Inhibition of HSP90 as a Strategy to Radiosensitize Glioblastoma: Targeting the DNA Damage Response and Beyond

Radiotherapy is an essential component of multi-modality treatment of glioblastoma (GBM). However, treatment failure and recurrence are frequent and give rise to the dismal prognosis of this aggressive type of primary brain tumor. A high level of inherent treatment resistance is considered to be the major underlying reason, stemming from constantly activated DNA damage response (DDR) mechanisms as a consequence of oncogene overexpression, persistent replicative stress, and other so far unknown reasons. The molecular chaperone heat shock protein 90 (HSP90) plays an important role in the establishment and maintenance of treatment resistance, since it crucially assists the folding and stabilization of various DDR regulators. Accordingly, inhibition of HSP90 represents a multi-target strategy to interfere with DDR function and to sensitize cancer cells to radiotherapy. Using NW457, a pochoxime-based HSP90 inhibitor with favorable brain pharmacokinetic profile, we show here that HSP90 inhibition at low concentrations with per se limited cytotoxicity leads to downregulation of various DNA damage response factors on the protein level, distinct transcriptomic alterations, impaired DNA damage repair, and reduced clonogenic survival in response to ionizing irradiation in glioblastoma cells in vitro. In vivo, HSP90 inhibition by NW457 improved the therapeutic outcome of fractionated CBCT-based irradiation in an orthotopic, syngeneic GBM mouse model, both in terms of tumor progression and survival. Nevertheless, in view of the promising in vitro results the in vivo efficacy was not as strong as expected, although apart from the radiosensitizing effects HSP90 inhibition also reduced irradiation-induced GBM cell migration and tumor invasiveness. Hence, our findings identify the combination of HSP90 inhibition and radiotherapy in principle as a promising strategy for GBM treatment whose performance needs to be further optimized by improved inhibitor substances, better formulations and/or administration routes, and fine-tuned treatment sequences.


INTRODUCTION
Glioblastoma (GBM) is the most aggressive type of primary brain tumor with a highly dismal prognosis and a 5-year overall survival of less than 5% (1). Standard treatment involves maximal safe resection-if possible-followed by radio(chemo)therapy, and maintenance chemotherapy according to the EORTC/NCIC protocol (2)(3)(4)(5). However, treatment failure and recurrence are frequent, and the major underlying reason appears to be the high level of inherent resistance against both chemo-and radiotherapy which represents a central hallmark of this cancer entity (6)(7)(8)(9). Moreover, recent data indicate that the degree of radioresistance further increases during therapy-particularly when radiotherapy is applied in classically fractionated regimens (10,11). Adaptive processes and an overt upregulation of the DNA damage response (DDR) have been reported to be crucial driving forces in this scenario (12)(13)(14)(15)(16). Since alternative fractionation regimens of radiotherapy have not shown relevant improvements (17- 19), and recurrence frequently occurs within the irradiated volume (20), the question arises if biological targeting of the DDR can contribute to break GBM radiation resistance. Intriguingly, the DDR relies on high molecular weight proteins and multi-protein complexes which essentially require folding assistance and stabilization by chaperones, such as heat shock protein 90 (HSP90) (21)(22)(23). Thus, HSP90 actively contributes to radioand chemoresistance of GBM and other cancer cells and represents an attractive target for biologically targeted radiosensitization, because HSP90 inhibition (HSP90i) -at least in principle-can affect multiple DDR pathways simultaneously (24)(25)(26). This was the focus of the present study for which we made use of the pochoxime-based, second generation HSP90 inhibitor NW457 with documented radiosensitizing potential in other cancer entities (27)(28)(29)(30)(31)(32). We observed that diverse DDR regulators are overexpressed in human GBM cells and that their protein levels decrease upon HSP90i at low nanomolar doses which per se exhibited only limited cytotoxicity. In HSP90i-treated GBM cells, DNA damage repair was clearly impaired translating into significantly reduced clonogenic survival upon irradiation in vitro. In vivo, HSP90i augmented the therapeutic efficacy of fractionated, conebeam CT (CBCT)-based irradiation in an orthotopic GBM mouse model, although less potently than expected. Interestingly, the invasive morphology of radiotherapytreated tumors was reverted by additional HSP90i, and in vitro migration analyses confirmed that HSP90i does reduce irradiationinduced GBM hypermigration.

Cell Lines and Reagents
The human GBM cell lines LN229 and T98G were obtained from ATCC (Manassas, VA, USA) and were cultured in Dulbecco's Modified Eagle medium (DMEM), supplemented with 10% heatinactivated fetal calf serum, 100 U/ml penicillin, and 0.1 mg/ml streptomycin (all from ThermoScientific, Schwerte, Germany) at 37°C and 7.5% CO 2 . The murine GBM cell line GL261 was obtained from the National Cancer Institute (NCI, Bethesda, MD, U.S.A.) and was cultured under same conditions. All cell lines were screened to be free from mycoplasma infection, and identity of human cell lines was confirmed by short tandem repeat (STR) typing (service provided by DSMZ, Braunschweig, Germany).
The HSP90 inhibitor NW457 (epi-pochoxime F) was previously described (27)(28)(29)(30)(31). For in vitro experiments, a 10 mM stock solution was prepared in DMSO (Sigma-Aldrich, Taufkirchen, Germany) and was further diluted to 100 µM with DMSO before final concentrations were adjusted in cell culture medium. Respective amounts of DMSO served as controls. For in vivo purposes, NW457 was dissolved at 100 mg/ml in DMSO and was further diluted in 0.9% NaCl (37°C), supplemented with 5% Tween-20 (all from Sigma-Aldrich). The vehicle formulation was used as control.

X-Ray Treatment In Vitro
Irradiation of cells was done with an RS225 X-ray tube (200 kV and 10 mA, Thoreaus filter, 1 Gy in 63 s, Xstrahl, Camberley, Great Britain) as described (30).

Quantitative Real-Time PCR (qRT-PCR)
Profiling of mRNA expression levels was performed by quantitative realtime RT-PCR as described (32). Briefly, total RNA was extracted from cells by NucleoSpin RNA II extraction kit (Macherey & Nagel, Dueren, Germany). 500 ng of isolated RNA were mixed with 5 µM random hexamers, 5 µM Oligo(dT) 18 , 500 µM dNTPs, 1 U/µl Ribolock RNase inhibitor, and 10 units/µl RevertAid transcriptase (all from ThermoScientific) and subjected to reverse transcription. Twenty nanograms of cDNA were employed for Realtime PCR runs with 300 nM primers in 1x Maxima SYBR Green qPCR Mastermix (ThermoScientific) on an LC480 qPCR platform (Roche Applied Science, Penzberg, Germany). Primer sequences are listed in Supplementary Table 1. Relative quantification was performed by the ddC T method. Results were normalized to a matrix of reference genes comprising 18S rRNA, d-Aminolaevulinate-synthase (ALAS), and b2-Microglobulin (B2M) and calibrated to the relative expression levels measured in primary human astrocytes (BioCat, Heidelberg, Germany). Three replicates were analyzed per cell line, and heatmaps were generated using the matrix visualization software Morpheus (https://software. broadinstitute.org/morpheus).

Clonogenic Survival Assay
Clonogenic survival was assessed by colony formation assays as described before (30,33). In brief, cells were detached by Trypsin/ EDTA (ThermoScientific), counted with a Neubauer counting chamber, and seeded as single cell suspensions at defined numbers anticipating 20-100 colonies per well depending on the different irradiation doses into 6-well plates. Adherence was allowed for 4 h. LN229 and T98G cells were treated with 10 nM NW457 or DMSO for 24 h, irradiated, and incubated in the presence of 10 nM NW457 for 13 d. GL261 cells were treated in a similar fashion, except that NW457 was removed by medium exchange after the 24 h of pre-incubation. Colonies were stained with methylene blue (dissolved at 0.3% in 80% ethanol, both from Merck, Darmstadt, Germany), and all colonies containing more than 50 cells were counted under a Stemi 305 stereomicroscope (Carl Zeiss, Oberkochen, Germany). The percentages of surviving cells were determined and calibrated to the corresponding plating efficiencies. Regression was performed according to the linearquadratic model.

Viability Assay
Viability was determined by Alamar Blue assays (BioRad, Puchheim, Germany) as described (30). Briefly, 5,000 cells were seeded into 96-well plates, adherence was allowed for 4 h, and cells were treated with NW457 at the indicated doses. Upon incubation for 24-96 h, medium was replaced by fresh medium supplemented with 1/10 volume of Alamar Blue reagent, and Resazurin conversion was allowed at 37°C and 7.5% CO 2 for 2-6 h. Resorufin fluorescence was measured on a Synergy Mx microplate reader platform (BioTek, Bad Friedrichshall, Germany), and results were calibrated to untreated controls.

Quantitative Fluorescence Microscopy
DNA damage repair was examined by immunofluorescence staining of phosphorylated histone variant H2AX (gH2AX) and p53-binding protein 1 (53BP1), followed by quantitative fluorescence microscopy as described (32). Cells were seeded into 24-well plates supplemented with coverslips, allowed to adhere overnight, and treated with 10 nM NW457 or DMSO for 24 h before being irradiated at 2 Gy. At the indicated time points, cells were fixed with 3.7% isotonic paraformaldehyde (Merck), containing 0.1% Triton X-100 (v/v Sigma-Aldrich) for 10 min before being permeabilized with 0.5% isotonic Triton X-100 for 5 min. Unspecific binding sites were blocked with 3% isotonic bovine serum albumin (w/v, Sigma-Aldrich) and 0.1% Triton X-100 at 4°C overnight. Cells were stained with monoclonal mouse anti-gH2AX (Merck Millipore) and polyclonal rabbit anti-53BP1 (Bio-Techne, Wiesbaden, Germany) antibodies diluted in 3% isotonic bovine serum albumin and 0.1% Triton X-100 for 2 h at room temperature. After extensive washing with PBS plus 0.1% Triton X-100, cells were stained with Alexa488-coupled goatanti-mouse IgG and Alexa568-coupled goat-anti-rabbit IgG (both from ThermoScientific) for 1 h. DNA was stained with 2 µg/ml Hoechst 33342 (Sigma Aldrich) for 10 min. Upon several washing steps with PBS plus 0.1% Triton X-100, coverslips were mounted with 4 µl mounting medium (Sigma-Aldrich) onto object slides. Microscopic analysis was performed with a Zeiss AxioObserver Z1 inverted microscope, equipped with an LCI Plan-Neofluar 63x/1.3 glycerol objective, an AxioCam MR Rev3 camera, and ZEN 2.3 software (all from Carl Zeiss). For image acquisition, 31 z-stacks with 250 nm interstack distance were captured, and deconvolution was performed with ZEN 2.3 software. For quantification of DNA damage repair, gH2AX/ 53BP1 double-positive foci were used. At least 20 nuclei of nondeformed morphology were selected for each condition, and foci were counted by hand. Results are depicted as individual data points with superimposed means and 95% confidence intervals.

Live-Cell Microscopy of Cell Death Morphology
For live-cell imaging, a Zeiss AxioObserver Z1 inverted microscope, equipped with an AxioCam MR Rev3 camera, an XL multi S1 incubation chamber, and a PS1 compact heating unit (both from Pecon, Erbach, Germany) was used. Briefly, cells were seeded into Ibidi µ-slides (Ibidi GmbH, Martinsried, Germany) and allowed to adhere for 4 h before treatment with 10 nM NW457 or DMSO for 24 h and irradiation at 4 Gy. Livecell imaging was initiated 1 h after irradiation and performed over 12 d. Images were captured in 12 min intervals, and movies were processed with Fiji software.

Wound Healing Assay
Migration of GBM cells was assessed by wound healing assays. Cells were seeded into Ibidi µ-slides supplemented with culture inserts (both from Ibidi) and allowed to adhere overnight. Cells were treated with 30 nM NW457 or DMSO for 24 h, irradiated at 3 Gy, and live-cell imaging was performed for 12 h. Images were captured in intervals of 3 min, and cell migration was analyzed using the manual tracking plugin tool (ImageJ) as previously described (31,34). Migration was quantified in form of colonized area (four regions of interest per condition in three independent experiments) and accumulated distance per cell over time (at least 25 randomly picked cells per condition).

Transcriptome Analysis Via RNA Sequencing
Transcriptome profiling was performed by 3'-RNA sequencing. Prior to sequencing, RNA integrity was assessed using the Agilent Bioanalyzer RNA 6000 Nano Kit (Agilent Technologies, Waldbronn, Germany) by calculating the percentage of fragments > 200 nucleotides (DV200). Sequencing libraries were prepared with 100 ng total RNA using the QuantSeq 3'-RNA-Seq Library Prep Kit FWD for Illumina (Lexogen GmbH, Vienna, Austria) according to the manufacturer's instructions for singleindexing and good RNA quality. For library amplification, PCR cycles were determined using the PCR Add-on Kit for Illumina (Lexogen), and the individual libraries were amplified with 17 PCR cycles. Quantity and quality of sequencing libraries were assessed using the Quanti-iT PicoGreen dsDNA Assay Kit (ThermoScientific) and the Bioanalyzer High Sensitivity DNA Analysis Kit (Agilent Technologies). Libraries were sequenced in 150 bp paired-end mode on a HiSeq4000 sequencer (Illumina, Berlin, Germany). The pool of individually barcoded libraries was distributed across the lanes of the same flow-cell aiming for approximately ten million paired-end reads per sample.
For sequence data processing, adapter sequences were removed using BBDUk (https://jgi.doe.gov/data-and-tools/ bbtools). Human fastq-files including forward-reads were subjected to alignment against the human genome reference genome (GRCh38) using STAR (36). Aligned reads were quantified via htseq-count employing appropriate transcriptome gtf-files (37). FastQC was utilized for analyzing quality of unaligned and aligned reads (https://www.bioinformatics. babraham.ac.uk/projects/fastqc/) followed by summarization via multiQC (https://multiqc.info). Genes with a raw read count (for the whole dataset) smaller than five times the total number of samples were excluded. Correlation heatmaps were employed to analyze data consistency and technical outlier detection, and shrinked (apeglm) log2 expression values were determined (38). Calculation of differentially expressed genes and geneset enrichment analyses (GSEAs) were performed on the basis of log2 expression values (39). Reactome functional interaction (FI) networks were constructed and analyzed in Cytoscape (40,41). iRegulon was employed to identify potential transcriptional regulators (42).

Orthotopic Mouse Glioblastoma Model and Contrast-Enhanced, Conebeam CT-Based, Fractionated Radiotherapy
All animal experiments were performed in accordance with the FELASA guidelines and upon ethical approval by the Regierung von Oberbayern. Female C57BL/6 mice were obtained from Charles River (Sulzfeld, Germany), and housed in groups of maximally four animals in individually ventilated cages (GM500, Tecniplast, Hohenpeißenberg, Germany) in a specified, pathogen-free animal facility with a 12 h day/night cycle. Standard rodent feed (from Ssniff, Soest, Germany) and water were provided ad libitum. Animals were inspected on a daily basis and sacrificed when reaching pre-defined health scores. Criteria for immediate sacrifice comprised the following: Strongly altered hygiene behavior, flattened breathing, body weight loss of ≥ 20%, ulcerating wounds, epileptic seizures or spasms, paralysis of extremities, bloody diarrhea, apathy, hunchbacked posture, self-mutilation, isolation from the group. In milder occurrence of these criteria, mice were sacrificed within 48 h. Intracranial implantation of GL261 cells was performed as described recently (43). Briefly, mice were medicated with 200 µg/g metamizol (WDT, Garbsen, Germany) and anesthetized by intraperitoneal injection of 100 µg/g ketamine and 10 µg/g xylazine (both from WDT). Mouse heads were mounted onto a stereotaxic frame (David Instruments, Tujanga, CA, USA), skulls were exposed by longitudinal skin incision, and a hole was drilled 1.5 mm laterally (right) and 1 mm anteriorly to the bregma using a pair of 23G and 21G microlances (BD Biosciences). Then, 90,000 GL261 cells (in 1 µl PBS) were slowly injected into the right striatum, using a stereotactically guided syringe (Hamilton, Bonaduz, Switzerland). Once the syringe was withdrawn, skin was closed with Ethibond Excel 5-0 suture material (Ethicon, Norderstedt, Germany), and mice were monitored until regaining consciousness. Starting at d7 after implantation, tumor growth was monitored by contrast-enhanced, conebeam computed tomography (CBCT) scans twice weekly using the small animal radiation research platform (SARRP, X-Strahl, Camberley, Great Britain) (44). For acquisition of CBCT scans, 360 projection images were captured (1°per image, x-ray tube settings: 60 kV, 0.8 mA, 1.0 mm aluminium filter). To enhance the contrast of soft tissue, 300 ml Imeron-300 (Bracco, Konstanz, Germany) were administered intravenously before CBCT scanning. Irradiation was performed at a weekdaily fractionation regimen of 2 Gy (2× 5× 2 Gy in total) with two contralateral beams (gantry positions − 90°and 90°) and 3× 9 mm 2 collimation (fixed nozzle, x-ray tube settings: 220 kV, 13 mA, 0.15mm copper filter) on d7-11 and d14-18. Isocenters were aligned to the centers of contrast enriching volumes, and treatment planning was executed with Muriplan software (X-Strahl). NW457 was administered intraperitoneally at 10 µg/g or 50 µg/g 24 h before irradiation. Tumor volumes were determined by Lx Wx H measurements of the 3 longest orthogonal axes as described (43), and 3D reconstructions were generated in 3D-Slicer (www.slicer.org/).

Statistical Analyses
Statistical analyses were performed using OriginPro 9.1 software (OriginLab Ltd., Northhampton, MA, USA). Results are shown as individual data points of all replicates, means ± s.d., or means ± 95% confidence intervals as indicated. For group comparisons, two-sided Student's t-tests or ANOVAs (one-way or two-way) were employed as indicated. Survival analyses were performed according to Kaplan-Meier with log-rank testing.

RESULTS
HSP90 Inhibition by NW457 Leads to Downregulation of DNA Damage Response Factors on the Protein Level, Impaired DNA Damage Repair, and Reduced Clonogenic Survival in Response to Ionizing Irradiation in Glioblastoma Cells A high degree of inherent radioresistance belongs to the signature hallmarks of GBM (6,8). On the molecular level, GBM radioresistance is considered to derive from constantly activated DNA damage response (DDR) mechanisms driven by the overexpression of oncogenes, persistent replicative stress, and other so far unknown reasons (45,46). Previous studies have shown that HSP90 plays an important role in DDR function via its crucial involvement in folding and stabilizing DDR proteins and/or multi-protein complexes (21,22). Accordingly, the present study was designed to examine whether HSP90 inhibition (HSP90i) can efficiently sensitize experimental model systems of GBM to ionizing irradiation in vitro and in vivo as a multi-target approach of pharmacological interference with the DDR (26,47,48). For our study, we made use of two human and one mouse GBM cell line with distinct alterations in the loci of TP53, MGMT, CDKN2A, PTEN, and IDH1/2 as described for primary GBM (Table 1) (51,52). Initial qRT-PCR profiling confirmed that the human GBM cell lines LN229 and T98G show a broad-range upregulation of diverse DDR regulators as compared to normal human astrocytes suggesting that DDR activity is indeed increased-irrespective of the O6methylguanine-DNA-methyltransferase (MGMT) status ( Figure  1A and Supplementary Figure 1A) (53). The highest levels of overexpression were detected for the replication and DDRassociated nucleases FEN1 and EXO1, members of the DNA double-strand break (DSB) detecting MRN complex (MRE11 and NBN), the DNA helicases BLM and PALB2, the singlestrand binding protein RPA1, and members of the XRCC family which are involved in non-homologous end joining (XRCC4 and XRCC6) and alternative non-homologous end joining (XRCC1).
Time course westernblot analyses of diverse DDR proteins revealed different clusters of responses in LN229 and T98G cells with several common motifs ( Figure 1B). Whereas the protein levels of few individual DDR regulators, such as NBN and RBBP8, increased upon HSP90i by NW457, clearly more candidates were downregulated, including a core cluster of CHK1, RAD51, and DNA2 with a particularly strong decrease in protein levels. Other downregulated DDR proteins comprised NHEJ1, KU80, XRCC3, ATR, CHK2, LIG4, and RPA1, and for some candidates mixed responses were observed. Our findings confirm and complement previous reports showing the critical dependence of certain DDR regulators on HSP90 chaperoning function (54)(55)(56). However, our data disclose also several DDR factors with so far unknown HSP90 dependence, including DNA2, an end resecting DNase with important functions in replication and DSB repair (57)(58)(59)(60), and NHEJ1, a scaffold protein that binds to and assists DNA ligase 4 (LIG4) in DSB repair (61). Overall, HSP90i affected proteins of various DDR pathwaysat least in the GBM cell lines used in our studythereby providing a strong rationale for combined modality approaches of HSP90i and radiotherapy. Importantly, downregulation of DDR proteins occurred already at very low concentrations of NW457 (10 nM) which per se exhibited only marginal cytotoxicity even during prolonged treatment-a characteristic which is of special interest for potential future clinical translation (Supplementary Figure 1B). These findings confirm the notion that DDR regulators, compared to other HSP90 client proteins, are particularly sensitive towards HSP90i (22).
In order to examine the consequences of HSP90i on DDR function, LN229 and T98G cells were pre-treated with NW457 for 24 h, irradiated at 2 Gy, and subjected to immunofluorescence staining for phosphorylated histone H2AX (gH2AX) and 53BP1. The kinetics of DNA damage foci formation and clearance was quantitatively analyzed (Figures 1C, D). HSP90i by NW457 resulted in significantly delayed clearance of gH2AX/53BP1 double-positive foci, indicating that DNA repair was obviously impaired. HSP90i also increased the overall numbers of foci as compared to the controls. This could either be due to false repair of irradiation-induced DNA damages, or irradiation-independent formation of damage sites. So far, our data show that HSP90i by low concentrations of NW457 leads to destabilization of DDR regulators in various pathways and reduced DNA damage repair capacity. In consequence, the question arises whether this also translates into reduced clonogenic survival of GBM cells upon irradiation. To this end, LN229 and T98G cells were pre-treated with NW457 for 24 h, irradiated at 0-8 Gy, and clonogenic survival was analyzed after 13 d incubation in the presence of NW457. As suggested by its marginal cytotoxicity (Supplementary Figure 1B), the effect of HSP90i monotherapy on clonogenic survival was modest in LN229 cells. In T98G cells it was clearly stronger, and for both cell lines this was statistically significant (Supplementary Figure 1C). Importantly, HSP90i by NW457 significantly reduced (C) Immunofluorescence microscopy of gH2AX and 53BP1 DNA damage repair foci in LN229 and T98G cells upon irradiation at 2 Gy ± HSP90i by NW457. Cells were treated with NW457 (10 nM) or DMSO for 24 h, irradiated, and fixed at the indicated times. Cells were stained for gH2AX, 53BP1, and DNA and subjected to deconvolution immunofluorescence microscopy. Scale bar depicts 10 µm. (D) Quantification of DNA damage repair kinetics from (C). gH2AX and 53BP1 double-positive foci in at least 20 randomly picked nuclei were counted by hand. Individual data points with superimposed means ± 95% confidence intervals are displayed, and overall curve comparison was performed by two-way ANOVA. (E) Clonogenic survival of LN229 and T98G cells upon irradiation at 0-8 Gy ± HSP90i by NW457. Cells were pre-treated with 10 nM NW457 or DMSO for 24 h followed by irradiation at the indicated doses, and colony formation was allowed for 13 d ± continuous NW457 treatment. Individual data points of 4 independent experiments are shown, linear-quadratic regression lines are superimposed, and overall curve comparison was performed by two-way ANOVA.
the clonogenic survival upon irradiation in both cell lines ( Figure 1E), confirming our hypothesis that multi-target interference with DDR function by HSP90i at per se non-toxic doses suffices to sensitize resistant GBM cells to irradiation. Similar findings were very recently reported for other cancer entities (62,63). Morphologically, the mode of cell death underlying reduced clonogenic survival upon HSP90i plus radiation was a highly disruptive, necrotic one which occurred after several rounds of aberrant mitosis and intermediate states of highly aneuploid cells with multiple and/or giant nuclei (Supplementary Movie File 1) (64).

HSP90 Inhibition by NW457 Stimulates a Compensatory Transcriptional Upregulation of Genes Involved in Protein Production Accompanied by Downregulation of Genes Engaged in Survival Signaling, Cell Stemness, and Integrin Signaling
Our results suggest-at first sight-that sensitization to radiotherapy upon HSP90i by NW457 derives from the downregulation of crucial DDR mediators on the protein level. Certainly, the immediate consequences of HSP90i affect the posttranslational level where abortive chaperoning leads to proteasomal degradation of HSP90 client proteins (65). Nevertheless, it is feasible to assume that this broad-range protein catabolytic remodeling can also stimulate complex responses on the transcriptome level which may contribute to radiosensitization as well. In order to address this question, we performed RNA sequencing analyses upon HSP90i by NW457 (10 nM) in LN229 and T98G cells. Differential gene expression analysis revealed an overt transcriptomic response in LN229 cells, and an attenuated but still detectable response in T98G cells ( Supplementary Figure 2A). The overlap in up-or downregulated genes was rather small (Supplementary Figure  2B), possibly pointing towards a regulatory involvement of p53 (T98G have dominant negative p53 M237I , LN229 carry p53 P98L with unclear functionality but intact DNA binding domain, see Table 1 and Supplementary Table 2). Construction of a functional interaction (FI) network of the upregulated intersect genes revealed a clear activation of the heat shock response, comprising many chaperones and members of the heat shock protein family (Supplementary Figure 2C). This FI network appeared to be predominantly controlled by heat shock factor 1 (HSF1) as suggested by iRegulon analysis (Supplementary Figure 2D). For the common downregulated genes shared by LN229 and T98G cells, a smaller FI network was constructed, basically comprising elements of DNA repair, mitosis regulation, NOTCH signaling, and protein folding (Supplementary Figure 2E). iRegulon analysis suggested a rather heterogenous pattern of transcriptional regulators, including EP300, JUND, BCL3, and others (Supplementary Figure 2F).  Figure 3A). The FI network of the compiled leading edge genes allowed the conclusion that this was in principle a compensatory response to HSP90i treatment, because interaction clusters representing basic functions of RNA polymerase II transcription, mRNA processing and splicing, RNA transport, translation initiation, protein folding, rRNA processing, and ribosome biogenesis were identified. Moreover, clusters involved in cell cycle regulation (G2/M transition, G1/S transition, and mitosis) and DNA repair were observed among the leading edge genes. Without prior GSEA, the FI network of all significantly upregulated genes showed additional interaction clusters of transcription factor activation (AP1, GR, and p53), focal adhesion and organization of the extracellular matrix (ECM), the heat shock response, and signaling by small GTPases (RAS, RAP1, and RHO) (Supplementary Table 2).
In terms of transcriptional downregulation upon HSP90i, no significantly enriched MSigDB hallmark genesets (FDR < 0.1) were observed in LN229 cells. However, on the level of significantly downregulated individual genes FI network construction revealed several interaction clusters whose decreased expression may contribute to radiosensitization by HPS90i. As such, clusters involved in survival signaling (EGFR and IGF1 signaling, PI3K/ AKT/mTOR signaling) and maintenance of cell stemness (NOTCH signaling) represent potential candidates, as well as clusters orchestrating integrin signaling, ECM receptor interaction, and ECM organization (Supplementary Table 3).

HSP90 Inhibition by NW457 Improves the Efficacy of Fractionated Radiotherapy in an Orthotopic, Syngeneic GBM Mouse Model
In the next step, we examined the performance of HSP90i plus radiotherapy in vivo. We made use of a syngeneic orthotopic mouse GBM transplantation model with i.p. injection of an in vivo NW457 formulation and contrast-enhanced, conebeam (CB)CTbased, fractionated radiotherapy (43) (Figure 2). Importantly, we first recapitulated the basic in vitro experiments with GL261 cells, the murine cell line that was used for transplantation. This cell line showed a particularly strong downregulation of DDR proteins upon HSP90i (Supplementary Figures 4A, B) and did not tolerate permanent NW457 incubation in colony formation assays so that NW457 needed to be removed after the 24 h preincubation time (Supplementary Figures 4C, D). Nevertheless, this treatment was already sufficient to facilitate radiosensitization of GL261 cells which have been described to exhibit particularly high levels of intrinsic treatment resistance (66).
Upon transplantation of GL261 cells into the right hemispheres of C57BL/6 mice, tumor progression was monitored by contrastenhanced CBCT scans over time (Figure 2A). Starting at d7 after implantation, mice were subjected to CBCT-guided, fractionated radiotherapy with 2× 5× 2 Gy using two contralateral beams at 3× 9 mm 2 collimation ( Figure 2B), and NW457 (or the vehicle control) was administered 24 h before each radiation treatment (10 or 50 µg/g i.p.). Tumor growth follow-up was accomplished by serial contrast-enhanced CBCT scans, and animals were sacrificed when reaching the pre-defined humane endpoints. Of note, d7 CBCT scans confirmed that tumor volumes were statistically not significantly different across different treatment groups at the start of therapy (Supplementary Figure 5A).
Overall, the treatment was tolerated well, and no significant differences in body weight in response to the treatment were observed (Supplementary Figure 5B). Tumors of vehicle-treated animals grew exponentially, and monotherapy with HSP90i delayed tumor growth only marginally ( Figure 2C). Fractionated radiotherapy exerted strong inhibitory effects on tumor growth, but responses were rather heterogeneous among the animals in this group. Additional HSP90i delayed tumor growth even further-yet compared to the effects observed in vitro, the in vivo performance was not as strong as expected. These findings basically mirrored the Kaplan-Meier survival analyses ( Figure 2D). HSP90i in mono-agent settings had only minor impact on animal survival, although this reached statistical significance compared to the vehicle controls at 50 µg/g NW457. Upon radiotherapy alone, animals revealed clearly prolonged survival, albeit again with heterogenous responses. Survival times were further increased by additional HSP90i, although to a rather limited, yet statistically significant extent in case of 50 µg/g NW457. Interestingly, two animals of the combined modality group (radiotherapy + 50 µ/g NW457) showed full tumor remission translating into persistent survival until the end of the experiment (Supplementary Figure 5C). In summary, these data indicate that HSP90i by NW457 can complement and improve the efficacy of fractionated radiotherapy in the used GBM in vivo model. Nevertheless, since the effects observed in vitro clearly outcompete the performance in vivo, further optimization of HSP90 inhibitor substances and/or treatment sequences would be needed.

HSP90i by NW457 Attenuates Irradiation-Induced Hypermigration and Invasiveness of GBM Cells In Vitro and In Vivo
Several previous studies have shown that non-lethal irradiation results in accelerated GBM cell migration-a phenomenon with implications for relapse and treatment failure (67,68). We therefore examined, whether HSP90i by low-dose NW457 treatment does interfere with GBM cell migration. LN229 cells were treated with NW457 for 24 h, irradiated at 3 Gy, and their migratory behavior was analyzed in wound healing setups by  Two contralateral beams with 3× 9 mm 2 collimation were used to administer 2× 5× 2 Gy. (C) Treatment schedule and tumor growth curves. Six days after orthotopic transplantation, mice were randomized into 6 groups (vehicle, 10 mg/g NW457, 50 mg/g NW457, 2× 5× 2 Gy + vehicle, 2× 5× 2 Gy + 10 mg/g NW457, 2× 5× 2 Gy + 50 mg/g NW457), and treatment was administered according to the indicated schedule. (D) Kaplan-Meier survival analysis of all treatment groups. Tumor-specific death was scored when mice showed pre-defined symptoms. p-values were obtained by log-rank test.
live-cell imaging ( Figure 3A and Supplementary Movies File 2). Migration was quantitated by the colonized area and the accumulated distance per cell over time as determined by tracking of at least 25 randomly picked cells per condition ( Figures 3B-D). With both approaches, the basal migratory activity of LN229 cells was observed to be significantly increased by radiation at 3 Gy, and this was almost completely reversed by pre-treatment with NW457. Thus, HSP90i, in addition to its radiosensitizing potential, does also efficiently counteract irradiation-induced hypermigration of GBM cells. In order to test whether this also holds true in vivo, we characterized the morphology of the tumors from our in vivo experiments. The aspect ratios of L axes (cranial-caudal, 90°to beam axes) and W axes (left-right, 0°to beam axes) were determined in contrast-enhanced CBCT scans of all mice at the day of sacrifice, and exemplary 3D reconstructions were generated ( Figures 4A-C).
Tumors of mice from the radiotherapy-only group showed significantly distorted aspect ratios and caudal-cranially stretched 3D reconstructions, implying tumor progress orthogonally to the irradiation field. This effect was fully reversed by additional NW457 treatment. Histologically, tumors from irradiated mice revealed clearly more invasive borders than tumors from vehicle or NW457-only-treated mice, and also more and larger areas of hemorrhage ( Figure  4D). Intriguingly, this morphotype was also fully reversed upon co-treatment with NW457, indicating that HSP90i does not only improve the therapeutic efficacy of radiotherapy but also counteracts GBM cell migration and tumor invasiveness in response to radiotherapy. These findings clearly strengthen the attractiveness of HSP90i as a partner for radiotherapy in combined modality settings.

DISCUSSION
Radiotherapy is a fundamental part of the standard of care for the treatment of glioblastoma (GBM) (69). However, treatment failure and (in-field) recurrence are frequent and form the basis for the dismal prognosis of this devastating disease (1). Significant advances in radiotherapy treatment and imageguidance technology as well as the addition of temozolomide (TMZ), a DNA alkylating chemotherapeutic drug, have led to modestly improved outcomes (2, 3), yet continued development remains urgently needed. GBM is characterized by a high level of inherent radioresistance which is considered to derive from overexpression of DNA damage response (DDR) genes and basally increased DDR activity (6,8,9,45,46). In this context, the molecular chaperone HSP90 is of particular interest and represents a promising target for radiosensitization approaches, since several key regulators of the DDR are known to crucially depend on HSP90 folding assistance (70)(71)(72)(73)(74)(75). However, in singleagent settings administration of HSP90 inhibitors was frequently associated with side effects of relevant severity, including gastrointestinal toxicity and hepatoxicity, because the concentrations needed to achieve anti-tumor effects-despite the relative selectivity for cancer versus normal cells-were rather high, and the employed substances exhibited suboptimal toxicity profiles and poor pharmacokinetic features (76). Intriguingly, quantitative mass spectrometric analyses revealed that pathways of the DDR are among the most sensitive ones in cancer cells that are perturbed by HSP90i already at very low inhibitor concentrations (22). Furthermore, a recent study showed that administration of very low, non-toxic doses of an HSP90 inhibitor of the third generation results in DDR protein disintegration in HNSCC and pancreatic cancer cells, while this was not observed in non-transformed, normal cells (62). This would allow targeting DNA damage repair mechanisms in cancer cells while not affecting the normal tissue and-in combination with radiotherapy-would imply a kind of biologically driven increase in radiation dose selectively at the tumor. Given that treatment-associated radionecrosis represents a major dose limiting factor in GBM radiotherapy, targeted radiosensitization of the tumor by HSP90i appears specifically attractive for this cancer entity, ideally in combination with modern high precision, image-guided radiotherapy (69).
In the present study, we report that treatment with very low concentrations of the pochoxime-derived HSP90 inhibitor NW457 which per se exhibit only limited cytotoxicity leads to DDR protein disintegration in GBM cells. We observed several key regulators of the DDR to be affected by HSP90i, previously published HSP90 client proteins as well as DDR regulators with so far unknown HSP90 dependence (32,(54)(55)(56)62). The cluster with the strongest decrease in protein levels upon HSP90i treatment in human and mouse GBM cell lines comprised CHK1, RAD51, DNA2, and NHEJ1. Accordingly, HSP90i represents a multi-target approach and affects various DDR pathways, including upstream checkpoint signaling (CHK1), double-strand break (DSB) repair by homologous recombination (RAD51), and non-homologous end-joining (NHEJ1), as well as crosslink repair and DNA replication (DNA2). This may explain why HSP90i is such a potent means of radiosensitization compared to mono-target approaches, for instance PARP inhibition (26,77). Nevertheless, due to the correlative nature of our observations we cannot exclude that other mechanisms, such as the HSP90itriggered proteasomal degradation of non-DDR proteins or the observed transcriptional downregulation of various genes, for instance genes involved in survival and/or integrin signaling, contribute to radiosensitization (78)(79)(80)(81). Functionally, degradation of DDR regulators upon HSP90i was accompanied by delayed DNA damage repair kinetics and significantly impaired clonogenic survival upon irradiation. In vivo, HSP90i by NW457 augmented the efficacy of fractionated radiotherapy in an orthotopic, syngeneic GBM mouse model, both in terms of tumor progression and survival. However, the observed effects were not as strong as expected from the convincing in vitro results. This may be due to limited GBM penetration by the inhibitor in vivo, although the family of pochoxime-derived HSP90 inhibitors has been shown to exhibit favorable brain pharmacokinetic profiles (82). Nevertheless, a very recent study with an orthotopic patient-derived GBM model reported similar therapeutic efficacy of a related pochoximederived HSP90 inhibitor in combination with whole brain irradiation (83). So, inhibitor substances with improved brain pharmacokinetic profiles, optimized formulations and/or administration routes, and/or fine-tuned treatment sequences may help to fully develop the synergistic potential of HSP90i and radiotherapy for the treatment of GBM.
In addition to its radiosensitizing effects, we observed that HSP90i by NW457 did reverse irradiation-induced GBM cell hypermigration in vitro and GBM invasiveness in vivo. This is of relevant interest, since GBM cells which survive radiotherapy and evade the target volume of radiotherapy may drive tumor relapse and dissemination. Our findings are in line with other reports showing that HSP90i efficiently decreases migration and invasion of human GBM cell lines (84)(85)(86). Although the detailed mechanisms of action remain elusive, we assume that the downregulation of migration regulating proteins is of importance in this scenario. On the protein level, mediators of protein (tyrosine) kinases have been reported to be particularly sensitive to HSP90i, and these are crucial regulators of migration-relevant signaling cascades (22). Additionally, our study shows that HSP90i stimulated the transcriptional downregulation of several interaction clusters involved in migratory processes, including integrin signaling, ECM receptor interaction, and signaling by small and large GTPases. HSP90i may thus offer a means to interfere with the highly infiltrative GBM phenotype which worsens with radiotherapy and represents another hallmark of this cancer entity contributing to its poor prognosis.
It should be noted, that HSP90i has also been shown to synergize with TMZ treatment in orthotopic models of GBM (87). This raises the question whether a triple combination of HSP90i and the current clinical standard of TMZ-based radiochemotherapy may improve the therapeutic outcome even further.
In conclusion, our study shows that HSP90i by low doses of NW457 potently interferes with the DDR in GBM cells leading to significant sensitization towards radiotherapy in vitro. The in vivo performance of this combined modality approach was less convincing than expected, although tumor growth was clearly delayed, survival was significantly prolonged, and radiationinduced invasive tumor morphology was reverted. Hence, our data reveal that the combination of HSP90i and radiotherapy is a promising strategy for GBM treatment whose performance needs to be further optimized by improved inhibitor substances, better formulations and/or administration routes, and fine-tuned treatment sequences.

DATA AVAILABILITY STATEMENT
The RNA sequencing datasets presented in this study can be found in online repositories. The names of the repository/ repositories and accession number(s) can be found here: NCBI GEO: GSE164717. All other datasets are available from the corresponding author upon reasonable request.

ETHICS STATEMENT
The animal study was reviewed and approved by the Regierung von Oberbayern.

ACKNOWLEDGMENTS
The murine glioblastoma cell line GL261 was obtained from the DCTP tumor repository (DTR, NCI, Frederick, MD, USA) in collaboration with Rainer Glass, Dept. of Neurosurgery, University Hospital, LMU Munich. The authors thank Manfred Felbermeier, Steffen Heuer, and Laura Holler for excellent technical assistance and animal husbandry.