The Irradiated Brain Microenvironment Supports Glioma Stemness and Survival via Astrocyte-Derived Transglutaminase 2

Glioblastoma (GBM) is the most aggressive and most common glioma, with less than a 10% 5-year survival rate (1). GBM is treated by surgery, radiation, and chemotherapy, yet nearly all tumors recur (2). Recurrent tumors generally form within or overlapping the original tumor volume (3) and within a gliotic region corresponding to the initial field receiving radiation, suggesting that recurring tumors form within the irradiated microenvironment. Recurred tumors have limited treatment options.

GBM recurrence is often attributed to radioresistant glioma cells, a phenotype that is related to stemness, in that tumor cells surviving radiotherapy display characteristics of stem cells, like enhanced self-renewal and increased drug-efflux capabilities, as compared with radiation-sensitive cells (4, 5). Understanding which factors promote treatment resistance of glioma cells is of intense interest. Microenvironmental cues such as hypoxia (6), extracellular matrix proteins (7, 8), or growth factors secreted by stromal cells (9, 10) may be sufficient to induce tumor cell therapeutic resistance. Microenvironmental regulation of stemness is further supported by the finding that stem-like tumor cells are enriched in specific niches, primarily the perivascular niche and hypoxic compartments (6, 11, 12). These niches contain a variety of stromal cell types including astrocytes, microglia, endothelial cells, and pericytes (13), which contribute to glioma aggressiveness (14, 15); however, little is known about how these cells respond to radiotherapy. We sought to determine whether irradiation of the brain tumor microenvironment affects the therapeutic response of glioma cells. We identified astrocyte-derived transglutaminase 2 (TGM2) as a potential radiation-induced modifier of the tumor microenvironment, which protected against radiation-induced cell death, and may serve as a potential therapeutic target.

Figure 2A–E: B6 (RRID: IMSR_TAC:b6) mice were irradiated 1 × 10 Gy. Figure 2F–L: Mice were irradiated with 0–10 Gy, injected 72 hours later with PIGPC intracranially, then euthanized upon glioma symptoms. Figures 3A–D, 4E–I, 5G and H, 6D–F: Male and female Nestin-tv-a mice (RRID: IMSR_NCIMR:01XH4) were injected with DF-1 cells (ATCC RRID:CVCL_0570) expressing RCAS-PDGFB and RCAS-shp53 (16, 17). Upon symptoms, mice were treated with 0–10 Gy radiation and euthanized 72 hours posttreatment. Figure 7A–C: Nestin-tv-a Ink4a/Arf^−/− mice were injected with DF-1 cells (ATCC) expressing RCAS-PDGFB and RCAS-shp53. Mice were euthanized upon glioma symptoms. Figure 7D–G: Nestin-tv-a mice were injected with DF-1 cells expressing RCAS-PDGFB and RCAS-shPTEN. Upon glioma symptoms, mice were euthanized for primary tumors or treated daily 5 × 2 Gy and euthanized upon symptom recurrence. Laboratory animal use was conducted in accordance with European Union directive on the subject of animal rights and approved by local ethical committees CEIyBA (CBA PA37_2013)/(IACUC 029–2013/001; CNIO), M-186/14 (Lund), and 5.8.18–16350/2017 (Uppsala). Animals were distributed randomly by weight and sex into treatment groups.

PIGPC were isolated as described previously (7). PIGPC and U251 (Sigma; RRID:CVCL_0021) cells were cultured in DMEM (Corning) supplemented with 10% FBS (Biological Industries) and 1% PenStrep (Corning). HBVP (ScienCell) were cultured in PM (ScienCell) on 2 μg/cm² poly-l-lysine mol wt 70,000 to 150,000 (Sigma) coated plastic and used below passage 4. HMC3 (ATCC, CRL-3304, RRID:CVCL_II76) were cultured in DMEM (Corning) supplemented with 10% FBS (Biological Industries) and 1% PenStrep (Corning). BEND3 (ATCC, CRL-2299, RRID:CVCL_0170) were cultured in Endothelial Cell Growth Medium MV2 (PromoCell) on 0.1% gelatin coated plastic. Primary human astrocytes (3H Biomedical) were cultured in Astrocyte Medium (3H Biomedical) and used below passage 15. U3082 glioma cells were obtained from HGCC (hgcc.se, RRID:CVCL_IR93) and were cultured in HGC medium on laminin-coated plates as described previously (18). Cells were acquired 2015 to 2020, tested quarterly for mycoplasma (Eurofins, MycoplasmaCheck), and not further authenticated.

Cryosections were fixed in 4% PFA, followed by permeabilization in 0.3% Triton-x-100 and blocking in 1% BSA in PBS or DACO blocking reagent. Cryosections of organotypic slice cultures were fixed in 4% PFA followed by permeabilization in 0.1% Triton-x-100 and 0.1% sodium citrate then overnight incubation in 1% BSA or DAKO Diluent. Antibody combinations can be found in Supplemental Information. Images were acquired using an Olympus BX63 microscope and DP80 camera and cellSens Dimension v 1.12 software (Olympus; RRID:SCR_014551). Quantification was performed using ImageJ (RRID:SCR_003070) or Cell Profiler. All images were blinded for analysis.

A total of 100,000 BEND3, HMC3, or PIGPC or 50,000 HBPV cells were pretreated with or without 10 Gy followed by 48 hours of coculture in DMEM with 100,000 PIGPC.

Irradiated pericytes, microglia, endothelial cells, and astrocytes were pretreated with 10 Gy in a CellRad x-ray irradiator (Faxitron).

SP was performed as described previously (19) with dual wavelength detection using 448/45 (Hoechst 33342-blue) and 613/18 (Hoechst 33342-red) filters.

Confluent astrocytes were irradiated with 0 or 10 Gy followed by 10 days of culture on 0.2% gelatin in astrocyte medium supplemented with 50 μg/mL L-ascorbic acid (Sigma). After 10 days, plates were decellularized in 0.4 mmol/L NH₄OH (Fluka), 0.5% Triton X-100 in PBS with 1 mmol/L CaCl₂ and 0.5 mmol/L MgCl₂ (PBS-MC) at 37°C, washed with PBS-MC, then treated with 10 μg/mL DNAse I (Roche) for 1 hour at 37°C and stored in PBS-MC at 4°C.

Matrix lysates were prepared by scraping into 175 μL 4X Laemmli buffer (Bio-Rad;161–0747) containing 150 mmol/L DTT and boiling for 10 minutes. Whole cell lysates were prepared directly in 1:2 4X Laemmli buffer:RIPA with 50 mmol/L DTT. Blotting details can be found in Supplementary Information.

Cell-derived matrix was solubilized for LC/MS-MS analysis according to Naba and colleagues (20). Detailed LC/MS-MS protocol can be found in Supplementary Information.

The Proteome Profiler phosphokinase array (RND, ARY003C) was used according to manufacturer's instructions on U3082 grown for 24 hours on plates coated with either 0.1 μg/cm² laminin or laminin plus 3.3 mU/cm² TGM2.

Human ELISA kits for lysates for TNFα (Sigma; RAB1089), IL6 (Sigma; RAB0308), and IL8 (Sigma; RAB0319) were used according to manufacturer's instructions on lysates from astrocytes derived from 4 individual donors 24 hours after irradiation with 10 Gy. TNFα values were normalized to total protein in lysates. IL6 and IL8 were used as markers of astrocyte activation.

U251 cells were plated at 200 cells/well in a 6-well dish on ADMc or ADMi in DMEM. After 24 hours, dishes were irradiated with 0 or 4 Gy. Cells were cultured 2 weeks or until visible colonies formed. Colonies were fixed in 4% PFA, stained with 0.01% Crystal Violet for 1 hour, photographed on a LAS-3000 imaging system, then counted using Image J followed by visual confirmation.

U3082 cells were cultured on ADMc or ADMi 5 days in HGC Media. For primary spheres, cells were dissociated with Accutase then replated in three wells each of 6-well dishes at 300 cells/mL in 2 mL HGC Media and cultured until visible spheres formed. Primary spheres were dissociated and replated as described above. Spheres were counted on a Zeiss AX10 inverted microscope.

U3082 cells were preconditioned on plates coated with either 0.1 μg/cm² laminin or laminin plus 3.3 mU/cm² TGM2 for 4 days followed by plating at a density of 300 cells/mL in 2 mL HGC media and culture until visible spheres formed. Primary spheres were dissociated then replated at 1,000 cells/mL in 300 μL in eight wells of a 48-well dish followed by serial 1:2 dilutions across eight columns in the presence or absence of 40 mU/mL TGM2 and the presence or absence of 10 nmol/L dasatinib until visible spheres formed. Any well containing a sphere was scored as positive. Limiting dilution assay was quantified using the analysis tool at http://bioinf.wehi.edu.au/software/elda/ (21).

Sequencing performed at the Center for Translational Genomics, Lund University and Clinical Genomics Lund, SciLifeLab, and National Genomics Infrastructure Stockholm Node, Science for Life Laboratory. Details in Supplementary Information.

The R2: Genomics Analysis and Visualization Platform (http://r2.amc.nl) was used to examine the Glioblasoma-The Cancer Genome Atlas (TCGA)-540 dataset (22). Astrocyte irradiation signature score was calculated using the z-score of the top 100 upregulated genes. Kaplan–Meier analysis using the scanning method was applied to the signature score using overall survival data and Bonferoni adjusted P-values of <0.05 were considered significant. Data from the Allen Institute for Brain Science IVY-GAP (http://glioblastoma.alleninstitute.org; RRID:SCR_005044) and the Chinese Glioma Genome Atlas (CGGA; ref. 23) were analyzed using the GlioVis data portal for visualization and analysis of brain tumor expression datasets (24, 25).

Freshly dissected brains were sliced into 300 μmol/L slices using a 5,100 mz vibrating blade tissue slicer (Campden Instruments). Slices were supported on polycarbonate tissue culture dish inserts with 0.4 μmol/L pore size for culture at 37°C, 5% CO₂ in HGC Medium. After overnight incubation, slices were treated with 0 or 5 μmol/L of the TGM2 inhibitor GK921 daily for 72 hours. Slices were PFA fixed then cryosectioned. Percent Olig2-positive nuclei were quantified using Cell Profiler (26).

All statistical analysis was performed in GraphPad Prism vs. 8.1.2 (RRID:SCR_002798). Statistical tests and number of replicates are indicated in figure legends. All t tests were two-tailed.

Human biological samples were collected according to the Swedish guidelines for use of clinical material, which includes written informed consent from patients and approvals by the regional ethical committee (Dnr.2018/37). IDH mutation status was determined according to clinical routine.

We devised a coculture scheme to test control or irradiated stromal cells for their ability to increase one glioma stemness indicator, the side population (SP; ref. 19). Stromal cells were irradiated before coculturing with nonirradiated PDGFB-induced glioma primary cells (PIGPC) derived from RCAS-PDGFB-induced gliomas in Nestin-tv-a Ink4a/Arf^−/− mice (PDG mice). Nonirradiated PIGPC were added to the pre-irradiated cells for coculture for 48 hours, followed by an SP assay (Fig. 1A).

We pretreated mouse brain endothelial cells (BEND3; Fig. 1B), human microglia (HMC3; Fig. 1C), primary human brain vascular pericytes (HBVP; Fig. 1D), and primary human astrocytes derived from individual donors (Fig. 1E) with 0 to 10 Gy radiation. PIGPC were used as controls. Only coculture with pre-irradiated astrocytes stimulated a consistent increase in SP of PIGPCs (Fig. 1E; Supplementary Figs. S1A and S1B), suggesting that irradiated astrocytes support glioma stemness. No SP was detected in stromal cells (Supplementary Figs. S1C and S1D). Coculture of PIGPC with irradiated PIGPC did not increase SP (Fig. 1B–E).

Coculture of astrocytes with glioma cells may increase the glioma cell SP through cell–cell contact, soluble factors, or modification of the extracellular matrix (ECM). We created conditioned media to test soluble factors by suspending astrocytes in sodium alginate beads followed by treatment with 0 to 10 Gy, then cocultured with PIGPC. No increase in SP of PIGPC cells cultured in conditioned media was detected (Supplementary Figs. S2A–S2C), suggesting the possibility of insoluble factors influencing the SP of PIGPCs.

To generate purified ECM, confluent astrocytes were treated with 0 to 10 Gy, cultured for 10 days, then de-cellularized, leaving behind insoluble matrix proteins (astrocyte derived matrix, ADM). Glioma cells were cultured on matrix from 10 Gy-irradiated astrocytes (ADMi) or nonirradiated astrocytes (ADMc) to examine the influence of ADM on SP, self-renewal, and survival after irradiation (Supplementary Fig. S2A).

PIGPC (Fig. 1F) and the serum-free cultured human glioma cell line U3082 (Fig. 1G) showed increased SP when cultured on ADMi, suggesting that changes in ADM proteins after irradiation increased the SP of glioma cells (Fig. 1E). No change was observed in mRNA of the stemness markers NANOG, SOX2, OCT4, and ID1 after culture of U3082 on ADMi (Supplementary Fig. S2D). We therefore looked at a functional assay measuring self-renewal by quantifying sphere-formation. U3082 cells were preconditioned for 5 days on ADMc or ADMi, then plated at clonal densities, followed by assessment of primary and secondary sphere formation. There were no changes in primary sphere numbers (Fig. 1H); however secondary spheres, the more accurate measure of self-renewal (27), increased from cells preconditioned on ADMi compared with ADMc (Fig. 1H). These data suggest that astrocytes modify the ECM after irradiation to support stemness.

To determine whether ADMi promotes radioresistance of glioma cells, we plated clonal densities of human glioma U251 cells on ADMc or ADMi, followed by irradiation at 4 Gy (Fig. 1I). Glioma cells cultured on ADMi showed increased colony formation after irradiation compared to cells cultured on ADMc (Fig. 1I), suggesting that matrix from irradiated astrocytes promotes radioresistance of glioma cells.

RNA sequencing of U3082 cells cultured on ADMc or ADMi for 48 hours revealed that culture on ADMi resulted in global gene expression changes of glioma cells. Among gene sets upregulated after culture on ADMi were pathways associated with the ECM, basement membrane, and integrin signaling (Fig. 1J); downregulated ones were associated with translation initiation and oxidative phosphorylation (Fig. 1J). No significant changes were recorded in individual stem cell marker genes including SOX2, ID1, OCT4, and PROM1 (adjusted P = 0.99).

Together, these data suggest that astrocytes modify matrix proteins after irradiation in a manner promoting stemness, including radioresistance of glioma cells.

Finding that tumor-naïve astrocytes irradiated in culture can promote the SP of glioma cells, we sought to determine whether astrocytes in tumor-naïve brains show increased activation in response to irradiation. We stained brains from tumor-naïve mice treated with 0 to 10 Gy for GFAP, a marker of astrocyte activation, and identified a trend of increased GFAP in the brains of irradiated mice (Fig. 2A–E), consistent with previous findings at higher radiation doses (28).

We tested whether radiation-induced changes in the microenvironment influence tumor growth by pre-irradiating mice with 0 to 10 Gy to the brain, followed by intracranial injection with PIGPC 72 hours post-irradiation. Mice were sacrificed upon displaying glioma symptoms (Fig. 2F). Pre-irradiated mice developed tumors with increased microvascular proliferation (MVP; Fig. 2G–I) and increased Ki67-positive nuclei (Fig. 2J and K), indicators of more aggressive, high-grade glioma. Astrocytes were abundant within the tumor and associated with sites of MVP (Fig. 2L), one site enriched with stem-like glioma cells (11). These findings suggest that irradiation of the microenvironment supports tumor growth.

Astrocytes are an abundant cell type in the brain and in gliomas and are likely to survive radiotherapy (29). It is established that astrocytes respond to damage to the brain (30), including radiation (28) in a process termed reactive gliosis. Although this response varies, some hallmarks are commonly seen, including GFAP and vimentin upregulation and morphologic reorganization (30). Notably, tumors frequently recur in the astrogliotic regions overlapping with previous irradiation fields (3).

To understand the response of astrocytes to radiation, we examined purified human astrocytes in vitro and astrocytes from normal and tumor-bearing mouse brains in vivo. PDG mice were irradiated upon glioma symptoms with 0 to 10 Gy, then sacrificed 72 hours later (Fig. 3A). Whole brain sections were stained for Olig2, which marks glioma cells, and GFAP, which marks astrocytes (Fig. 3B and C). In 0 Gy-treated tumor-bearing brains, reactive astrocytes were enriched at the tumor borders and were found throughout the tumor, as seen in whole brain sections (Fig. 3B). Tumor area is demarcated with a line. Astrocytes were found within stromal compartments (white arrow in high-magnification, merged image) and among the Olig2-positive glioma cells (blue arrow; Fig. 3B).

After irradiation, Olig2-positive cells decreased to nearly undetectable levels; however, GFAP staining was retained in areas of the likely original tumor volume (Fig. 3C, white borders, identified by GFAP staining). This can be quantified by measuring the ratio of GFAP to Olig2 in the areas demarcated by strong GFAP staining (white borders; Fig. 3D), which revealed an increased ratio of GFAP to Olig2 in irradiated brains compared with control.

Because in vitro cultivation of primary astrocytes shifts them to a more reactive phenotype than seen in vivo, we tested whether primary cultured human astrocytes underwent features of reactive gliosis. Astrocytes exposed to 10 Gy developed the classic morphology of somatic hypertrophy (Fig. 3E) and expressed elevated levels of the reactive marker vimentin (Fig. 3F; ref. 30). Somatic hypertrophy and elevated vimentin could also be seen after three daily doses of 2 Gy (Supplementary Figs. S3A–S3C). Furthermore, irradiated astrocytes expressed increased levels of the soluble proteins IL6 (Fig. 3G) and IL8 (Fig. 3H), which remained for up to 10 days, indicating a persistent radiation response.

To characterize the astrocytic response to radiation in culture, we performed RNA sequencing on purified human astrocytes 24 hours after a dose of 0 to 10 Gy in vitro. Irradiated astrocytes had a distinct gene expression pattern from control astrocytes (Fig. 3I), including upregulation of GFAP (P = 4.6E−16). Gene set enrichment data revealed an atypical response to irradiation compared with other cell types, such as suppression of the p53 pathway, which has been seen previously in astrocytes (31), as well as elevation of gene sets involved in DNA repair and cell cycle (Supplementary Table S1).

A gene signature consisting of the 100 most upregulated genes in irradiated astrocytes as compared with controls was significantly associated with worse survival in GBM, based on analysis of TCGA GBM dataset (Fig. 3J; ref. 22).

To identify proteins in the ECM responsible for increased radioresistance of cells cultured on ADMi, we performed mass LC/MS-MS analysis on ADMc and ADMi. Supplementary Table S2 lists proteins in the matrix that showed the greatest changes. Among the potential targets identified within this altered ECM, we selected TGM2 for further inquiry because of its proposed roles in matrix modification and stemness (32, 33), as well as promising results in a previous glioma xenograft model (33, 34) and reports that it is a prognostic indicator of shorter time to relapse after chemoradiotherapy (35).

We first confirmed by Western blot increased expression of TGM2 in matrix from astrocytes from different donors (Fig. 4A). We also confirmed increased TGM2 in total cell lysates from astrocytes after stimulation with 3 or 10 Gy (Fig. 4B). Previous studies have shown that TGM2 in glioma cells is upregulated in response to TNFα stimulation (34). In primary human astrocytes, stimulation with TNFα increased expression of TGM2 (Fig. 4C) and TNFα was elevated after irradiation (Fig. 4D), suggesting that TGM2 increase after irradiation may be dependent on TNFα.

We followed this by examining TGM2 expression in vivo in PDG mice (Fig. 3A). Whole brain sections were stained for Olig2, TGM2, and GFAP (Fig. 4E and F). Relative to nontumor bearing brain tissue, TGM2 expression was elevated in tumors (white bounded area, Fig. 4E), was extensively expressed in the stromal compartment, and showed areas of co-staining with GFAP (Fig. 4E, example: blue arrow, merged high magnification image). In the tumors treated with 10 Gy (Fig. 4F), TGM2 expression was retained in the likely original tumor volume and even increased, resulting in a greater percent area covered by TGM2 (Fig. 4G). Within the area of TGM2-positive staining, quantification of TGM2 relative to Olig2 (Fig. 4H) indicates that TGM2 remains even after the Olig2-positive tumor cells have been depleted by irradiation. In the irradiated tumors, regions of TGM2 staining were also found in close association with GFAP-positive astrocytes, often co-staining or showing astrocytes surrounding areas of intense TGM2-positivity (Fig. 4F, example: blue arrow, merged high magnification image).

Like GFAP, TGM2 expression is elevated in areas of MVP and the normal vasculature (Fig. 4I). Comparable to our findings in mice, analysis of the Ivy Glioblastoma Atlas Project (IVY-GAP; ref. 24) reveals that TGM2 is highly elevated in areas of MVP (Supplementary Fig. S4).

Taken together, these data suggest that irradiation promotes increased expression of TGM2 from astrocytes, potentially providing a radio-protective environment for a newly expanding tumor. We therefore returned to our in vitro experimental systems to determine if there was a functional relationship between TGM2 and stemness and radiation resistance of glioma cells.

To determine whether TGM2 expression might be associated with glioma cell stemness or radiation resistance, we examined the effects of either purified TGM2 or TGM2 inhibitors on our in vitro model of glioma cell–astrocyte matrix interaction. Purified TGM2, when coated onto ADMc, raised the SP of PIGPC near to that seen with PIGPC cultured on ADMi (Fig. 5A). In contrast, two different TGM2 inhibitors, GK921 (Fig. 5B) and dansylcadaverine (DC; Fig. 5B), decreased the SP of PIGPC cultured on ADMi to the levels seen on ADMc. GK921 also decreased survival after irradiation of PIGPC cultured on ADMi to levels seen on ADMc, while not affecting PIGPC cultured on ADMc (Fig. 5C; Supplementary Fig. S5). These data suggest that extracellular TGM2 contributes to the increased stemness and radiation resistance of glioma cells cultured on ADMi as compared with those cultured on ADMc.

In vivo, tumors treated with 10 Gy increased CD44 expression (Fig. 5D–G), a stem cell marker in glioma (36) and a marker of mesenchymal subtype GBM. CD44 strongly colocalized with TGM2 (Fig. 5F and G). Like TGM2, CD44 expression was retained in the likely original tumor volume after irradiation, even as the Olig2-positive cells receded, as seen by increased expression relative to Olig2 (Fig. 5G). These data provide evidence that TGM2 is associated with increased stemness in vivo and is elevated after irradiation.

We next sought to determine which TGM2-stimulated signaling pathways may promote stemness in glioma cells. We performed a phosphokinase array to screen for potential changes in signaling in U3082 cells cultured on plates coated with 3.3 mU/cm² TGM2 (Fig. 6A and B; Supplementary Table S3). Among the most highly increased phosphorylation events, proteins involved in the integrin-β1 pathway were increased, including GSK3β, Fgr, β-catenin, EGFR, JNK1/2/3, and Src (37). GSEA of U3082 cultured on ADMc and ADMi indicated a trend toward enrichment of the Integrin1 pathway in cells cultured on ADMi (Fig. 6C). In addition, integrin-β1 is linked to fibronectin signaling, which is directly affected by TGM2 (38). We therefore examined expression of integrin-β1 in tumors from PDG mice 72 hours post-irradiation. Integrin-β1 expression was closely associated with and frequently costained with TGM2 in tumors (Fig. 6D). Like TGM2, integrin-β1 remained in the original tumor volume after irradiation even after Olig2-positive cells were depleted, as can be seen when measured as ratio of integrin-β1 to Olig2 (Fig. 6E and F). According to the IVY-GAP database, like TGM2, integrin-β1 is associated with MVP (Fig. 6G). Furthermore, there is a significant correlation between TGM2 expression and the PID_INTEGRIN1_ PATHWAY in the TCGA GBM dataset (Fig. 6H) when analyzed in the Gene Expression Profiling Interactive Analysis (GEPIA2) web server (39).

To determine whether there was a functional relationship between the integrin-β1 pathway and the ability of TGM2 to promote stemness features of glioma cells, we examined whether inhibition of Src family kinases with dasatinib could block the stemness promoting properties of TGM2. We performed a limiting dilution self-renewal assay on U3082 preconditioned on either 0 or 3.3 mU/cm² TGM2 for 4 days. Preconditioned U3082 were allowed to form primary spheres in the presence or absence of 40 mU/ml TGM2, then dissociated spheres were replated in a limiting dilution assay in the presence or absence of TGM2 and 10 nmol/L dasatinib. TGM2 increased sphere-forming potential of U3082 relative to control cells while dasatinib decreased sphere formation in TGM2-treated cells but not in control cells (Fig. 6I; Supplementary Table S4). Taken together, these data suggest that stemness is promoted by TGM2 through a Src-related pathway as part of enhanced integrin-β1 signaling.

To further examine the effects of TGM2 inhibition, we used an ex vivo organotypic slice model of GBM. We initiated tumors in PDG mice by injection with RCAS-PDGFB and shp53, then sliced freshly-harvested brains upon detection of glioma symptoms. Slices were cultured overnight, followed by treatment with 0 or 5 μmol/L of the TGM2 inhibitor GK921 for 72 hours, then fixed and cryosectioned, followed by immunofluorescent detection of Olig2 to determine the presence of tumor cells. TGM2 inhibitor significantly decreased the percent of Olig2-positive nuclei in the sections, in some cases reducing Olig2-positive cells to undetectable levels (Fig. 7A–C).

To examine TGM2 expression in a model of recurrent glioma, we initiated tumors in PDG mice with RCAS-PDGFB and shPten. Mice were treated at symptoms with either 0 Gy or a fractionated dose of 2 Gy daily for 5 days, then sacrificed upon recurrence of glioma symptoms. Immunodetection of TGM2 and Olig2 revealed that elevated TGM2 is retained in the tumor recurrence (Fig. 7D–G), showing increases both relative to Olig2 (Fig. 7F) as well as in percent area covered in the tumor (Fig. 7G) as compared to the untreated primary tumors.

A small sample of human isocitrate dehydrogenase wild type (IDHwt) GBM showed elevated TGM2 protein expression in recurrent versus primary GBM (Fig. 7H and I). Recurrent tumors were from patients receiving radiotherapy of 60 Gy and temozolomide adjuvant therapy. While the sample size is small, this finding, in combination with the results from our mouse model, suggests higher TGM2 has the potential to remain in the GBM microenvironment at recurrence. This is further supported by an analysis of recurrent tumors in the TCGA and CGGA database, which both show a small but significant increase in TGM2 mRNA in recurrent tumors compared to primary tumors (Supplementary Figs. S6A and S6B).

In primary human glioma, TGM2 mRNA is elevated as tumor grade increases (Fig. 7J). TGM2 mRNA is significantly increased in IDHwt tumors (Fig. 7K), the most aggressive tumor classification. Elevated TGM2 correlates with shorter survival times (Fig. 7L). Although these shorter survival times are possibly the result of the association of elevated TGM2 with increasing grade, it is notable that within IDHwt GBM, tumors with the top 10% of TGM2 mRNA expression have the shortest survival times (Fig. 7M). Taken together, these data suggest that TGM2 may be a viable target in GBM.

On the basis of our findings, we propose a model in which radiation stimulates tumor-associated astrocytes to secrete TGM2, which then promotes stemness and survival of glioma cells, allowing tumor recurrence (Fig. 7N).

GBM treatment options are limited after tumor recurrence. As surgery can rarely eliminate all GBM cells, identifying and targeting pathways that support GBM cell survival after radiotherapy is essential to improving outcomes. To this end, a better understanding of the effects of radiation on all aspects of a tumor's biology is needed. Studies so far have identified that radiation can promote the migration and invasion of tumor cells (40, 41), particularly by stimulating inflammation in the tumor microenvironment, such as that mediated by IL1β in a pre-irradiation model of glioma in rat brains (41, 42). Although some studies have examined how the tumor microenvironment can promote radiation resistance and GBM stemness (10, 43–45), little study has been made of how the stromal cells themselves respond to treatment modalities to support these features (46). This is an important question, as findings presented here suggest that stromal cells may respond to radiation in ways that could prime the microenvironment for protection of GBM cells against further radiation insults.

It is likely that effects of radiotherapy on the tumor microenvironment are not limited to its effect on astrocytes described here. Although our coculture screen of pre-irradiated stromal cells detected a SP increase only from pre-irradiated astrocytes, our screen may favor our ability to detect the effect of irradiated astrocytes on glioma cells over other cell types. The use of the SP assay as a readout and the combination of human stromal cells with mouse glioma cells may have limited our findings. Other model systems may reveal further important interactions. However, despite astrocytes being derived from multiple different human donors, which reflects the variability seen in human populations in the TCGA and IVY-GAP datasets, we were still able to detect radiation-driven pathways that resulted in identification of a potential target for glioma therapy.

This study provides a promising new strategy in GBM treatment by examining lasting changes to the tumor microenvironment made by stromal cells in response to radiotherapy, changes that may support radiation resistance of GBM cells. Although previous work has shown that in untreated tumors, astrocytes react to the GBM microenvironment and can promote glioma growth (14), we have identified astrocytes as persisting within the original tumor volume even after the tumor bulk is killed by radiation, altering that environment in a radioprotective manner. We demonstrate that they do this by secreting TGM2, which supports stemness and radioresistance of GBM cells. This is an important addition to studies examining TGM2 expression in glioma cells themselves, and may shed light on previous findings that the TGM2 promoter is hypermethylated in glioma, yet TGM2 expression in GBM is elevated (47). Understanding the role of astrocytes in radiation response and TGM2 release may also further elucidate previous findings that TGM2 inhibition increased survival in a glioma xenograft model (33, 34).

Although we identified an important role for the integrin-β1 and Src pathways in TGM2 mediation of radiation resistance and stemness of glioma cells, the mechanisms by which TGM2 protects GBM should be further explored. Fibronectin bundling, perhaps the most well-described function of TGM2 (38), influences radiation resistance (48), angiogensis (49), and invasiveness, and integrin-β1 signaling (50), suggesting further study of fibronectin bundling and other matrix alterations post-irradiation is warranted. Our findings suggest a new strategy for sensitizing GBM cells to radiotherapy by short-circuiting the normal brain's protective response to radiation.

A. Pietras reports grants from Ragnar Söderberg Foundation, The Swedish Cancer Society, The Swedish Research Council, The Swedish Childhood Cancer Fund, Ollie & Elof Ericssons Foundation, Jeanssons Stiftelser, The Crafoord Foundation, Gösta Miltons Donationsfond, and Stiftelsen Cancera during the conduct of the study. No disclosures were reported by the other authors.

T.J. Berg: Conceptualization, data curation, formal analysis, investigation, methodology, writing–original draft, writing–review and editing. C. Marques: Formal analysis, investigation, writing–review and editing. V. Pantazopoulou: Formal analysis, investigation, writing–review and editing. E. Johansson: Formal analysis, investigation, writing–review and editing. K. von Stedingk: Formal analysis, investigation, writing–review and editing. D. Lindgren: Formal analysis, investigation, writing–review and editing. P. Jeannot: Formal analysis, investigation, writing–review and editing. E.J. Pietras: Formal analysis, investigation, writing–review and editing. T. Bergström: Formal analysis, investigation, writing–review and editing. F.J. Swartling: Formal analysis, investigation, writing–review and editing. V. Governa: Investigation, writing–review and editing. J. Bengzon: Formal analysis, investigation, writing–review and editing. M. Belting: Formal analysis, supervision, investigation, writing–review and editing. H. Axelson: Formal analysis, supervision, investigation, writing–review and editing. M. Squatrito: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, investigation, methodology, writing–original draft, project administration, writing–review and editing. A. Pietras: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, investigation, methodology, writing–original draft, project administration, writing–review and editing.

The authors thank Christina Möller for technical assistance. The authors acknowledge support from Science for Life Laboratory, the National Genomics Infrastructure, NGI, and Uppmax for providing assistance in massive parallel sequencing and computational infrastructure. They acknowledge the Center for Translational Genomics, Lund University and Clinical Genomics Lund, SciLifeLab Sequencing for sequencing performed in their labs. They acknowledge the Preclinical Cancer Treatment Center, SciLifeLab, Uppsala University, for technical assistance and support with radiation therapies experiments in vivo. Support from the Swedish National Infrastructure for Biological Mass Spectrometry is gratefully acknowledged. The results shown here are in whole or part based upon data generated by the TCGA Research Network: https://www.cancer.gov/tcga.

This study was supported by grants from the Ragnar Söderberg Foundation, the Swedish Cancer Society, the Swedish Research Council, the Swedish Childhood Cancer Fund, Ollie & Elof Ericssons Foundation, Jeanssons Stiftelser, the Crafoord Foundation, Gösta Miltons Donationsfond, Stiftelsen Cancera, donations from Viveca Jeppson and Maj-Britt and Allan Johansson and the Seve Ballesteros Foundation. J. Bengzon was supported by Region Skane and ALF.

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.