Clinical samples (tumour tissue and blood) from patients with histologically confirmed glioblastoma multiforme were obtained through the South Australian Neurological Tumour Bank (SANTB) and use of the tissues was approved by the Central Adelaide Local Health Network Human Research Ethics Committee (CALHN HREC; approval number R20160727). All human specimens were used in accordance with the Declaration of Helsinki, and participants provided written consent.

To interrogate genes related to glioblastoma T-cell homing, we used our previously published scRNA-seq dataset based on a total of 13,903 cells from three primary glioblastoma specimens (22). Refer to Ebert et al. (2020) for detailed methods on tumour specimen processing for single-cell sequencing, as well as single-cell library processing (22). Briefly, using R version 3.6.3 and Seurat version 3.1.5, cells with (i) fewer than 200 genes, (ii) gene numbers outside ±2 standard deviations from the mean, or (iii) a mitochondrial gene fraction greater than 10% were excluded. Remaining cells were log normalised by total expression and scaled to 10,000 transcripts/cell with the NormalizeData function in Seurat. Using the FindIntegrationAnchors (dims = 1:30, k.filter = 200) and IntegrateData (dims = 1:30) functions, cells from different libraries were combined by assessing the pairwise correspondence between a set of representative genes (anchors). The FindVariableGenes function was used to identify variable genes returning 2000 features using vst as the selection method. The data were then scaled and principal component analysis applied using the ScaleData and RunPCA functions respectively. Cells were clustered based on gene expression profiles using the FindNeighbors and FindClusters functions with resolution set to 0.5. UMAP plots were generated using the first 20 principal components, and clusters annotated based on expression of reported cell-lineage markers (23).

To identify different subsets of lymphocytes, the lymphocyte clusters were first isolated from non-lymphocyte clusters. Subsequently, the isolated lymphocyte cells were re-clustered as described above. UMAP plots were generated using the first 11 principal components, and clusters annotated based on expression of cell-lineage markers. EBF1 and CD19 delineated B cells and CD14 and CD68 identified contaminating macrophages. CD3D and NCR1 identified T-cell and NK-cell-clusters respectively. CD4⁺, CD8⁺ T cells and Tregs were visualised using CD4, CD8 and FOXP3 respectively. T cells expressing high levels of TCF7, SELL, CCR7, IL7R, PTGER2, and KLRG1 were considered Tcm-like. GZMB, PDCD1, HLA-DRA, and KLRC1 were used to classify effectors or recently activated T cells. Additionally, ITGA1, ITGAE, and CXCR6 were used to identify Trm cells (13).

To screen for chemokine receptor, integrin, chemokine, and adhesion molecule gene expression, the frequency of cells positive for genes of interest were obtained using the FetchData function from the merged and normalised libraries of the BT20, BT23 and BT26 scRNA-seq dataset. Cells were considered positive for a gene of interest if at least one unique molecular identifier (UMI) corresponding to the gene was detected in the cell. Cell populations were identified as previously described (22).

Glioblastoma samples were obtained as pieces of resected tumour tissue, or as tumour fragments recovered from aspirates following Cavitron Ultrasonic Surgical Aspirator (CUSA) ablation, and dissociated as described previously (22). For tumour biopsies, a portion of the sample was embedded in optimal cutting temperature (OCT) medium, flash-frozen using supercooled isopentane, and stored at -80°C. Remaining tumour biopsy was rinsed to remove blood and dissociated to generate single-cell suspensions using Accutase (Thermo Fisher Scientific) or the Human Tumor Dissociation kit in combination with the gentleMACS Octo Dissociator (Miltenyi Biotec) according to the manufacturer’s recommendations. Tissue fragments in CUSA aspirates were first sedimented by slow centrifugation at 70g for 1 min and the supernatant was discarded. The remaining tissue fragments were dissociated using Accutase or the Human Tumor Dissociation kit as described above. Following digestion, dissociated samples were centrifuged at 300g for 5 min to pellet cells, and supernatants for chemokine analyses harvested and stored at -80°C. Dissociated cells were resuspended and filtered through a 70µm strainer. For cryopreservation, a portion of the cells were resuspended in 90% foetal bovine serum (FBS) + 10% dimethyl sulphoxide (DMSO), frozen in a Mr Frosty freezing chamber (Thermo Fisher Scientific) at -80°C overnight, then stored in liquid nitrogen for flow cytometry analyses. The remaining cells were used to generate glioma neural stem (GNS) cell cultures as previously described (22, 24). GNS cell-line supernatants were harvested when the cells were 75-95% confluent and stored at -80°C for chemokine analyses.

Peripheral blood mononuclear cells (PBMC) were harvested from patient blood via density centrifugation (Lymphoprep), resuspended in 90% FBS + 10% DMSO, then frozen in a Mr Frosty freezing chamber (Thermo Fisher Scientific) at -80°C overnight and stored in liquid nitrogen for flow cytometry analyses.

Sections cut from OCT-embedded fresh tissues (5-12µm thickness) were fixed for 20 minutes in Cytofix/Cytoperm (BD) at room temperature (RT), washed in PBS and blocked using 10% human plasma, 10% normal donkey serum and 10% normal goat serum in PBS containing 1% bovine serum albumin (BSA) for 30 minutes at RT. 5µg/mL purified mouse anti-CD3 (UCHT1; BioLegend) and 0.67µg/mL rabbit anti-CD31 (polyclonal; Bethyl Labs #IHC-00055) antibodies were incubated with tissue sections for 12 hours at 4°C. Following washing, tissue sections were incubated with goat anti-mouse IgG-AF488 (Abcam) and donkey anti-rabbit IgG-AF555 (Abcam) detection antibodies at a dilution of 1:500 for 1 hour at RT. Following washing, tissue sections were blocked using 10% normal mouse serum in PBS containing 1% BSA for 30 minutes at RT. Sections were then incubated with mouse anti-GD2-AF647 (14.G2A; BD) at a dilution of 1:100 for 2 hours at RT. After washing, sections were mounted using ProLong Gold anti-fade mounting medium with DAPI (Life Technologies), cured overnight, and sealed with nail varnish before viewing.

Confocal imaging was performed on a Zeiss LSM 800 confocal microscope equipped with 405nm, 488nm, 561nm and 640nm lasers, using 20x objective and Zen 2.6 system software. Fluorescence overlays were created by combining Z-stacks, merging channels, and applying false colour using ImageJ (National Institutes of Health).

For flow cytometry analysis, cryopreserved, dissociated tumour specimens were rapidly thawed at 37°C, washed, and incubated in pre-warmed Roswell Park Memorial Institute (RPMI) media and 20U/mL DNase I (Sigma) for 15 minutes at 37°C. Cells were then washed and resuspended in a 30% isotonic Percoll (Sigma)/RPMI solution with 2mM EDTA. Using a syringe and needle, 70% isotonic Percoll/RPMI with 2mM EDTA was underlaid beneath the cell suspension, then centrifuged at 500g for 30 mins without the brake at RT. Lymphocytes were harvested from the interphase and washed with complete RPMI media to remove excess Percoll. Isolated lymphocytes were resuspended in PBS and plated at a density of 3x10⁵ cells/well in 96-well V-bottom plates. Cryopreserved patient PBMC were rapidly thawed at 37°C, washed, and resuspended in PBS to be plated at a density of 5x10⁵ cells/well. Five flow cytometry panels were designed to simultaneously analyse chemokine receptor and/or integrin expression while identifying T-cell subsets. A core combination of antibodies to identify T-cell subsets including: anti-CD3-PerCP-Cy5.5 (UCHT1; BD), anti-CD4-BV510 (SK3; BD), anti-CD8-BV786 (RPA-T8; BD), anti-CD25-BB515 (2A3; BD), anti-CD69-PE-Cy7 (FN50; BD), anti-CD45RO-BUV395 (UCHL1; BD), anti-CCR7-APC (2-L1-A; BD) or anti-CCR7-BV421 (G043H7; BD), anti-CD127-PE (A019D5; Biolegend), and anti-CD103-BV605 (Ber-ACT8; Biolegend) were used in each panel with an additional 2-3 antibodies to detect different chemokine receptors and/or integrins. Patient samples were randomly allocated to the different flow panels with larger biopsies that provided a greater yield of T cells used in multiple flow panels. Patient tumour samples digested with Accutase were allocated to the flow cytometry panel containing CXCR3 and CCR6 antibodies as the standard dissociation protocol with GentleMACS reagents (Miltenyi) significantly reduced the detection of these receptors. Cells were pelleted via centrifugation at 400g for 3 minutes, resuspended in PBS with 1:1000 Fixable viability stain 780 (BD) and incubated for 20 minutes at RT. Cells were washed with FACS buffer (PBS, 1% BSA) and Fc receptors blocked with human Fc block (BD) for 5 minutes on ice. For chemokine receptor analyses, the following antibodies were used: anti-CCR2-BUV737 (LS132.1D9, BD), anti-CCR4-BV421 (1G1; BD), anti-CCR5-PE-CF594 (2D7; BD), anti-CCR6-BUV737 (11A9; BD), anti-CCR7-APC, anti-CCR7-BV421, anti-CXCR3-APC (G025H7; Biolegend), anti-CXCR4-BV421 (12G5; BD), anti-CXCR6-BV421 (13B 1E5; BD), and anti-CX3CR1-PE-CF594 (2A9-1; BD). For panels with anti-CCR5, chemokine receptor staining was staggered as previously recommended (25). First, cells were incubated with chemokine receptor antibodies, except anti-CCR5, in Brilliant Staining Buffer Plus (BSB+; BD) for 10 mins at RT. Anti-CCR5-PE-CF594 was then added and cells were incubated for a further 10 minutes at RT. For panels without anti-CCR5, cells were incubated with chemokine receptor antibodies in BSB+ for 20 mins at RT. To the above suspensions, our core T-cell antibody cocktail diluted in BSB+ (BD) was added and incubated for a further 20 minutes on ice. For integrin panels, cells were incubated with anti-CCR7 at RT for 20 minutes, then incubated for a further 20-minutes on ice with the above T-cell cocktail supplemented with anti-CD11a-BV421 (HI111; BD) and anti-CD49d-BUV737 (9F10; BD), or anti-CD51-APC (NKI-M9; Biolegend) and anti-CD49a-BUV737 (SR84; BD). Stained cells were washed with PBS and fixed with Cytofix/Cytoperm (BD) for 20 minutes on ice. Lastly, cells were washed with PBS, resuspended in FACS buffer, and stored at 4°C protected from light until acquisition. Samples were acquired on a BD LSR Fortessa Special Order Research Product using FACS Diva Software version 8.0. Analysis was performed using FlowJo™ version 10.8.1 (BD).

To detect CXCL16 and CCL20 in supernatants harvested from GNS cell-lines and dissociated glioblastoma and CUSA samples, the DuoSet ELISA Human CXCL16 kit (DY1164; R&D Systems) and DuoSet ELISA Human CCL20 kit (DY36005; R&D Systems) were used according to the manufacturer’s recommendations. Samples were analysed at 450 nm (FLUOstar Omega microplate reader), and concentrations of chemokines calculated from standard curves generated with GraphPad Prism version 8.4.3.

To detect the chemokines CCL2, CCL3, CCL4, CCL5, CXCL9, CXCL10, CXCL11, CXCL12 and CX3CL1 in supernatants harvested from GNS cell-lines and dissociated glioblastoma and CUSA samples, a LEGENDplex Custom Human Chemokine Panel kit (Biolegend) was used according to the manufacturer’s recommendations. Standards and samples were acquired on a BD LSR Fortessa Special Order Research Product and chemokine concentrations determined using the LEGENDplex data analysis software version 2021-07-01 (Biolegend).

Statistical analysis and graphing were performed in GraphPad Prism version 8.4.3. Pairwise comparisons were performed using two-tailed paired Student’s t test. For all tests, p values under 0.05 were considered significant. Significance was represented as: *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001.

To identify glioblastoma tissues with infiltrating T cells, frozen sections were prepared from patients’ biopsies (n=6) for confocal microscopy. Sections were stained for CD3 (green) to identify glioblastoma-infiltrating T cells, GD2 (magenta) to visualise tumour cells, and CD31 (white) to identify ECs ( Figure 1A ). In all instances, T cells were visualised near CD31⁺ blood ECs and among GD2⁺ glioblastoma cells ( Figure 1A ), but not within blood vessels. Importantly, this confirmed that glioblastoma tissues contained endogenous T cells that had infiltrated the glioblastoma parenchyma and were not restricted to the blood vessel lumen.

Three of these glioblastoma biopsies, all of which contained tumour-infiltrating T cells, (BT20, BT23, BT26; refer to Table 1 ), were matched to our previously published scRNA-seq dataset (22). This dataset was generated by dissociating three primary glioblastoma patient specimens ( Table 1 ) to single-cell suspensions and subjecting them to scRNA-seq using the 10X Chromium platform. Detailed methods and data analysis pipelines are described in the materials and methods section and our recent publication (22). Upon clustering of cells based on their transcriptome, expression of reported lineage-markers was used to identify the cell-type of each cluster. Subsequently, clusters were identified as ECs, glioblastoma-associated macrophages/microglia (GAMs), pericytes, and lymphocytes ( Supplementary Figure 1A ). Clustered cells expressing a range of genes associated with glial and neural cell types but lacking stromal lineage-markers were annotated as tumour cells ( Supplementary Figure 1A ). Utilising this existing dataset, expression of chemokine receptors and integrins was interrogated in glioblastoma-infiltrating T cells for this study.

To screen for chemokine receptor and integrin genes expressed by glioblastoma-infiltrating T cells, the lymphocyte cluster was isolated and re-clustered with the bioinformatics pipeline described (see above and Materials and Methods) to further separate lymphocyte populations according to differentially expressed genes ( Figure 1B ). B cells were identified via EBF1 and CD19 expression, contaminating macrophages via CD14 and CD68 expression, T cells via CD3D expression and NK cells via NCR1 expression ( Figure 1B and Supplementary Figures 1B, C ). Frequencies of T cells positive for chemokine receptor ( Figure 1C ) and integrin genes ( Figure 1D ) in each patient were determined. Within glioblastoma-infiltrating T cells, a large proportion of cells were positive for CXCR4 and the proportion of cells positive for chemokine receptors CCR4, CCR5, CCR6, CCR7, CXCR3, and CXCR6 ranged between 5-25% ( Figure 1C ). Additionally, integrin genes ITGAE, ITGAL, ITGAV, ITGA1, ITGA4, ITGA5, ITGA6, ITGB1, ITGB2, and ITGB7 were detected in a moderate proportion of T cells ( Figure 1D ). All other receptors investigated were either undetectable or were expressed in low frequencies of T cells. Remaining cells in the lymphocyte cluster were also screened for chemokine receptor and integrin gene expression. B cells, NK cells and contaminating macrophages also expressed chemokine receptor and integrin genes with some overlap in the expression of receptors with T cells ( Supplementary Figures 2A, B ). Together, these data identify infiltrating T cells in glioblastoma biopsies and define a putative expression profile of glioblastoma-homing markers.

Chemokine receptor-mediated T-cell recruitment is directed by complementary chemokine ligands expressed in the glioblastoma microenvironment; thus, the expression of chemokines was screened in non-lymphocyte clusters ( Figure 1E ). There were varying proportions of cells positive for chemokines CCL2, CCL3, CCL4, CCL5, CCL20, CXCL1, CXCL2, CXCL3, CXCL10, CXCL11, CXCL12, CXCL16, and CX3CL1 ( Figures 1F, G ), among non-lymphocyte cell populations. Subsequently, chemokine receptors identified in glioblastoma-infiltrating T cells were selected for further investigation based on expression of the receptor at the gene level within T-cell clusters, and/or expression of the cognate ligand(s) identified within non-lymphocyte clusters. Thus, the chemokine receptors CCR2, CCR4, CCR5, CCR6, CXCR3, CXCR4, CXCR6, and CX3CR1 were selected for further validation by flow cytometry. Similarly, integrins CD49a (ITGA1), CD49d (ITGA4), CD51 (ITGAV), and CD11a (ITGAL) were selected for further validation based on elevated gene expression in glioblastoma-infiltrating T cells.

Previous studies have identified that memory T cells, including Tcm, Tem and Temra, and CD4⁺ Tregs can infiltrate glioblastoma (10, 13, 14, 20). Additionally, Trm cells have been characterised in the healthy brain (18). In keeping with this, our scRNA-seq data revealed heterogenous expression of genes associated with a Tcm-like signature (TCF7, SELL, CCR7, IL7R, PTGER2, and KLRG1), and effector-like genes (GZMB, PDCD1, HLA-DRA, and KLRC1) ( Supplementary Figure 1C ) (13). Additionally, our scRNA-seq dataset revealed FOXP3 expression in a small proportion of glioblastoma Tregs, and expression of ITGA1, ITGAE, and CXCR6 which are associated with a Trm signature ( Supplementary Figure 1C ) (13). Considering these subsets play distinct roles in tumour immunity (10, 17, 26), a multicolour flow cytometry panel was designed to simultaneously identify these populations and interrogate chemokine receptor and integrin expression ( Figures 2A, B ). Paired blood ( Figure 2A ) and dissociated glioblastoma samples ( Figure 2B ) were analysed to identify homing receptors enriched on glioblastoma-infiltrating T-cell subsets compared to paired blood samples as a means of identifying migratory axes that may be of biological importance. Tregs were identified as CD4⁺CD25⁺CD127^low (27), and CD4⁺ or CD8⁺ Trm cells as CD69⁺CD103⁺ (18). Trm cells were not detected in the peripheral blood samples and only identified within glioblastoma. Within the CD103^- population, CCR7 and CD45RO expression patterns distinguished naïve (CCR7⁺CD45RO^-), Tcm (CCR7⁺CD45RO⁺), Tem (CCR7^-CD45RO⁺), and Temra (CCR7^-CD45RO^-) populations for both CD4⁺ and CD8⁺ T cells (10, 16). Following identification of T-cell subsets, frequencies of these subsets were compared between matched patient blood and glioblastoma biopsies ( Figure 2C ). Of note, significantly fewer naïve CD4⁺ and CD8⁺ T cells were detected in glioblastoma biopsies compared to paired blood samples, indicating limited blood contamination of these resected glioblastoma biopsies obtained at surgery. In contrast, there was a significant enrichment of CD4⁺ and CD8⁺ Tcm, Tem, and Trm populations in glioblastoma biopsies, with CD4⁺ Tcm and CD8⁺ Tem and Trm populations most abundant overall. There was also a significant decrease in the proportion of CD8⁺ Temra cells in glioblastoma compared to blood samples, while the proportions of CD4⁺ Temra cells in blood and glioblastoma were negligible. Proportions of Tregs between the blood and tumour samples were similar.

Following identification of T-cell subsets ( Figure 2 ), expression of CCR2, CCR4, CCR5, CCR6, CXCR3, CXCR4, CXCR6, and CX3CR1 was investigated on glioblastoma and blood CD4⁺ ( Figure 3 ) and CD8⁺ ( Figure 4 ) T-cell subsets. Here, naïve blood T cells served as an internal negative control for chemokine receptor staining as this population lacks the expression of inflammatory chemokine receptors, except for CXCR3 and CXCR4, where naïve T cells and live CD3^- cells containing distinct negative and positive populations were used to set gates for these receptors respectively ( Supplementary Figure 3 ).

Our standard process for dissociation of tumour tissue specimens using gentleMACS reagents was found to significantly diminish CXCR3 and CCR6 staining (data not shown). Hence, other patient samples digested with Accutase were used to interrogate these receptors. Representative histograms for CD4⁺ T-cell subsets illustrate the typical chemokine receptor expression patterns observed, and demonstrate that staining for CCR2, CCR5, CXCR3 and CXCR6 was robust in glioblastoma-infiltrating T-cell populations compared to matched blood naïve CD4⁺ T cells ( Figure 3A and Supplementary Figure 4 ). Quantitative analysis of chemokine receptor expression revealed a significant increase in the proportion of CCR2⁺CD4⁺ Tcm and Treg cells in glioblastoma compared to blood samples from the same patients ( Figure 3B ). CCR4 expression was most abundant in Tregs, however there was no difference in the proportion of CCR4⁺ Tregs in glioblastoma compared to peripheral blood ( Figure 3C ). Strikingly, all glioblastoma CD4⁺ T-cell subsets highly expressed CCR5, and CCR5 expression was significantly enriched in glioblastoma T cells compared to blood ( Figure 3D ). CCR6 expression was relatively low in CD4⁺ T-cell subsets compared to other chemokine receptors analysed, and no significant differences in CCR6 expression were observed between T-cell subsets in glioblastoma and blood ( Figure 3E ). CXCR3⁺ cells were abundant in all CD4⁺ T-cell subsets, however enrichment was only statistically significant in CD4⁺ Tem and Tregs in glioblastoma biopsies compared to paired blood samples ( Figure 3F ). Equally high proportions of CD4⁺ T-cell subsets in both blood and glioblastoma expressed CXCR4, hence no enrichment of this receptor was observed in glioblastoma ( Figure 3G ). As high proportions of both blood and glioblastoma-associated CD4⁺ T-cell subsets expressed CXCR4, the geometric mean fluorescence intensity (gMFI) was compared to determine whether CXCR4 abundance on T-cell subsets differed between glioblastoma and blood samples. Strikingly, there was a significant increase in the abundance of CXCR4 on all CD4⁺ T-cell subsets in glioblastoma compared to blood samples ( Supplementary Figure 5A ). CXCR6, an established marker of CD69⁺CD103⁺ Trm cells, was expressed on CD4⁺ Trm cells as anticipated, but also enriched in all other CD4⁺ T-cell subsets in glioblastoma compared to blood ( Figure 3H ). Lastly, CX3CR1 expression was low among all CD4⁺ T-cell subsets except CD4⁺ Temra cells, where CX3CR1 expression was enriched in the blood and significantly lower in glioblastoma ( Figure 3I ). Together, our flow cytometry analysis demonstrates that CCR5 and CXCR6 are expressed by large proportions of all glioblastoma-infiltrating CD4⁺ T-cell subsets, whereas CCR2 and CXCR3 enrichment was only observed in glioblastoma-infiltrating CD4⁺ Tcm and Tregs, and Tem and Tregs respectively.

Within CD8⁺ T-cell subsets, representative histograms highlighted robust expression of CCR5, CXCR3, and CXCR6 compared to naïve peripheral blood CD8⁺ T cells ( Figure 4A and Supplementary Figure 6 ). Similar to CD4⁺ T cells, there was enrichment of CCR2⁺CD8⁺ Tcm cells in glioblastoma compared to paired blood samples, and approximately 50% of CD8⁺ glioblastoma Trms expressed CCR2 ( Figure 4B ). CCR4⁺ cells were most abundant within the CD8⁺ Tcm population, but expression of this receptor in glioblastoma samples was not enriched for any CD8⁺ T-cell subsets compared to matched blood samples ( Figure 4C ). Like CD4⁺ T-cell subsets, CCR5 was expressed by a high proportion of all CD8⁺ T-cell subsets and was significantly enriched for all CD8⁺ T-cell subsets in glioblastoma ( Figure 4D ). CCR6 was expressed at similar proportions between blood and glioblastoma specimens for all CD8⁺ T-cell subsets ( Figure 4E ). Like CD4⁺ glioblastoma T cells, CXCR3 was expressed moderately by all CD8⁺ T-cell subsets, and there was a significantly greater proportion of CXCR3⁺CD8⁺ Tem cells in glioblastoma biopsies compared to paired blood samples ( Figure 4F ). CXCR4 expression was uniformly high for all CD8⁺ T-cell subsets and significantly enriched in glioblastoma compared to peripheral blood ( Figure 4G ). Additionally, CXCR4 gMFI was significantly increased for all CD8⁺ T-cell subsets in glioblastoma ( Supplementary Figure 5B ). Although all CD8⁺ glioblastoma T-cell subsets expressed modest levels of CXCR6, expression was only enriched within glioblastoma compared to blood for Tcm and Temra cells ( Figure 4H ). Lastly, CX3CR1 expression was greatest in the blood for CD8⁺ Tem and Temra populations with a significant reduction in the frequency of CX3CR1⁺CD8⁺ Tem and Temra cells in glioblastoma samples. Overall, CCR5 and CXCR4 show the most prominent expression on CD8⁺ T-cell subsets in glioblastoma tissues, with CCR2, CXCR3 and CXCR6 displaying subset-specific glioblastoma enrichment in Tcm, Tem, and Tcm and Temra populations respectively.

Utilising the same core panel for identification of T-cell subsets as Figures 3 , 4 , integrins identified by scRNA-seq were investigated by flow cytometry. Subsequently, proportions of T cells expressing integrins were compared between paired blood and glioblastoma samples ( Figure 5 ). All T-cell subsets in blood and glioblastoma biopsies uniformly expressed high levels of CD11a (integrin αL; Figures 5A, B and Supplementary Figures 4 , 6 ). Positive gates for CD49a and CD51 expression on glioblastoma T-cell subsets were set based on paired blood naïve T cells, which were uniformly CD49a^- and CD51^- ( Figures 5A, B and Supplementary Figure 3 ). Live CD3^- cells containing distinct populations of CD49d⁺ and CD49d^- cells were used to set positive gates for CD49d⁺ T cells ( Supplementary Figure 3 ). Despite high proportions of T cells positive for ITGAV, we did not detect CD51 protein in any T-cell subsets in glioblastoma biopsies or paired blood samples ( Figures 5A, B ). Importantly, we demonstrated efficacy of this CD51 antibody clone with the Malme-3M human melanoma cell-line (data not shown). In contrast, CD49a (integrin α1) expression was significantly enriched within CD4⁺ Tcm, Tem, Temra and Tregs, and CD8⁺ Tcm, Tem and Temra subsets in glioblastoma infiltrates compared to blood, and both CD4⁺ and CD8⁺ Trm cells expressed this integrin ( Figures 5C, D ). Additionally, CD4⁺ Tcm, Tem and Tregs expressing CD49d (integrin α4) were enriched in glioblastoma ( Figure 5D ). High proportions of CD8⁺ T-cell subsets expressed CD49d, and enrichment for CD49d expression in glioblastoma was significant for the CD8⁺ Tcm population ( Figure 5D ). Given large proportions of T cells in blood and glioblastoma expressed CD11a and CD49d, the abundance of these integrins on T cells, as measured by gMFI, was compared between blood and glioblastoma samples ( Supplementary Figure 5 ). We observed no differences in the abundance of CD11a for CD4⁺ T-cell subsets, while there was an increase in the abundance of CD11a on the CD8⁺ Temra population ( Supplementary Figures 5A, B ). There was a significant decrease in the abundance of CD49d on the CD4⁺ Tem population in glioblastoma compared to blood samples, and a significant increase in CD49d gMFI on CD4⁺ Tregs in glioblastoma ( Supplementary Figure 5A ). However, there was no difference in the abundance of CD49d on CD8⁺ T-cell subsets between glioblastoma and paired blood samples ( Supplementary Figure 5B ). Physiological T-cell surveillance of the brain includes migration across post-capillary blood-vessels into perivascular spaces adjacent to the brain parenchyma (12). This pathway requires adherence of T cells, which occurs via the integrins CD11a and CD49d interacting with adhesion molecules ICAM-1 and VCAM-1, respectively, on BBB ECs. Indeed, we identified expression of ICAM-1 and VCAM-1 at the gene level within glioblastoma-associated clusters via scRNA-seq ( Supplementary Figure 7A ). Together, these data suggest that CD11a-ICAM-1, CD49a and CD49d-VCAM-1 contribute to the infiltration of subsets of T cells into glioblastoma.

Having identified that CCR2⁺, CCR5⁺, CXCR3⁺, CXCR4⁺, and CXCR6⁺ T-cell subsets were enriched in glioblastoma, we subsequently validated the expression of complementary chemokines in dissociated glioblastoma biopsies and patient-derived glioblastoma cell-lines ( Figure 6 ). We also tested for the complementary chemokines to CCR6 and CX3CR1 because these receptors have been implicated in brain-homing in other studies (28, 29). The CCR4 chemokines were not selected for further analysis as we observed low proportions of non-lymphocyte cell populations expressing the complementary chemokines CCL17 and CCL22 via scRNA-seq ( Figure 1G ) and observed no enrichment of CCR4⁺ T cells in glioblastoma ( Figures 3C, 4C ). To measure chemokines present within tumour tissue, the supernatant fraction from dissociated glioblastoma biopsies and CUSA aspirates, expected to contain soluble factors released from extracellular tissue spaces, was retained for analysis. In addition, we also collected supernatants from short-term cultures of tumour cells cultured under defined GNS cell conditions (24) to measure active secretion of chemokines by tumour cells. Chemokines were detected in both sample types via ELISA or cytometric bead-based assay. Notably, this approach favours the detection of soluble chemokines, as chemokines sequestered intracellularly or tethered to the cell surface via transmembrane domains or glycosaminoglycans (GAGs) are likely to remain in the cellular fraction. Of all chemokines assayed, the CCR2/CCR4 ligand CCL2 was the most abundant chemokine detected in tumour biopsy, CUSA and cultured GNS cell supernatants, and was present above the limit of detection in all samples analysed ( Figure 6A ). Interestingly, CCR5 ligands: CCL3, CCL4 and CCL5 were detected in the majority of supernatants harvested from dissociated tumour biopsy and CUSA, but little was detected in GNS cell supernatants ( Figures 6B–D ). Of the CXCR3 ligands, CXCL9 was detected in the majority of dissociated tumour specimens but was below the level of detection in all GNS cell cultures analysed ( Figure 6E ), CXCL10 was largely absent from dissociated tumour specimens, but highly expressed by few GNS cell samples ( Figure 6F ), and CXCL11 was below the lower limit of detection in all samples ( Figure 6G ). This was in contrast to expression of these ligands detected via scRNA seq ( Figure 1F ), but may reflect the high affinity of CXCL10 and CXCL11 for GAGs (30, 31). CXCL12, the sole chemokine ligand for CXCR4, was abundant in 6 out of 12 supernatants from the GNS cell cultures, however it was undetectable in all dissociated tumour biopsy and CUSA samples analysed ( Figure 6H ). CXCL16, the chemokine ligand of CXCR6, was detected in supernatants from the majority of dissociated tumour and GNS cell samples ( Figure 6I ). CX3CL1 was below the lower limit of detection in supernatants from all dissociated glioblastoma tumour biopsy, CUSA and GNS cell samples analysed ( Figure 6J ). However, measures here for both CXCL16 and CX3CL1 may underestimate their abundance as they are expressed as transmembrane chemokines and CXCL16 exhibits selective GAG binding (32–34). CCL20, the chemokine ligand of CCR6, was not detected in either dissociated tumour or GNS cell supernatants ( Figure 6K ). Comparing chemokine abundance between dissociated tumour and cultured GNS cell supernatants, CCL3, CCL4, CCL5 and CXCL9 were abundant in dissociated tumour samples, with little detected in supernatants from cultured tumour cells, suggesting non-tumour-cell populations were the source of these chemokines in glioblastoma. In contrast, glioblastoma tumour cells may be the main source of CCL2, CXCL10 and CXCL12 production in the tumour microenvironment given the elevated CCL2 detected in GNS cell culture supernatants and the minimal levels of CXCL10 and CXCL12 detected in dissociated tumour supernatants. Considered together with the analyses of receptor expression on glioblastoma T cells, these data implicate the CCR2-CCL2, CCR5-CCL3/4/5, CXCR3-CXCL9, CXCR4-CXCL12 and CXCR6-CXCL16 chemokine axes in T-cell infiltration and/or retention in glioblastoma.