nonetheless, based upon the impressive functional outcomes reported here and previously (11–14, 27–29), clinical applications should be rapidly and pragmatically implemented (30). In this light, it remains important to point out that standard radiotherapy approaches currently use in clinics (intensity-modulated radiotherapy, stereotactic ablative radiotherapy, proton therapy, etc.) remain incompletely understood, pointing to the need to move more potentially efficacious treatments forward despite a thorough understanding of mechanism of action.

For instance, in the context of adult and especially pediatric patients with brain tumor, novel strategies are desperately needed for improving the therapeutic index of radiotherapy where progressive and debilitating normal tissue toxicities can severely compromise quality of life (3, 4). Given this backdrop, brain tumors were logical targets for the assessment of fractionated FLASH-RT efficacy since previous demonstrations of the FLASH effect in the brain were primarily done with single large doses and volumes (cm³). An orthotopic murine glioblastoma model was selected for its clinical relevance and feasibility to perform reliable and quantitative comparisons between the efficacy of FLASH-RT and CONV-RT. The best treatment regimen was subsequently implemented using a human orthotopic glioblastoma model and data confirmed that our findings could be extended to other tumors including humans. In all tested regimens, the antitumor efficacies of CONV-RT and FLASH-RT were equivalent and increased with the BEDs (Fig. 5A and B). In addition, HBI results showed the benefits of a pseudoconformal approach combined with FLASH-RT, where higher FLASH doses could be used on smaller irradiated volumes to achieve tumor control. These data support the future implementation of state-of-the-art imaging to further optimize the advent of bona-fide conformal FLASH-RT. At the normal tissue level, only FLASH irradiated animals showed a preservation of cognition after hypofractionated and single-dose radiation exposures. Importantly, although variation in neurocognitive capability was observed in nonirradiated animals, sparing was achieved in FLASH-treated tumor-bearing animals after various treatment regimens including 10 Gy, 2 × 7 Gy and 3 × 10 Gy with respective BED of 43.3, 46.7, and 130 Gy (using an α/β ratio of 3 for normal brain) pointing to the promise of achieving similar normal tissue sparing in the clinic (BED on the normal brain around 100 Gy) following various hypofractionated regimens. Interestingly, we showed that the use of FLASH-RT was particularly beneficial when the toxicity triggered by conventional dose rate irradiation was substantial, highlighting the possibility to deliver larger doses per fraction with FLASH-RT.

To our knowledge, these findings are the first demonstration in an orthotopic tumor-bearing mouse model of an intervention capable of improving short-term neurocognitive outcome without compromising the radiation response of the tumor. In the presence of an aggressively growing tumor, neurocognitive testing is typically confounded, which highlights the need to conduct testing at relatively early times following irradiation, as opposed to tumor-free animals that can be evaluated at later postirradiation intervals. Nonetheless, our success at controlling both tumor growth and adverse cognitive outcomes highlights the potential promise of safely increasing the total dose for tumor cure, through ultra-high dose rate FLASH-RT. Because normal tissue tolerances currently limit the dose delivered to the tumor bed, the capability of FLASH to avoid typical normal tissue complications provides an exciting clinical prospect that promises to stimulate further investigation in all disciplines of radiation oncology. Interestingly, the decreased production of ROS that we have previously described as one possible mechanism for normal brain protection (14) does not appear to impair tumor response, suggesting that additional tumor-specific mechanisms are involved.

The tumor cell line (H454) chosen in this study was isolated from the FVBN mouse strain that exhibits many behavioral abnormalities, rendering them nonsuitable for NOR testing. Therefore, we implanted the H454 cells in immunocompromised mice, which prohibits the investigation of immune cell contributions. Nevertheless, differences between tumor, normal cell metabolism and the microenvironment might provide some explanation for the FLASH effect. A recent report investigating FLASH and conventional dose rate irradiation has indicated the importance of oxygen tension in clonogenic survival assays (31), corroborating several past results (14, 32). Increased oxygen tension afforded by carbogen breathing was also found to negate the neurocognitive benefits of FLASH, again suggesting a role for oxygen and ROS in the FLASH effect (14). Moreover, tumor-related hypoxia in the context of treatment resistance remains to be investigated after CONV and FLASH-RT. Differences in redox metabolism between normal and tumor cells have also been proposed to account for some of the differential responses and have recently been proposed to model the FLASH effect (33). The capacities of normal cells to more efficiently regulate redox stress, including hydroperoxides and labile Fe, along with a decreased production of ROS could explain the differential effect observed between the tumor and normal tissue after FLASH-RT. While details remain to be validated by experimentation, multiple mechanisms are likely to operate simultaneously to elicit the FLASH effect and we among other labs are actively seeking to identify the best parameters to reproducibly and reliably optimize FLASH-RT.

In summary, while further in-depth experimentation is clearly needed to fully characterize the physical, physicochemical, and biological mechanisms of FLASH-RT, cautious implementation of this promising new cancer treatment seems increasingly plausible as the necessary technology becomes available.

P. Gonçalves Jorge reports grants from IRA during the conduct of the study. K. Petersson reports grants from FNS, FNS/ANR, ISREC Foundation, and MESR during the conduct of the study. M.-C. Vozenin reports grants from SNF, ISREC, and NIH during the conduct of the study. No disclosures were reported by the other authors.

P. Montay-Gruel: Conceptualization, data curation, formal analysis, investigation, writing-original draft, writing-review and editing. M.M. Acharya: Conceptualization, data curation, investigation, writing-original draft, writing-review and editing. P. Gonçalves Jorge: Data curation, investigation. B. Petit: Data curation, investigation. I.G. Petridis: Data curation, investigation. P. Fuchs: Data curation, investigation. R. Leavitt: Data curation, investigation. K. Petersson: Data curation, investigation. M. Gondré: Data curation, investigation. J. Ollivier: Data curation, investigation. R. Moeckli: Supervision. F. Bochud: Supervision. C. Bailat: Supervision. J. Bourhis: Supervision, funding acquisition. J.-F. Germond: Formal analysis, supervision, validation. C.L. Limoli: Conceptualization, formal analysis, supervision, funding acquisition, validation, writing-original draft, writing-review and editing. M.-C. Vozenin: Conceptualization, resources, formal analysis, supervision, funding acquisition, validation, writing-original draft, project administration, writing-review and editing.

The study was supported by a Synergia grant from the FNS CRS II5_186369 (to M.-C. Vozenin and F. Bochud), a grant from lead agency grant FNS/ANR CR32I3L_156924 (to M.-C. Vozenin and C. Bailat), ISREC Foundation thank to Biltema donation (to J. Bourhis and M.-C. Vozenin) and by NIH program project grant PO1CA244091 (to M.-C. Vozenin and C.L. Limoli), NINDS grant NS089575 (to C.L. Limoli), KL2 award KL2TR001416 (M.M. Acharya). P. Montay-Gruel was supported by Ecole Normale Supérieure de Cachan fellowship (MESR), FNS N°31003A_156892 and ISREC Foundation thank to Biltema donation, K. Petersson by FNS/ANR CR32I3L_156924 and ISREC Foundation thank to Biltema donation; P. Gonçalves Jorge by ISREC Foundation thank to Biltema donation; I.G. Petridis and R. Leavitt by Synergia grant from the FNS CRS II5_186369.

We would like to thank K. Shchors and D. Hanahan for the H454 orthotopic murine glioblastoma model, and the animal facilities of Epalinges for husbandry.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.