Microbial peptides activate tumour-infiltrating lymphocytes in glioblastoma – Nature

Patient information and materials

Patient 1635WI, who was treated with a personalized neopeptide vaccination, was a 55-year-old male diagnosed with glioblastoma. After resection, he was treated with standard radiochemotherapy and additionally with the immune checkpoint inhibitor pembrolizumab (anti-PD1) and the anti-VEGF monoclonal antibody bevacizumab. He was interested in participating in an individual treatment attempt with a personalized peptide vaccination therapy²⁵. He was fully informed about the potential risks of such a vaccination, which, by definition, is an individual medical treatment (‘compassionate use’) and not subject to special regulations for medical research according to the Therapeutic Products Act (TPA) or Swiss Federal Human Research Act (HRA). Consequently, individual medical treatments are not approved by the Ethics Committee or Swiss Agency for Therapeutic Products (Swissmedic), as they do not constitute clinical trials for the purpose of systematically gaining knowledge. The patient gave written informed consent to receive the individualized vaccination.

The cohort of 18 additional patients with glioblastoma includes the primary tumour of 16 patients and 2 recurrent tumours from 2 patients, all from the University Hospital Zurich, Zurich, Switzerland. Tumour tissues were immediately frozen after surgery and stored at −80 °C until immunoaffinity purification of HLA–peptide complexes. Autologous tumour cells were also isolated and frozen from several samples (n = 6) and immunopeptidomes were analysed (Supplementary Table 1). Written informed consent was obtained from each patient in accordance with the local ethical requirements (KEK-ZH-Nr. 2015-0163). All included patients underwent elective brain tumour surgery at the Department of Neurosurgery at the University Hospital Zurich and were diagnosed with glioblastoma (IDH1 wild type) by a board-certified neuropathologist. No other inclusion or exclusion criteria were applied so we do not expect other biases. Sample size was determined based on the availability of biological materials. To generate autologous tumour cell lines, tumour tissue was washed thoroughly using PBS. Next, they were cut into small pieces and treated with digestion mix including RPMI-1640 medium (Sigma-Aldrich) containing 10% FCS (Eurobio), 1 mg ml⁻¹ of collagenase IV (Sigma-Aldrich) and 0.1 mg ml−1 of DNase I (Sigma-Aldrich), and incubated for 42 min at 37 °C. The reaction was stopped by adding PBS with 2 mM EDTA (Invitrogen) and the lysate was passed through a filter mesh. Next, tumour cells were isolated using Percoll (Sigma-Aldrich) gradient. Isolated cancer cells were passaged up to 30 times to make sure no other cells would survive before they were snap frozen and used for the immunopeptidome analyses. All cells were regularly tested for mycoplasma contamination and were negative.

DNA extraction and genotyping

DNA was extracted from PBMCs, tumour cells and tissues of patients with glioblastoma using DNeasy Blood & Tissue Kits (Qiagen). The genotyping of HLA class I and HLA class II was done by high-resolution HLA sequence-based typing (Histogenetics) (Supplementary Table 2).

PBMCs, cell lines and TCCs

PBMCs of patient 1635WI were isolated from a leukapheresis, and allogenic PBMCs were isolated from buffy coats by Ficoll (Eurobio) density gradient centrifugation.

BLS-DRB1*03:01 (kindly gifted by W. W. Kwok, Benaroya Institute, WA, USA), BLS-DRB1*04:02, BLS-DRB3*02:02 and BLS-DRB4*01:01 are EBV-infected B cells from a patient with BLS type II that are transfected with single HLA-DR molecules of the patient with glioblastoma. Thus, each BLS-DR transfectant expresses a single HLA-DR molecule. In brief, various HLA-DRB complementary DNAs were inserted after the CMV promoter in a pLenti-CMV vector (kindly provided by M. Scharl, University of Zurich). Plasmid vectors carrying different HLA-DRBs were used to prepare lentiviruses using pCMV-dR8.91 (packaging plasmid) and pMD2.G (VSV-G envelope plasmid) (both kindly gifted by M. Scharl). Next, BLS cells expressing HLA-DRA1*01:01 (kindly provided by W. W. Kwok) were transduced with various HLA-DRB lentiviruses and cells were selected using blasticidine after 7 days. BLS cells were maintained and expanded using RPMI-1640 medium (Sigma-Aldrich) containing 10% heat-inactivated FCS (Eurobio), 2 mM l-glutamine (Thermo Fisher Scientific), 100 U ml⁻¹ penicillin (Corning) and 100 µg ml⁻¹ streptomycin (Corning). BLS cell lines were regularly examined for expression of HLA-DR via FACS. Bulk TILs and isolated TCCs including TCC88 and other TCCs were expanded using irradiated allogenic PBMCs (45 Gy) in T cell medium containing IMDM (GE Healthcare), 5% heat-inactivated human serum (Blood Bank Basel, Switzerland), 1 μg ml⁻¹ PHA, 20 U ml⁻¹ of human IL-2 (h-IL-2), 2 mM l-glutamine (Thermo Fisher Scientific), 100 U ml⁻¹ penicillin (Corning) and 100 µg ml⁻¹ streptomycin (Corning). The medium was changed every 3–4 days and new h-IL-2 was added each time.

Proliferation assay

All peptides were synthesized with acetylated N terminus and amidated C terminus (Peptides & Elephants) (Supplementary Table 10). Proliferation of TCCs and bulk TILs was measured using the ³H-thymidine incorporation (Hartmann Analytic) assay after 3 and 5 days, respectively. Isolated TCCs (2 × 10⁴ cells per well) were co-cultured with irradiated BLS cells (300 Gy) (5 × 10⁴ cells per well) in 200 µl of X-VIVO 15 medium (Lonza) and seeded in 96-well U-bottom plates (Greiner Bio-One). Bulk TILs (5 × 10⁴ cells per well) were co-cultured with irradiated autologous PBMCs (45 Gy) (2 × 10⁵ cells per well) in 200 µl of X-VIVO 15 medium (Lonza) and seeded in 96-well U-bottom plates (Greiner Bio-One). Peptides were then used at a final concentration of 10 µM to stimulate the TCCs or bulk TILs. Results are shown as SI, that is, the ratio of count per minute in the presence of the peptides divided by the negative control (no peptide). SI ≥ 2 is considered positive.

For testing peripheral blood memory T cells, CD45RA⁻ cells were isolated using human CD45RA microbeads (Miltenyi). Negatively selected CD45RA⁻ cells were seeded (2 × 10⁵ cells per well) in 200 µl of X-VIVO 15 medium (Lonza) in 96-well U-bottom plates (Greiner Bio-One). Peptides were then added at a final concentration of 10 µM and incubated for 7 days before measuring their proliferation via ³H-thymidine incorporation.

To isolate antigen-reactive T cells from the TILs or peripheral T cells, CFSE (Sigma-Aldrich) was used to label the cells. In brief, cells were incubated with CFSE (final concentration of 0.5 µM) for 3 min at room temperature. Next, the labelling was immediately stopped using a serum-containing medium and washed before cells were seeded (5 × 10⁴ cells per well) in the presence of irradiated autologous PBMCs (45 Gy) (2 × 10⁵ cells per well) in 200 µl of X-VIVO 15 medium (Lonza) and the peptide of interest. After 10 days, cells were pooled, washed once with X-VIVO 15 medium (Lonza) and stained with live-dead yellow (Invitrogen) as well as surface markers and cells sorted with a SH800S Cell Sorter (Sony).

Sorting bulk T cells

TILs expressing specific TCRBVs were isolated from bulk TILs using a panel of 22 fluorescently labelled TCRBV antibodies that detect 27 different TCRBVs. Twelve TCRBV antibodies were labelled with PE and 10 with FITC (Beckman Coulter). Five million TILs that had been expanded once with the above protocol were thawed and washed once with PBS before incubation with human IgG (1:25) (Sigma-Aldrich) for 30 min to block Fc fragments and reduce nonspecific antibody staining. After another wash, cells were stained with CD4 (1:100) (APC anti-human CD4, eBioscience), CD8 (1:100) (Pacific blue anti-human CD8, BioLegend), TCRBV-PE (1:25) (Beckman Coulter) and TCRBV-FITC (1:25) (Beckman Coulter) antibodies (Supplementary Table 12) for 30 min. Finally, cells were sorted, seeded at 5 × 10⁴ cells per well in T cell medium without h-IL-2 and rested for 1 day before stimulation with 2 × 10⁵ cells per well of irradiated autologous PBMCs (45 Gy) and peptide (10 µM).

Furthermore, TILs were labelled with CFSE and 5 × 10⁴ cells per well were stimulated with irradiated autologous PBMCs (2 × 10⁵ cells per well) and pools of 22 glioblastoma-associated peptides that had been used in an individualized vaccination of patient 1635WI²⁵ or 23 bacteria-derived and microbiota-derived peptides, which had been identified by an unbiased target search for TCC88 (Supplementary Table 13) (see below). Both peptide pools have a final concentration of approximately 25 µM for all the peptides in the pool, and the DMSO concentration was always lower than 0.5%.

Regarding isolation of peripheral memory T cells, CD45RA⁻ cells were sorted using human CD45RA microbeads (Miltenyi) and labelled with CFSE. Next, 2 × 10⁵ cells per well were seeded in 200 µl of X-VIVO 15 medium (Lonza) in 96-well U-bottom plates (Greiner Bio-One) in the presence of the bacteria-derived and microbiota-derived peptide pool or the IPdBP pool. The IPdBP pool contains 37 HLA class II-derived bacterial peptides at a final concentration of 20 µM for all the peptides in the pool (Supplementary Table 13). After 10 days, cells were Fc-blocked using human IgG (1:25) (Sigma-Aldrich) and labelled with live-dead yellow (1:1,000) (Invitrogen), PerCP-Cy5.5 anti-human CD3 (1:100) (BioLegend), APC anti-human CD45RA (1:50) (BioLegend), APC-Cy7 anti-human CD4 (1:100) (BioLegend) and Pacific blue anti-human CD8 (1:100) (BioLegend). CD45RA⁻CD3⁺CD4⁺CFSE^dim cells were then sorted and 2 × 10⁴ cells per well were seeded in 96-well U-bottom plates (Greiner Bio-One) with irradiated allogenic PBMCs, PHA and h-IL-2 as described above. The medium was changed every 3–4 days and new h-IL-2 was added each time. If necessary, each well was split into two. After 14 days, cells were cryopreserved.

T cell cloning and characterization

To generate CD4⁺ TCCs, bulk TILs were labelled with CFSE and stimulated with SIN3A* peptide (10 µM). After 10 days, live proliferative (CFSE^dim), CD3⁺ (1:100) (PerCP/Cy5.5 anti-human CD3 antibody, BioLegend) and CD4⁺ (1:100) (APC-Cy7 anti-human CD4 antibody, BioLegend) T cells were single-cell sorted into 96-well U-bottom plates (Greiner Bio-One). Cells were then expanded using irradiated allogenic PBMCs in the presence of PHA and h-IL-2 in T cell medium. The medium was changed every 3–4 days and h-IL-2 was added each time until day 12. Cells were then subjected to rapid expansion to expand TCCs. TCCs (5 × 10⁴ cells) were transferred into T25 flasks (TPP) and were expanded using 3 × 10⁷ irradiated allogenic PBMCs (45 Gy) in T cell medium containing anti-CD3 monoclonal antibody (OKT3) (Ortho Biotech). After 1 day, h-IL-2 was added at 50 U ml⁻¹; at day 5, the content of the flask was harvested, spun down and resuspended in new T cell medium containing 50 U ml⁻¹ of h-IL-2. At day 8, cells were split into new T25 flasks and 20 U ml⁻¹ of h-IL-2 was added. At day 16, cells were harvested, washed and cryopreserved in freezing medium including 90% heat-inactivated FCS and 10% DMSO (Applichem).

TCCs were characterized using primers for TCRVβ chains and further sequencing of amplicons³¹. In addition, the LEGENDplex Multi-Analyte Flow Assay kit (BioLegend) was used to identify the functional phenotype of TCC by measuring multiple cytokines in the supernatant following activation with anti-CD2/CD3/CD28 antibody-loaded MACSibead (Miltenyi). The HLA-DR restriction of TCCs was defined by co-culturing TCCs with the four aforementioned irradiated HLA-DR transfected BLS cells (300 Gy), presenting different concentrations of the cognate antigen and measuring proliferation after 3 days using ³H-thymidine incorporation.

Cytokine measurements

Supernatants from each experiment were harvested a day before adding thymidine and frozen at −20 °C. Cytokines were measured using ELISA for human IFNγ ELISA MAX (BioLegend), human GM-CSF DuoSet ELISA (R&D Systems) and bead-based assays. A Synergy H1 hybrid multi-mode reader (Agilent) was used to measure the absorbance. The LEGENDplex Multi-analyte Flow Assay kit (BioLegend) is a 13-plex human T_H cytokine panel, which includes IFNγ, TNF, IL-2, IL-4, IL-5, IL-6, IL-9, IL-10, IL-13, IL-17A, IL-17F, IL-21 and IL-22. Staining was conducted as suggested by the manufacturer. LSR Fortessa Flow Cytometer (BD Biosciences) was used to measure cytokines, and data were analysed using FlowJo (Tree Star). The Human XL Cytokine Luminex Performance Assay Fixed Panel (R&D Systems) contains 45 analytes covering a broad range of various chemokines and cytokines. Bio-Plex 200 systems (Bio-Rad) was used to measure cytokine and chemokine secretion as suggested by the manufacturer.

ps-SCL and biometrical analysis

A library of l-amino acid decapeptide ps-SCL mixtures was prepared as mentioned before³⁰. TCC88 is a CD4⁺ TCC isolated from the TILs and is restricted by HLA-DRB3*02:02. Irradiated BLS-DRB3*02:02 cells (300 Gy) were incubated with 200 mixtures from the library at 200 and 100 µg ml⁻¹ concentration. TCC88 was then co-cultured with these cells. Each well contained 2 × 10⁴ TCC88 and 5 × 10⁴ BLS cells. After 72 h, supernatants were harvested and tested for IFNγ secretion by ELISA (BioLegend). Next, a scoring matrix was generated via log₁₀ transformation of median IFNγ values of three independent experiments²⁶. Next, biometrical analysis²⁹ was used to score all 10-mers from several large databases, that is, all human proteins (Swiss-Prot-20201202), the translated transcriptomes of the primary and recurrent autologous glioblastoma tumour (patient 1635WI), all viruses, which are annotated for infecting humans in UniProt (Swiss-Prot-20210305), as well as all bacteria (Swiss-Prot-20201202) and human gut microbiota³⁴. To assess the validity of the in silico predictions of stimulatory peptides for TCC88, consecutive 10-mer peptides with the highest predicted scores in each database were synthesized and tested for their capacity to stimulate TCC88. Peptides were selected based on their predicted stimulatory scores and not other factors such as their source. Peptide scores ranged between 80% and 100% of the maximum theoretical score for all databases beside the tumour immunopeptidome (annotated by human proteins), which contained only a few peptides with a score over 80%. Therefore, we also synthesized several peptides within an extended score range between 60% and 80%. Peptides were tested at a final concentration of 10 µM on irradiated BLS-DRB3*02:02 and co-cultured with TCC88 as mentioned before. ³H-thymidine incorporation (Hartmann Analytic) was used to measure proliferation.

T cell cytotoxicity assay

In brief, TCC88 (2 × 10⁴ cells per well) was stimulated using the tumour antigen SIN3A* and Victivallis gut bacterium HGM3 peptides (2 µM) separately on irradiated autologous PBMCs (1 × 10⁵ cells per well) for 72 h in X-VIVO 15 medium (Lonza). Next, the cells were pooled and washed once with X-VIVO 15 medium (Lonza). Autologous glioblastoma cells were stained with PKH26 Red Fluorescent Cell Linker Kit (Sigma-Aldrich) for 3 min and immediately washed three times with serum-containing medium. Target cells were then resuspended in X-VIVO 15 medium (Lonza) and seeded at 2 × 10⁴ cells per well. Activated and rested TCC88 cells were co-cultured at 1:1 and 5:1 effector to target ratios in duplicates. Next, the plate was centrifuged at 80g for 1 min to ensure cell–cell contact. After 24 h, supernatants were removed and cells were detached using accutase (Thermo Fisher Scientific), washed and stained with LIVE/DEAD Fixable Near-IR Dead Cell Stain Kit (Thermo Scientific), and the percentage of PKH26⁺Near-IR⁺ dead tumour cells was assessed using BD LSR Fortessa. HLA-DR expression of cancer cells was also measured using PE-Cy7 anti-human HLA-DR antibody (1:50) (BioLegend). To calculate the specific T cell-mediated cytotoxicity, target cells were separately seeded without TCC88 and their death (%) was considered as unspecific background death.

TCRVβ sequencing

To compare the TCRVβ sequence of TILs responding to the vaccine peptide pool with those responding to the bacteria-derived and microbiota-derived peptide pool, CFSE-labelled TILs were seeded at 5 × 10⁴ cells per well in a 96-well U-bottom plates (Greiner Bio-One) and stimulated with the corresponding peptide pools incubated with irradiated autologous PBMCs (2 × 10⁵ cells per well). After 10 days, between 5 × 10³ and 10 × 10³ live CD3⁺CD4⁺CFSE^dim cells were sorted from each sample. Moreover, to isolate peripheral T cells responding to bacteria-derived and microbiota-derived peptides or the IPdBP pool, CFSE-labelled cells were sorted. Sorted cells from both experiments were spun down and supernatants were removed. The pellet was snap-frozen immediately. The DNeasy Blood & Tissue Kit (Qiagen) was used to extract the DNA, and amplification and sequencing of TCRVβ CDR3 was performed using the ImmunoSEQ platform (Adaptive Biotechnologies). For the TCR data analysis, non-productive TCRVβ sequences were excluded, and shared unique productive TCRVβ sequences were specified based on identical CDR3 sequence as well as V and J chains. Representation of data was done using VENNY^2.1 (BioinfoGP, CNB-CSIC)⁵¹.

IFNγ fluorospot assay

TILs (2 × 10⁴ cells per well) were seeded with 5 × 10⁴ irradiated (300 Gy) autologous EBV-transformed B cell line cells per well as APCs primed with bacteria-derived and microbiota-derived peptides (10 µM final concentration). Cells were seeded in X-VIVO 15 medium (Lonza) in human IFNγ/IL-10/IL-17A pre-coated fluorospot plates (FSP-010703-10, Mabtech) and were incubated for 44 h at 37 °C. Before seeding, fluorospot plates were washed three times using PBS and then blocked with RPMI-1640 medium (Sigma-Aldrich) containing 10% FCS (Eurobio) as instructed by the manufacturer. After 44 h of incubation, plates were washed using PBS and incubated with detection antibody, anti-IFNγ monoclonal antibody 7-B6-1-BAM, for 2 h. Next, plates were washed using PBS and incubated with anti-BAM-490 fluorophore conjugate for 1 h. Both detection antibody and fluorophore conjugate were diluted with PBS containing 0.1% BSA (Sigma-Aldrich) as instructed by the manufacturer. Next, plates were washed using PBS and incubated with fluorescence enhancer (Mabtech) for 15 min. Plates were emptied by flicking and kept in the dark for 24 h to completely dry. Finally, spot analysis was performed using the automated Mabtech IRIS fluorospot reader (Mabtech).

Immunopeptidome analyses

To analyse the immunopeptidomes of the tumour tissues and cell lines, frozen tissues and frozen cell pellets were homogenized in CHAPS-containing lysis buffer to perform standard immunoaffinity chromatography²⁴. In brief, tumour tissues were washed thoroughly using PBS. Next, they were cut into small pieces and treated with digestion mix including RPMI-1640 medium (Sigma-Aldrich) containing 10% FCS (Eurobio), 1 mg ml⁻¹ of collagenase IV (Sigma-Aldrich) and 0.1 mg ml⁻¹ of DNase I (Sigma-Aldrich), and were incubated for 42 min at 37 °C. The reaction was stopped by adding PBS with 2 mM EDTA (Invitrogen) and the lysates were passed through a filter mesh. Next, tumour cells were isolated using Percoll (Sigma-Aldrich) gradient. HLA class II molecules were isolated using the HLA-DR-specific antibody (L243)⁵² as well as the pan-HLA class II-specific antibody Tü39 (ref. ⁵³) (mixed 1:1, at least 1 mg or 1 mg per 1 g of tissue, both produced in-house at the Department of Immunology, Interfaculty Institute of Cell Biology, University of Tübingen) and subsequent elution of the HLA-bound peptides. Eluated peptides were analysed by nanoflow high-performance liquid chromatography (Dionex UltimateTM 3000 Series liquid chromatography system) and tandem mass spectrometry (MS; Orbitrap Fusion Lumos, Thermo Fisher Scientific). Data-dependent acquisition of data was always conducted in technical triplicates²⁴. The raw MS data files were converted into ‘mgf’ using ThermoRawFileParser⁵⁴ and analysed on MASCOT software with a non-redundant eubacteria UniProtKB/Swiss-Prot protein database (12 December 2020 with 334,492 entries) and search parameters were 0.5-Da fragment mass tolerance and 5 ppm precursor with no enzyme. All spectra were analysed manually. We only considered a spectrum valid if we clearly identified a sequence of four B-ions or Y-ions. The samples were also run on Scaffold (Scaffold version 5.2, Proteome Software) software and the ‘protein prophet’ reported false discovery rates between 2.4 and 5.4. All the acquired sequences were again re-evaluated by comparing them with UniPort human proteins, viral proteins, bacteriophage protein database⁵⁵ and genomic transposable elements (data not shown)⁵⁶. Synthetic IPdBPs were measured with the same MS/MS method and device (Orbitrap Fusion Lumos, Thermo Fisher Scientific) and their MS/MS spectra were compared with those of the immunopeptidome peptides. As our synthetic peptides were synthesized with modifications at the N-terminal (acetylation) and C-terminal (amidation) sites, spectra often look quite different in terms of intensities and the presence or absence of B-ion and Y-ion series. The modification changed the charge of the peptides and the character of ionization patterns. The reason for synthesizing peptides with the above-end modifications stems from a convention that we previously established based on the ps-SCLs search algorithm. The latter decamer peptide mixtures, which are used for the unbiased antigen discovery studies, have N-terminal acetylation and C-terminal amidation, and consequently, we always synthesized the predicted peptides with the same ends. The peptides for the sequence validation of the IPdBPs should have been synthesized with free ends. Despite this limitation, peptide sequencing identified the exact same amino acid sequences for 37 IPdBPs via the synthetic peptides. The comparison of MS/MS spectra of ten synthetic peptides per original spectra of IPdBPs are shown in Supplementary Table 16. Universal Spectrum Explorer (USE) was used to annotate the spectra and Proteome Discoverer (Thermo Fischer) was utilized to do the research in the raw files.

The core binding motif of peptides to DRB1*03:01, DRB1*04:02, DRB3*02:02 and DRB4*01:01 was predicted using the NetMHCII 2.3 server⁵⁷, and iceLogo⁵⁸ (https://iomics.ugent.be/icelogoserver/) was used for graphical representation of the core binding motifs.

16S rRNA gene sequencing

16S rRNA gene sequencing of the tumour tissues was conducted by Microsynth AG. In brief, fresh-frozen tissues from patients were sent to Microsynth AG. Tissues were cut manually using a sterile scalpel, homogenized using Fastprep 24 (45 s, speed 6.5) and total DNA was extracted using ZymoBiomics DNA Mini Kit D4300 (Zymo Research). Negative extraction controls were included. Enrichment PCR of the whole 16S gene (primer pair 27F and 1492R) was performed, followed by amplicon library generation (Nextera library preparation protocol) using the locus-specific primer pair 341F (5′-CCT ACG GGN GGC WGC AG-3′) and 805R (5′-GAC TAC HVG GGT ATC TAA TCC-3′) and sequencing on a MiSeq platform 2 × 250 bp v2 (Illumina).

Reads were demultiplexed by separating reads into individual FastQ files, quality controlled and trimmed of Illumina adaptor sequences using locus-specific bcl2fastq software version v2.20.0.422, FastQC version 0.11.8 (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/) and cutadapt v3.2 (http://journal.embnet.org/index.php/embnetjournal/article/view/200), respectively. Trimmed forward and reverse reads of each paired-end read were merged to in silico reform the sequenced molecule considering a minimum overlap of 15 bases using the software USEARCH (version 11.0.667)⁵⁹. Merged reads were denoised using the UNOISE algorithm implemented in USEARCH to form operational taxonomic units, discarding singletons and chimaeras in the process. The resulting operational taxonomic unit abundance table was then filtered for possible barcode bleed-in contaminations and against the negative controls using the UNCROSS algorithm. Operational taxonomic unit sequences were compared with the reference sequences of the RDP 16S database⁶⁰. In addition, we used the previously reported data on the tumour microbiome²⁰ with in-depth contamination filtering to further remove potential bacterial contaminants.

RNA sequencing

RNA sequencing of the tumour tissue (1635WI) and analysis were conducted as previously reported²⁵. In brief, RNA sequencing was conducted using the HiSeq 4000 System (Illumina) at the Functional Genomics Center Zurich. Quantile normalized gene counts were compared with The Cancer Genome Atlas (TCGA) glioblastoma cohort.

Statistical analyses

All analysis were conducted using GraphPad Prism 8.0. Data are expressed as mean ± s.e.m. Proliferation results from incorporation of ³H-thymidine are always shown with SI mean ± s.e.m. and, as mentioned before, SI ≥ 2 is considered significant^26,27,32.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.