Cell lines
The syngeneic mouse GBM model SB28 was provided by H. Okada, and GL261 was obtained from the Developmental Therapeutic Program, NCI. CPA was provided by the Castro-Lowenstein Laboratory, and KR158 was provided by L. Deleyrolle. The mouse bladder cancer cell line MB49 was obtained from the Animal Tumour Core at the Cleveland Clinic. The mouse melanoma B16-F10 cells were a gift from T. Stappenbeck. The GBM patient-derived xenograft cell model L1 was obtained from B. Reynolds (originally from the laboratory of A. Vescovi) and GBM23 was obtained from E. Sulman.
After thawing, all cell lines were treated with 1:100 MycoRemoval agent (MP Biomedicals) and regularly tested for Mycoplasma spp. (Lonza). GBM cell lines were maintained in complete RPMI 1640 (Media Preparation Core, Cleveland Clinic) supplemented with 10% FBS (Thermo Fisher), 1% penicillin–streptomycin (Media Preparation Core) and GlutaMAX (Gibco). MB49 and B16-F10 cells were cultured in DMEM (Media Preparation Core, Cleveland Clinic) supplemented with 10% FBS, 1% penicillin–streptomycin, GlutaMAX and sodium pyruvate (Thermo Fisher Scientific). GBM patient-derived xenograft cells were cultured in DMEM/F12 medium (ThermoFisher) supplemented with 1% penicillin–streptomycin (ThermoFisher Scientific), 1% B27 without vitamin A (ThermoFisher), 20 ng ml–1 EGF (R&D Systems) and 20 ng ml–1 FGF (R&D Systems). Cells were cultured in humidified incubators at 37 °C and 5% CO2 and were not allowed to exceed 15 passages.
Mice
All animals were kept in a specific pathogen-free facility of the Biological Resource Unit at the Lerner Research Institute, Cleveland Clinic, with a 12-h light–dark cycle. All animal procedures were performed in accordance with the guidelines and protocols approved by the Institutional Animal Care and Use Committee at the Cleveland Clinic. All mouse strains used in the study are listed in Supplementary Table 7.
Castration
Two weeks before tumour implantation, 5–6-week-old male mice underwent either castration or sham surgery. Mice were maintained under inhalation anaesthesia (2–2.5% isoflurane) through a nose cone and administered an ophthalmic lubricant to prevent corneal dryness. The scrotal area was disinfected using betadine and alcohol. A small horizontal incision was made in the skin of the scrotum and the inner skin membranes, and the testicles were exteriorized. Using resorbable vicryl sutures, testicular arteries were ligated, followed by the removal of testicles. The incision was closed using surgical clips (Fine Science Tools). For pain control, subcutaneous injections of buprenorphine (0.1 mg kg−1) and bupivacaine (5 mg kg–1) were administered. In sham-surgery mice, the same procedure was performed, excluding the ligation and removal of the testis.
Microglia depletion
Two weeks after castration, mice were fed either a control diet (AIN-76A) or a PLX3397-supplemented diet (660 mg kg–1; Research Diets) ad libitum. Mice then received intracranial tumour cell injections and remained on the assigned diet until the experimental end point.
Tumour implantation and treatments
For intracranial tumour implantation, mice were anaesthetized by inhalation anaesthesia (2–2.5% isoflurane), secured in a stereotaxis apparatus and intracranially injected with tumour cells suspended in 5 µl RPMI-null medium with the following numbers: SB28, 10,000–15,000 cells per mouse; GL261, 20,000–25,000 cells per mouse; CPA, 10,000 cells per mouse; KR158, 50,000 cells per mouse; MB49, 5,000 cells per mouse; and B16-F10, 40,000 cells per mouse. Tumour injections were targeted to the left hemisphere approximately 0.5 mm rostral and 1.8 mm lateral to the bregma with a depth of 3.5 mm from the scalp, which does not directly contact or lie near the hypothalamus. The needle was held in place an additional 60 s before slow and measured removal. The animals were monitored to detect the onset of neurological and behavioural symptoms indicative of the presence of a brain tumour.
For the GEMM, we used CRISPR-mediated somatic gene deletion using a previously described in vivo electroporation approach30. In brief, plasmids carrying Cas9 and CRISPR guide RNAs targeting Nf1, Pten and Trp53 were injected into the lateral ventricle of postnatal day 2 mouse pups, followed immediately by electroporation to enable plasmid uptake into neural stem cells. Mice were weaned at postnatal day 21, and only male mice were used in the study.
For subcutaneous tumour implantation, mice were anaesthetized by inhalation anaesthesia (2–2.5% isoflurane). A total of 500,000 SB28 cells were suspended in 100 µl RPMI-null medium and subcutaneously injected into the right flank of the mouse. Tumour size was measured starting from day 10, when the tumour becomes palpable, and measurements were taken every 2 days.
In some experiments, gonadally intact 5–6-week-old male or female mice received intraperitoneal injections of enzalutamide (10 mg kg−1; SellekChem) or vehicle (corn oil) beginning 2 days before tumour implantation. The injections were repeated every 2–3 days until the experimental end point was reached. In other experiments, intraperitoneal injections of mifepristone (25 mg kg–1; Cayman Chemical) or vehicle (corn oil) were initiated 2 days before tumour implantation and were repeated every 2–3 days until reaching the end point.
To restore testosterone levels in castrated male mice or female mice, testosterone cypionate injections (12.5 mg kg–1; Hikma Pharmaceuticals) were given subcutaneously 1 week before tumour implantation and repeated once a week.
GBM cell lysates were prepared by sonicating SB28 cells for 30 min followed by a freeze–thaw cycle. The concentration of the tumour cell suspension used for lysate preparation matched that used for tumour implantation. For the latex beads experiment, FluoSpheres polystyrene microspheres (15 μm; Invitrogen) were intracranially implanted (10,000 beads per mouse).
To inhibit cytokine signalling, mice were intraperitoneally injected with 200 µg anti-IL-1β, anti-TNF antibody or isotype antibody (BioXcell) on day −1 and day 0 of tumour implantation and injections were repeated every 2–3 days until the experimental end point.
Tumour and tissue dissociation for flow cytometry
At the indicated time points, mice were euthanized and brain tumour, spleen and lymph nodes (inguinal) were collected. Brain tumour tissue was minced into small pieces with scalpels and subjected to enzymatic digestion in the presence of collagenase D (1 mg ml–1; Roche) and DNase I (0.1 mg ml–1; Roche) at 37 °C. Digested tissue was filtered through a 70 µm cell strainer. To enrich immune cells, gradient centrifugation was performed using 30% Percoll solution (Sigma). Red blood cells (RBCs) were lysed using RBC lysis buffer (BioLegend). For spleen and lymph nodes, tissue was ground onto a 40 µm cell strainer, followed by RBC lysis. All single-cell suspension samples were filtered once more with a 40 µm cell strainer before staining for flow cytometry.
Flow cytometry
Cells were stained with the antibodies listed in Supplementary Table 8. In brief, after live/dead staining with LIVEDEAD Blue (Thermo Fisher Scientific) on ice for 15 min, cells were washed and incubated with FcR blocker (Miltenyi Biotech) diluted in PBS and 2% BSA on ice for 10 min. For surface staining, cells were incubated in an antibody mixture diluted in brilliant buffer (BD Biosciences) at 1:100 to 1:250 on ice for 30 min. After washing with PBS and 2% BSA buffer, cells were fixed with FOXP3/transcription factor fixation buffer (eBioscience) at 4 °C overnight. For intracellular staining, antibodies were diluted in FOXP3/transcription factor permeabilization buffer at a ratio of 1:250, and cells were incubated at room temperature for 45 min. For intracellular cytokine detection, cells were stimulated using Cell Stimulation Cocktail plus protein transport inhibitor (eBioscience) in complete RPMI for 4 h, followed by the cell-staining procedures described above. Stained cells were acquired using a Cytek Aurora instrument with SpectroFlow software (v.3.3.0, Cytek Biosciences) or a BD Fortessa instrument with FACSDiva software (v.9.0, BD Biosciences) and analysed using FlowJo software (v.10, BD Biosciences) following a previously published gating strategy16 (shown in Extended Data Fig. 4b). t-SNE analysis was performed using FlowJo software.
Image-localized biopsy deconvolution and analysis
A total of 202 biopsy samples collected from 58 patients (22 women, 36 men) with high-grade glioma were analysed for bulk RNA-seq25,52. Using CIBERSORTx53 and a scRNA-seq reference dataset of the glioma microenvironment with clustered cell states, including T cells54, deconvolution of bulk RNA-seq data was performed to produce estimates of T cell abundances in each sample. Owing to the limited storage available on the CIBERSORTx online interface, snRNA was downsampled 3 times to produce 100 of each cell state as input into the algorithm, each run 6 times. We present an average across runs. Statistics presented for this dataset are a result of t-test within patient sex. T cell values were averaged within patients not to violate the assumption of independent samples.
Phosphorylation array
Hypothalamic tissue samples were isolated from castrated mice and sham-surgery mice 14 days after tumour implantation. Proteins were extracted following the manufacturer’s instructions and phosphorylation events were analysed using a Phospho Explorer Antibody array (Full Moon Biosystems), with array scanning and quantification performed by the service provider. Phosphorylation signals with a fold change greater than 1.3 or less than 0.7 were considered for further analyses. Canonical pathway analysis and upstream regulator prediction were conducted using Ingenuity Pathway Analysis (Qiagen).
Immunofluorescence staining and image analysis
For immunofluorescence analysis, animals underwent high-pressure transcardiac perfusion with 4% formalin. Brains were post-fixed in the calvarium for an additional 24 h before careful dissection and sequential dehydration, first in 30% sucrose then in a 1:1 solution of 60% sucrose and optimal cutting temperature (OCT) compound. Cryoprotected brains were embedded in OCT compound, and 30 µm fixed-frozen sections were prepared using a Leica CM1950 cryostat. For staining, tissue sections were blocked overnight at 4 °C in a PBS-based blocking solution of 5% normal donkey serum, 1 mg ml–1 bovine serum albumin and 0.3% Triton X-100. Primary and secondary antibodies were bound during sequential overnight incubations at 4 °C in blocking solution. Nuclei were counterstained using Hoechst 33342 dissolved in PBS with 0.1% Triton X-100. The following primary antibodies were used at 1:1,000 dilution: chicken anti-GFP (Aves Labs), rabbit anti-cleaved caspase-3 (Cell Signaling), rabbit anti-phospho-histone H3 (Cell Signaling), rabbit anti-c-FOS (EnCor Biotechnology), goat anti-IBA1 (Abcam) and rabbit anti-ASC (Adipogen). Donkey or rabbit raised secondary antibodies were used that were conjugated to Alexa Fluor 488, 555 or 647 fluorescent dyes. To visualize cell death with cleaved caspase-3, tissue sections underwent heat-induced epitope retrieval in 10 mM citrate buffer with 0.5% Tween 20 for 45 min at 95 °C before the first blocking step. Stained tissue was digitized at ×10 magnification using an Akoya Biosciences PhenoImager HT. Digital fluorescent micrographs were analysed using the Positive Cell Detection Pipeline in QuPath55. Fluorescent images were acquired using a Leica Stellaris5 confocal microscope. IMARIS software (Oxford Instruments) was used to prepare 3D reconstructions of the images.
Tissue mRNA extraction and RT–qPCR
At the indicated time points, brain tissue was collected and flash-frozen and stored at −80 °C until processing. Total RNA was isolated using QIAzol Lysis reagent (Qiagen) following the standard protocol, including tissue homogenization. cDNA was synthesized using a High-capacity cDNA Reverse Transcription kit (Applied Biosystems). qPCR reactions were performed using TaqMan probes and Fast Advanced mater mix (Applied Biosystems) on an Applied Biosystems QuantStudio 5Real-Time PCR system. The threshold cycle (Ct) value for each gene was normalized to the expression levels of Gapdh, and relative expression was calculated by normalizing to the average delta Ct value of the control group.
In vitro tumour cell proliferation assessment
Tumour cell proliferation was monitored and quantified using an IncuCyte Live-Cell Analysis system. For these experiments, 4 technical replicates of SB28 (500 cells per well, 200 µl) and GL261 (1,000 cells per well, 200 µl) cells were plated in flat-bottom 96-well plates and treated with testosterone cypionate or vehicle (corn oil, 2 µl per well). Data were captured for up to 120 h of incubation.
ELISA
Serum was collected at the indicated time points or at the neurological end point. ACTH levels were measured using a mouse/rat ACTH ELISA kit (Abcam) following the manufacturer’s instructions. Serum was diluted at a ratio of 1:2 to 1:4. Serum testosterone levels were measured using a Testosterone ELISA kit (Cayman chemical) according to the manufacturer’s protocol. Plates were read at 600 nm or 450 nm using a Victor Nivo (Perkin Elmer) multimode plate reader.
Immunoblotting
Brain tissue samples were collected and flash-frozen using liquid nitrogen and kept in −80 °C until use. Tissue protein lysates were prepared as previously described56. In brief, tissue samples were homogenized in RIPA buffer using a motorized hand-held pestle. Lysates were then subjected to 3 freeze–thaw cycles and spun down at 13,000g for 45 min. The protein content of the isolated supernatants was quantified using a BCA kit (Pierce). A total of 20–25 μg protein lysate was loaded onto 4–20% Tris Glycine gels (Thermo Fisher) and proteins were transferred to nitrocellulose membranes. After blocking with Intercept Blocking buffer (LiCor Biosciences), membranes were incubated with primary mouse anti-caspase-1 antibody (1:1,000; Adipogen), followed by IR-680-conjugated anti-mouse secondary antibody (1:10,000; LiCor Bio). Fluorescent western blots were imaged using a LiCor Odyssey CLx system (LiCor Biosciences). Membranes were re-probed with β-actin antibody (1:10,000; Sigma-Aldrich) and β-actin bands were visualized by enhanced chemiluminescence (Thermo Scientific). Unprocessed immunoblot images are provided in Supplementary Fig. 5.
In vitro treatment of macrophages overexpressing ASC-mCerulean
Immortalized macrophages overexpressing ASC-mCerulean57 were provided by D. Abbott. Cells were cultured in DMEM supplemented with 10% charcoal-stripped FBS (Gibco) to avoid residual hormone effects and 1% penicillin–streptomycin (Media Preparation Core) and maintained in 5 µg ml–1 puromycin until use. To assess the effect of androgen signalling, cells were plated onto glass coverslips and incubated with either dihydrotestosterone (DHT; Sigma-Aldrich) or vehicle control (methanol) for 72 h. After washing, cells were stimulated with LPS (500 ng ml–1; Sigma) or GBM lysates (SB28) for 4 h. Nigericin (10 nM; Sigma) was then added during the final 45 min of stimulation. After stimulation, 2.5 µM DRAQ5 (Life Technologies) was added to the cells for 30 min. Cells were then fixed with 4% paraformaldehyde for 20 min and washed with PBS 3 times. Coverslips were mounted onto slides using Aqua-Mount (Epredia). For confocal microscopy analysis, fluorescent images were acquired using a Leica Stellaris5 confocal microscope. To assess the percentage of ASC-speck-positive cells, the number of speck-positive cells for each image was divided by the number of DRAQ5-positive cells. The mCerulean signal intensity via flow cytometry was also measured using BDSymphony. Supernatants were collected, and secreted IL-1β levels were quantified by ELISA (Abcam).
MS analysis
Freshly collected mouse serum samples were stored at −80 °C until analysis. Concentrations of glucocorticoids and testosterone were measured by LC–MS/MS as previously described58. In brief, 60 µl thawed serum was spiked with an internal standard mix (androstene-3,17-dione-2,3,4-13C3, 5α-dihydrotestosterone-16,17,17-D3 and cortisol-9,11,12,12-D4). Protein precipitation was followed by adding acetonitrile, and the supernatant was collected to extract glucocorticoids and testosterone using methyl-tert-butyl ether through a liquid–liquid extraction procedure. The steroid fraction was collected, dried and reconstituted in 140 µl of 50% methanol. The reconstituted sample underwent LC–MS/MS analysis on a Shimadzu UPLC system with a C18 column (Zorbax Eclipse Plus C18 column, 150 mm × 2.1 mm, 3.5 μm, Agilent) coupled to a QTrap 5500 mass spectrometer (AB Sciex). Data acquisition and processing were performed using MultiQuant (AB Sciex; v.0.3).
Bulk RNA-seq
For tumour cell sequencing, tumour tissue samples were collected from castrated mice and sham-surgery mice 14 days after intracranial implantation of GBM cells (SB28, 15,000 cells per mouse). For hypothalamus sequencing, hypothalamus tissue samples were collected from castrated mice and sham-surgery mice 14 days after intracranial implantation of SB28 or injection with medium. Total RNA was isolated using a Maxwell RSC simplyRNA Tissue kit, and RNA quality and quantity were measured using a Tapestation 4200 (Agilent Technologies) and a Qubit Flex Fluorometer (Invitrogen), respectively. Libraries were generated using an Illumina TruSeq Stranded mRNA kit following the manufacturer’s protocol with 300 ng RNA as input. The process involved poly-A-containing mRNA purification using poly-T oligonucleotide-attached magnetic beads, followed by fragmentation of the mRNA into small pieces using divalent cations under elevated temperature. First-strand cDNA synthesis was performed using reverse transcriptase and random primers, with actinomycin D included to improve strand specificity. Second-strand cDNA synthesis was achieved using DNA polymerase I and RNase H, replacing dTTP with dUTP in the mix to ensure strand specificity. The 3′ ends of the cDNA fragments were adenylated to prevent self-ligation, and adapters were ligated to prepare the cDNA for hybridization onto a flow cell. The cDNA libraries were enriched by PCR and purified. Libraries were quantified using a Qubit Flex Fluorometer and their size distribution was assessed using a Tapestation 4200 (Agilent Technologies). Samples were sequenced on an Illumina NovaSeq 6000 for 200 cycles according to the manufacturer’s recommendations for a depth of 20 million paired (40 million total) reads per library.
scRNA-seq
For scRNA-seq, samples were prepared 14 days after intracranial implantation of GBM cells (SB28, 15,000 cells per mouse). Single-cell suspensions were prepared from tumour and spleen samples obtained from castrated mice and sham-surgery mice as described above, and CD45+ immune cells were sorted using a BigFoot Spectral Cell Sorter (ThermoFisher). The scRNA-seq libraries were prepared from the sorted immune cells following the manufacturer’s Chromium Single Cell 3′ protocol (10x Genomics). In brief, we targeted 10,000 single cells using the 10x Genomics Chromium Controller for cDNA synthesis and barcoding, followed by evaluation of the quality and quantity of cDNA in each sample using a Bioanalyzer High Sensitivity DNA assay. The cDNA was used as the initial material for the subsequent steps, including fragmentation, end repair, adapter ligation and sample indexing. To ensure the proper construction of sample libraries, the same Bioanalyzer assay was used. Once constructed, the libraries were pooled and quantified using a Quantabio Q cycler. They were then denatured and sequenced on an Illumina Novaseq 6000 high-throughput sequencing platform. The sequencing protocol included 28 cycles for the forward read and 91 cycles for the reverse read.
Bioinformatic analysis of bulk RNA-seq and scRNA-seq data
Bulk transcriptomic FASTQ data quality was assessed using FastQC (v.0.11.8)59. Reads were aligned to the GRCm38 (mm10) mouse reference genome using STAR (v.2.7.3a)60. Expression quantification of the transcripts was done using Salmon (v.0.14.1)61 with the GRCm38 (mm10) mouse reference genome. Gene-level expression quantification was calculated using tximeta (v.1.24.0)62 in R (v.4.4.1) with the Ensembl GRCm38 release 98 GTF file. Differential expression analyses between castrated mice and sham-surgery mice were performed using DESeq2 (v.1.46.0)63. Volcano plots of DEGs were generated using EnhancedVolcano (v.1.24.0)64. Statistically overrepresented GO terms and pathways were identified using clusterProfiler (v.4.14.6)65 and the PANTHER database66.
scRNA-seq data of CD45-sorted immune cells were mapped to the mouse reference mm10 (v.1.2.0) using 10x Genomics Cell Range (v.7.2.0). Filtered feature matrices were loaded to Seurat (v.5.2.1)67. Sample batch effects were controlled during the integration process using the reciprocal PCA method with pairwise anchoring in Seurat. Integrated data were clustered using the Leiden clustering algorithm68 at a resolution of 0.6. The Leiden method was implemented using the Python module leidenalg (v.0.10.2) in Python (v.3.12.8), which was loaded with reticulate (v.1.40.0) into R. After clustering, immune-cell-type classification was done using ScType (v.1.0)69 with the ‘Immune system’ tissue selection. For each cell type, differential expression analysis after pseudobulking was performed in Seurat with the DESeq2 method. Immune cell–cell communications were estimated using CellChat (v.2.1.2)70. Subsequently, T cell populations in the brain and macrophage populations in the spleen were selected, and we performed subclustering to identify finer cell population compositions. GSEA was performed using clusterProfiler (v.4.14.6) with genome-wide annotation for mouse org.Mm.eg.db (v.3.20.0) and the Molecular Signatures Database71 R implementation msigdbr (v.10.0.1)72.
Spatial transcriptomics
For digital spatial profiling, mice were subjected to cardiac perfusion with a solution of 4% formaldehyde in PBS. The brain was post-fixed in the calvarium for an additional 16–24 h before the intact brain was isolated into a solution of 70% ethanol. Preserved brain specimens were dehydrated with ascending concentrations of ethanol, cleared with xylene and embedded in paraffin. Formalin-fixed paraffin-embedded mouse brain samples were sectioned onto four slides for GeoMx digital spatial profiling slide processing. Each slide contained sections from three biological replicates for each time point per condition. All slides were processed for GeoMx Mouse Whole Transcriptome Atlas (MuWTA) collections according to Nanostring protocols. Slides were baked for 18 h at 60 °C before deparaffinization to increase tissue adhesion. After deparaffinization, slides underwent target retrieval in 1× Tris-EDTA at 99 °C for 20 min, followed by protease K digestion at a concentration of 1 μg ml–1 for 15 min and post-fixation with 10% NBF. MuWTA probes were added and hybridized overnight at 37 °C. The next day, slide preparation continued with stringent washes, blocking with Nanosting Buffer W and then staining with pre-conjugated antibodies. For this assay, the slides were stained with Syto13 (1:10; Nanostring), CD45 (1:40; Nanostring, AF594 conjugate) and IBA1 (1:100; Cell Signaling Technologies, AF647 conjugate). The slides were scanned on a GeoMx instrument (v.3.1.2.12), and ROIs were placed on each slide in cortical, hypothalamus and intratumoral regions in each replicate. Each ROI was divided into IBA1+ and IBA1– cell segments by thresholding on the basis of IBA1 antibody fluorescence. For each unique segment, the barcoded MuWTA probes were collected into a 96-well plate. The collected segment probes were amplified by PCR, pooled based on their corresponding segment area and sequenced on an Illumina NovaSeq X Plus platform. Fastq files were converted into digital count conversion files using the GeoMx NGS pipeline (v.3.1.3.6) and uploaded to the GeoMx Analysis suite (v.3.1.2.12) for data analysis.
Raw count data were normalized using the DESeq2 variance-stabilized transformation function. Negative control probes were used to establish a variance threshold, which was defined as the mean plus two standard deviations of the negative probe variance. Only genes exceeding this threshold were retained for downstream analyses. Differential gene expression analysis was performed using the limma package. Linear models were fitted to normalized expression data using the lmFit function followed by empirical Bayes moderation with eBayes. The design matrices incorporated treatment effects while controlling relevant covariates including spatial regions and time point (Supplementary Table 6). GSEA was performed using the fgsea package with pre-ranked gene lists from the MsigDBR package, including Hallmark, KEGG, Reactome, WIKIPATHWAYS and a customized set of ontologies found from the literature. Results were visualized using GGplot2. Cell signatures were taken from previous publications38,40 and used to score GeoMx samples for each signature and visualized using Complexheatmap. No multiple testing correction was applied to individual gene P values in the limma analysis, but pathway-level statistics were adjusted for multiple testing using the Benjamini–Hochberg method. R (v.4.4.1) was used for all analyses.
Real-world data analysis
Real-world data were obtained from linked SEER data and Medicare claims for male patients with GBM diagnosed from 1 January 2008 to 31 December 2019 with corresponding Medicare claims and follow up for survival until 31 December 2021. The standard-of-care therapy for GBM is maximum surgical resection and concomitant chemoradiation73,74. Therefore, the analysis was limited to individuals who received surgical resection (categorized as biopsy only, subtotal resection, gross total resection and surgery not otherwise specified), radiation and temozolomide, which led to 1,333 cases for analysis. Survival was assessed in individuals receiving temozolomide only compared with those who received both temozolomide and supplementary testosterone using Cox proportional hazard models adjusted for demographics, extent of resection and Charlson comorbidity scores to estimate HR, 95% CI and P values. Significance was assessed at an alpha level 0.05 for the P values. Additional details are provided in the Supplementary Information.
Statistical analysis
GraphPad Prism (v.9, GraphPad Software) software was used for data presentation and statistical analyses. Unpaired two-sided t-test or one-way or two-way ANOVA was used with Tukey’s multiple comparisons test, as indicated in the figure legends. Survival analysis was performed using log-rank tests. P < 0.05 was considered significant.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
