B-cells Drive Response to PD-1 Blockade in Glioblastoma Upon Neutralization of TGFβ-mediated Immunosuppression

Immunotherapy has revolutionized cancer treatment but has yet to be translated into brain tumors. Studies in other solid tumors suggest a central role of B-cell immunity in driving immune-checkpoint-blockade efficacy. Using single-cell and single-nuclei transcriptomics of human glioblastoma and melanoma brain metastasis, we found that tumor-associated B-cells have high expression of checkpoint molecules, known to block B-cell-receptor downstream effector function such as plasmablast differentiation and antigen-presentation. We also identified TGFβ-1/TGFβ receptor-2 interaction as a crucial modulator of B-cell suppression. Treatment of glioblastoma patients with pembrolizumab induced expression of B-cell checkpoint molecules and TGFβ-receptor-2. Abrogation of TGFβ using different conditional knockouts expanded germinal-center-like intratumoral B-cells, enhancing immune-checkpoint-blockade efficacy. Finally, blocking αVβ8 integrin (which controls the release of active TGFβ) and PD-1 significantly increased B-cell-dependent animal survival and immunological memory. Our study highlights the importance of intratumoral B-cell immunity and a remodeled approach to boost the effects of immunotherapy against brain tumors.


Introduction Glioblastoma (GBM) is an immunologically cold tumor characterized by minimal lymphocyte in ltration
and signi cant myeloid cell prevalence. Besides the intrinsically low mutational burden of GBM 1 , the suppressive tumor microenvironment (TME) plays a major role in developing a nonin amed tumor 2,3 . Key drivers of this immunosuppression include a combination of secreted soluble factors such as cytokines and prostaglandins. Immunosuppression is also mediated by cellular components such as regulatory Tcells (Tregs), regulatory B-cells (Bregs), myeloid-derived suppressor cells (MDSCs), and tumor-associated macrophages (TAMs) 3,4 . Together through pathways such as the programmed cell death protein/programmed death ligand 1 (PD-1/PDL-1) axis or CD155 (poliovirus receptor) pathway, these players work together to inhibit effector functions of immune cells 5,6,7 .
While immune checkpoint blockade therapy has revolutionized the treatment of many solid cancers, these results have yet to be translated to GBM 8 . In contrast to the lymphocyte-excluded nature of GBM, melanoma brain metastases (MBM) are characterized by an increase in lymphocyte in ltration and a decrease in myeloid cells with clinically meaningful intracranial e cacy for immune checkpoint blockade 9,10 . However, patients with MBMs exhibit worse response to checkpoint blockade compared to patients with metastatic melanoma that does not involved the central nervous system 11 . Thus, melanoma brain and extracranial metastases serve as a control tumor model to study both the larger CNS role and tumorintrinsic properties in GBM immune suppression.
B-cells are an understudied yet critical player in brain tumor immunity that can serve both regulatory and anti-tumor functions 12,13 . B-cells begin at a naïve state and upon BCR and cytokine stimulation, undergo differentiation into non-switched IgM + memory B-cells within the germinal center 14 . Further activation B-cell Localization and Activation Status in Brain Tumors First, we characterized the abundance and localization of intratumoral B-cells in GBM using single-cell deconvolution of spatially resolved transcriptomic pro ling and B-cell marker expression (CD20, IGHM, and CD27) of 16 human GBM specimens ( Fig. 1A and B). Distribution and accumulation across all samples revealed high heterogeneity, including higher B-cell abundance (> 20 B-cell-enriched spots, n = 4), intermediate (5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15)(16)(17)(18)(19) B-cell-enriched spots, n = 7), and low B-cell abundance (0-4 B-cell-enriched spots n = 5) ( Fig. 1B and C). However, not all tumors had in ltration of B-cells. This was evidenced by samples that had robust B-cell accumulation in benign peritumoral regions but minimal B-cells in ltration within the tumor mass (Fig. 1B). Thus, the spatial analysis excluded those B-cells located in the peritumoral brain tissue. Study of co-localization demonstrated an absence of explicit assignment to transcriptional niches in contrast to myeloid or T-cell accumulation as described recently 50 (Fig. 1C). Moreover, B-cells could be characterized relative to different zones of the GBM tumor, including regions with abundant radial glia, oligodendrocyte progenitor cells (OPC), neuronal development, reactive immune response, and reactive hypoxia. We grouped tumor samples based on levels of B-cell in ltration, and in tumors with high in ltration, B-cells tended to cluster within zones of reactive immune response and hypoxia (Fig. 1D).
Further spatial analysis of intratumoral B-cell accumulation revealed three patterns encompassing tissueresident B-cell in ltration, and perivascular and intravascular B-cell accumulation. The most common pattern was a perivascular accumulation, further con rmed by spatial weighted correlation analysis in which perivascular space was highly signi cantly co-localized (Sup Fig. 3B, vascular analysis).
To gain an understanding of the functional status of B-cells in brain tumors, we studied their activation status in both GBM and MBM tissues. Using single-cell RNA sequencing, we analyzed the TME of primary GBM tumor tissues pooled across multiple patient cohorts and identi ed the B-cell population (Fig. 1E). For a non-GBM CNS cancer control, we also analyzed the TME of both MBM and extracranial tumors (ECM), identifying B-cells in both brain metastasis and extracranial samples (Fig. 1J-K). Upon stratifying B-cells into naïve, activated, and plasmablast subsets, we found that B-cells in extracranial melanoma tumors were over 75% plasmablasts. In contrast, MBM B-cells mainly consisted of activated B-cells with less than 50% plasmablasts (Fig. 1L). Our data suggest that tumor-in ltrating B-cells across GBM and MBM are antigen-experienced, with GBM having signi cantly reduced plasmablasts than both MBM and ECM. However, MBM have fewer plasmablasts compared to ECM.
Building from our observations in human GBM and MBM, we determined that similar B-cell phenotype exists in our murine CT2A orthotopic glioma models. Nur77-GFP mice (where lymphocytes express GFP upon receptor activation via antigen crosslink 51 ) were injected with CT2A glioma cells and allowed to develop tumors. We found that B-cells in the tumor expressed elevated GFP compared to B-cells in the periphery (deep cervical lymph node), suggesting that tumor B-cells had undergone B-cell receptor (BCR) activation (Sup Fig. 1A Fig. 1C). In melanoma samples, we saw higher levels of CD22 in intratumoral B-cells of MBM than in ECM B-cells (Fig. 2B).
Using single-cell transcriptomic data, we found that B-cells in murine CT2A tumors also express elevated Cd72, Fcgr2b, and Cd22 (Sup Fig. 1D-E). These checkpoint molecules, as well as downstream ITIM signaling molecules, SHIP1 and SHIP2 60 , were upregulated in the tumor B-cells compared to splenic Bcells at both the transcription and protein levels (Sup Fig. 1D-E).
We then conducted an unbiased interactome analysis in GBM and murine CT2A tumors using single-cell transcriptomic data to rank B-cell surface molecules and paired receptors or ligands on other cells based on expression levels (Sup Fig. 2A-B). From this analysis, we found elevated expression of CD22, CD72, and SIGLEC10 paired with elevated expression of their respective ligands PTPRC, SEMA4D, and CD52 on other B-cells, TAMs, or tumor cells (Fig. 2C).
Our results suggest that B-cells in human GBM and melanoma brain metastases, as well as murine CT2A glioma models, are activated but express elevated levels of key checkpoint molecules such as CD22/Cd22 and CD72/Cd72, leading to an inability to differentiate into plasmablasts fully.
Brain Tumors Produce B-cell-suppressing Cytokines with Corresponding Receptors Expressed by Intratumoral B-cells.
Having established that B-cells in GBM, MBM, and murine CT2A models are unable to fully differentiate into plasmablasts and have elevated expression of B-cell checkpoint molecules, we then sought to identify factors produced by the tumor and its TME that prevented B-cells from becoming fully activated and entering a plasmablast state.
Several known cytokines and pathways inhibit B-cell function, such as IL10, IL7, IL4, and the TGFβ superfamily 32 . Using our single-cell transcriptomic data for the GBM TME, we analyzed the expression of suppressive cytokines based on cell type clustering. We found that TGFβ-1 was highly expressed by Tcells and tumor-associated macrophages (TAMs) (Fig. 3A, Sup Fig. 3A). BMP7, which is part of the TGFβ superfamily, was also found to be expressed by tumor cells. Next, we analyzed B-cells expressing paired cytokine receptors and found that TGFBR2 was elevated (Fig. 3B). The three different TGFβ-R forms (TGFβ-R1, TGFβ-R2, and TGFβ-R3) can form both homo-and hetero-dimers, which allows TGFβ-1 signaling to occur through TGFβ-R2 61 . While single-cell transcriptomic data pointed to myeloid cells as the primary producers of TGFβ-1 in the tumor environment, spatial transcriptomics highlighted myeloid cells as correlated most with B-cells for spatial interactions (Sup Fig. 3B).
Because B-cells in MBM exhibited similar functional suppression as observed in GBM, we also analyzed TME suppressive cytokine expression and intratumoral B-cell cytokine receptor expression using singlenuclei transcriptomic data. As we observed in GBM, the TME in MBM also expresses elevated levels of TGFB1 associated most strongly with myeloid cells (Fig. 3C, Sup Fig. 3C). Inhibitory cytokine receptor analysis on intracranial B-cells showed the highest expression of TGFBR2 compared to other receptors and elevated TGFBR2 on intracranial B-cells compared to extracranial B-cells (Fig. 3D).
We ran the same analysis in our murine CT2A model and found Tgfb1 and Bmp7 expressed by myeloid and tumor cells, respectively (Sup Fig. 3D). Tgfb3 was also expressed by tumor cells. Comparison between brain tumor and peripheral myeloid cells showed that expression of TGFβ-1 on myeloid cells was unique to tumor-resident myeloid cells (Sup Fig. 3E). Receptor analysis on intratumoral B-cells also showed high levels of Tgfbr2 and Tgfbr1 (Sup Fig. 3F), and Tgfbr2 was elevated on intratumoral B-cell compared to splenic B-cells within the same animal (Sup Fig. 3G-H). GSEA enrichment analysis showed signi cantly increased expression of TGFβ responsiveness genes in B-cells in the tumor compared to Bcells in the spleen (Sup Fig. 3I).
TGFβ-1 + cells in the TME consisted mainly of SOX2 + tumor cells, but the proportion of TGFβ-1 + within each cell population was similar between tumor cells, myeloid cells and microglia despite lower cell counts for the latter two cell types (Fig. 3E). We also analyzed the distance between B-cells and other cell types since TGFβ activation and signaling require proximity between recipient B-cells and the TGFβ source 62 . We characterized the different cell types within a 15 µm radius of B-cells, as previous studies had demonstrated that 15 µm is a close distance enough to both form an immune synapse and initiate TGFβ-1 release 62,63 , and found that myeloid cells were the most common cell type within 15 µm radius of B-cells (Fig. 3F).
Our data suggest that tumor and myeloid cells within the GBM TME produce of suppressive cytokines, particularly TGFβ-1, that may interact with B-cells and impact their maturation and function.

TGFβ Inhibits B-cell Differentiation and Induces Checkpoint Molecule Expression
To determine the direct effects of TGFβ signaling on B-cells, we cultured human B-cells from healthy donors in a B-cell expansion media supplemented with or without exogenous TGFβ. We found that the Bcell expansion media could expand naïve B-cells. TGFβ-1 signi cantly inhibited B-cell expansion as measured by cell proliferation and cell counts (expansion index mean±SD: Day 7 B-cell + DMSO vs. Day 7 B-cell + TGFβ-1 14.83 ± 2.53 vs 6.60 ± 0.59, p < 0.01, Fig. 4A-B). This inhibitory effect of TGFβ-1 could be prevented using SB431542, a TGFβ-R inhibitor (expansion index mean ±SD: B-cell + TGFβ-1 + DMSO vs Bcell + TGFβ-1 + SB431542 0.97 ± 0.02 vs. 2.41 ± 0.12 p < 0.0001, Fig. 4C). We also found that expansion and activation of B-cells were correlated with the decrease in checkpoint molecules such as CD22 and CD72 over time, suggesting these B-cells could differentiate into a plasmablast state. However, culturing B-cells with TGFβ-1 prevented the decline of these checkpoint molecules ( Fig. 4D-E) which demonstrated that TGFβ-1 could hinder B-cell differentiation. After culturing B-cells with exogenous TGFβ-1, western blots for TRAF6, a molecule recruited downstream of TGFβ-R, con rmed the presence of TGFβ signaling (Fig. 4F). We also showed that TGFβ-1 stimulation increased SHIP1 and SHIP2, critical molecules of checkpoint-mediated ITIM suppression (Fig. 4F).
Because we found that both myeloid and tumor cells produce TGFβ-1 in our murine CT2A model, we In summary, a screen of B-cell suppression cytokines in the TME environment revealed that myeloid cells across all tumor types produce TGFβ-1, with corresponding high expression of its ligand TGFβ-R2 on tumor B-cells. The data also show that TGFβ-1 directly inhibits B-cell proliferation and downstream activation by maintaining expression of inhibitory checkpoint molecules within the TME.
Neoadjuvant PD-1 Checkpoint Blockade Therapy Induces Bcell Checkpoint Molecules and Susceptibility to TME Suppression We evaluated how current immune checkpoint blockade treatment regimens affected the TME B-cell compartment in GBM. Recurrent GBM patients were categorized based on whether they had received neoadjuvant Pembrolizumab treatment. B-cells in Pembrolizumab treated and untreated were analyzed for checkpoint molecules and suppressive cytokine receptors.
The data showed that neo-adjuvant pembrolizumab treatment was correlated with a 2-fold increased B

Blocking TGFβ and PD-1 Eradicates Tumors in a B-cell Dependent Manner
While targeting the TGFβ pathway has been attempted in GBM, strategies such as using depletion antibodies or TGFβ receptor blockade has shown minimal clinical e cacy 64 65 . We targeted the TGFβ pathway in B-cells by blocking the αVβ8 integrin. αVβ8 integrin functions to cleave latency-associated peptide (LAP) to release activated TGFβ 66 , and is elevated in tumoral B-cells compared to peripheral Bcells (Fig. 6A). To evaluate the therapeutic bene t of αVβ8 integrin blockade and its effects on PD-1 e cacy, wild-type mice were implanted with CT2A tumors and treated with anti-αVβ8 Fab blocking antibodies with or without PD-1 blockade. We use the anti-αVβ8 Fab antibodies to prevent a Fc-mediated B-cell depletion. 200 ug anti-αVβ8 Fab antibodies were given every other day for two weeks. The same regimen was used for PD-1 blockade. While integrin blockade alone provided a signi cant survival bene t, a combination of anti-αVβ8 Fab and anti-PD-1 antibodies showed a synergistic effect, with nearly 60% of animals having complete tumor eradication (median survival days: No Tx vs. anti-αVβ8 Fab vs. anti-PD-1 vs. anti-αVβ8 Fab + anti-PD-1, 28 vs. 41.5 vs. 29 vs. Unde ned, Fig. 6B). Additionally, immune phenotyping of intratumoral B-cells revealed that dual αVβ8 and PD-1 blockade increased B-cell in ltration into the brain tumor, promoted B-cell proliferation, decreased expression of checkpoint inhibitory molecules, and decreased overall population of myeloid cells (Sup Fig. 5A).
To evaluate the effects of B-cells on immunological memory, long-term survivors from αVβ8 blockade and dual αVβ8 + PD-1 blockade groups were split into two groups, with one group receiving intracranial anti-CD20 B-cell depleting antibody and the other group having an intact B-cell compartment. Mice were then re-challenged with CT2A tumors in the contralateral hemisphere. Animals who had initially received dual-αVβ8 + PD-1 blockade and had an intact B-cell compartment were able to mount a memory response, with 50% of mice surviving the tumor re-challenge (median survival days: No Tx rechallenge vs. anti-αVβ8 Fab + anti-PD-1 rechallenge, 24 vs. 84, Fig. 6C). On the other hand, the group with B-cell depletion via anti-CD20 depleting antibody did not show a memory response and extended survival, even with initial dual-αVβ8 + PD-1 blockade treatment (median survival days: anti-αVβ8 Fab + anti-PD-1 rechallenge vs. anti-CD20 + anti-αVβ8 Fab + anti-PD-1 rechallenge, 84.5 vs. 34, Fig. 6C). To further evaluate the role of B-cells on the therapeutic bene t, the dual-αVβ8 + PD-1 blockade was given to B-cell knockout mice. Whereas wild-type mice saw signi cant survival bene t and tumor clearance from the combinational treatment with dual integrin and PD-1 blockade therapy (

TME Suppresses B-cell Plasmablast Differentiation and Antibody Production
Having shown that mitigating B-cell suppression using αVβ8 integrin blockade can increase the e cacy of checkpoint blockade and promote tumor clearance, we then evaluated how the combinational therapy affected the function of intratumoral B-cells. Using our in-vitro B-cell culture system, we studied the effects of TGFβ-1 on human B-cell differentiation into plasmablasts (CD38 + CD20 − ) 67 68 . First, we con rmed that B-cells in our expansion media could differentiate into plasmablasts, as B-cell proliferation was associated with a loss of CD20, but TGFβ-1 prevented both proliferation and CD20 loss (Fig. 7A).
Additionally, analysis of B-cells for plasmablast phenotype markers (CD38 + CD20 − ) showed increases in plasmablast differentiation over seven days of culture with the B-cell expansion media. Plasmablast differentiation was signi cantly suppressed by TGFβ-1 ( Fig. 7B-C). Moreover, western blots for IgG in our B-cell culture media showed that B-cells cultured with TGFβ produced less IgG (Fig. 7D).
To evaluate how integrin and checkpoint blockade could affect B-cell differentiation into plasmablasts invivo, CT2A tumor-bearing mice were treated with αVβ8 or PD-1 blockade monotherapy or a combination of αVβ8 and PD-1 blockade every other day for two weeks. Two weeks after tumor implantation, B-cells were isolated from the brain and analyzed for plasmablast differentiation. As we saw in our single-cell GBM analysis of B-cell phenotype in the tumor, CT2A tumor-bearing mice without any treatment showed minimal amounts of intratumoral plasmablasts (Fig. 7E). While treatment with only αVβ8 blockade or only PD-1 blockade did not show a signi cant increase of plasmablasts in the brain, dual αVβ8 and PD-1 blockade led to a substantial increase of total plasmablasts (CD38 + CD20 − ) and proliferating B-cells (%CD38 + CD20 − CD19 + plasmablasts mean ± SD: isotype control treatment vs. anti-αVβ8 + anti-PD-1 treatment, 0.1920 ± 0.384 vs. 8.076 ± 4.442, p < 0.01) (%Ki67 + CD19 + proliferating B-cells mean ± SD: isotype control treatment vs. anti-αVβ8 + anti-PD-1 treatment, 10.61 ± 2.051 vs. 22.96 ± 9.175, p < 0.05, Fig. 7E).
Thus, our results suggest that combination treatment with αVβ8 and PD-1 blockade can rescue B-cell function by promoting B-cell proliferation and plasmablast differentiation even in the suppressive TME. The data suggest that mitigating B-cell suppression in the TME can enhance the function of T-cells by leveraging the antigen presentation function of B-cells. Our study provides a strategy to combine T-cell checkpoint blockade and B-cell TGFβ pathway blockade to prompt a synergistic anti-tumor e cacy.

Discussion
Our study highlights a mechanism of brain tumor-mediated B-cell suppression that can be leveraged to augment the effects of immunotherapy. We quanti ed B-cell in ltration in GBM using spatial transcriptomics and characterized them as antigen-experienced but functionally inhibited via ITIM checkpoint molecules, a shared phenotype across GBM and MBM. Analysis across different types of human tumors suggests that a brain-intrinsic property prevents complete differentiation of B-cells into plasmablasts, as GBM B-cells had the fewest proportion of plasmablasts, and MBM had fewer plasmablasts compared to ECM. Furthermore, GBM B-cells had decreased checkpoint molecule expression compared to PBMC B-cells but still maintained expression of CD72 and CD22, suggesting that B-cells are activated upon recruitment to the brain, but the TME prevents complete transition into plasmablasts. While GBM and MBM share a lack of plasmablasts, their inhibitory checkpoint molecule expression pro le had subtle differences, indicating that GBM and MBM still have unique tumor properties and TMEs. Combining T-cell checkpoint blockade and indirect B-cell checkpoint blockade via integrin blockade allowed for more dramatic TME immune remodeling and anti-tumor effects. αVβ8 integrin is also expressed on other cell types such as CNS cells and the tumor 71 72 -thus αVβ8 integrin blockade can potentially have direct anti-TGFβ effects on tumor cells as well.
The role of B-cells in the TME can be broadly categorized in 3 main categories: antibody production 40 73 , antigen presentation 21 , and cytokine secretion 74 . We demonstrate an increase in plasmablast differentiation in the TME after dual αVβ8 integrin and PD-1 blockade, and future work will be done to characterize potential antibody reactivity. While our study focused on IgG production, previous studies have shown that TGFβ-R2 deletion in murine B-cells leads to a complete absence of IgA in the serum 75thus further work can be done to study the effects of our dual αVβ8 integrin and PD-1 blockade on Ig class switching.
We also identi ed increased spatial co-localization and co-expression of HLA and both CD4 and CD8 molecules in B-cell-rich zones, suggesting formation of potential germinal center-like structures with T-cell activation. Moreover, cell-cell interactome analysis based on spatial transcriptomic identi ed increased interaction between B-cells and NK cells. Such B-cell-NK interactions could suggest antibody dependent cytotoxicity or B-cell cytokine production 76 . The reverse interaction can also be possible, where NK cells in the TME can regulate or activate B-cells through interferon signaling 77 .
Overall, our study proposes the importance of leveraging B-cell immunity to target both GBM and other brain tumors. Blockade of αVβ8 integrin promotes B-cell effector function in a B-cell-speci c manner, making its clinical translation highly feasible. Future directions of this study should revolve around utilizing insights into TME immunosuppression to develop a new generation of therapies with potent cytotoxicity and mechanisms of protection against the TME integrated into their designs.

Human Samples
The Nervous System Tumor Bank at Northwestern University was used for the collection of all human samples. Approval for collection was granted under the institutional review board protocol number STU00202003 and the study was conducted following the U.S. Common Rule of ethical standards. All patients signed written consent forms. Collected samples included tumor, peripheral blood, and frozen tissue from GBM patients with at least 50% tumor cellularity, as determined by neuropathologist review of H&E sections. Mice C57BL/6, B-cell Knockout (µMT), Nur77-GFP, CD19 Cre , LyzM Cre , Tgfb1 ox , and Tgfbr2 ox were all purchased from The Jackson Laboratory and bred for use in experiments. Studies were initiated when the mice were 6-8 weeks old. Approval for all animal experimental protocols were approved by the Institutional Animal Care and Use Committee at Northwestern University under the protocol number ISO16696. All animals were housed at the Center for Comparative Medicine at Northwestern University in a dedicated pathogen-free animal facility with 12-hour light/12-hour dark cycles and ad libitum access to food and water.

Generation of single gene knockout of TGFβ-1 in CT2A glioma cells
Single gene knockout clones were generated in lentiCRISPRv2 (one vector system). The vector backbone was purchased from Addgene (lentiCRISPR v2 was a gift from Feng Zhang (Addgene plasmid # 52961; http://n2t.net/addgene:52961 ; RRID:Addgene_52961) 78 . The protocol for guide cloning and generation of the virus was as described in Sanjana et al. 78 . The guide sequence for mouse TGFβ-1 KO is "CACCGTTGACGTCACTGGAGTTGTA" and non-targeting control (NTC) is "CACCAATATTTGGCTCGGCTGCGC". The TGFβ-1 KO and control clones were selected using puromycin from Sigma (2ug/ml) in CT2A mouse glioma cell line. The TGFβ-1 KO was con rmed using western blotting (TGFβ-1 antibody from Abcam [EPR21143] (ab215715)). Readings were taken every 24 hours for the next 5 days

Intracranial Tumor Injections
Each mouse was injected with 1x10 5 tumor cells in a total volume of 2.5ml of PBS. Mice were anesthetized with ketamine (100 mg/kg) and xyalizine (10 mg/kg) via intraperitoneal injection. After shaving of surgical site and disinfection with povidone-iodine and 70% ethanol, an incision was made at the midline for access to the skull. A 1 mm-diameter burr hole was drilled 2mm posterior to coronal suture and 2 mm lateral to the sagittal suture. Injections were performed using a Hamilton syringe tted with a 26-guage blunt needle at a depth of 3.5mm. The injection site was then sutured closed.

Intracranial Cannula Implantation
Mice were anesthetized, and a skin incision ∼10 mm in length was made over the middle frontal to the parietal bone to expose the surface of the skull. A 26-gauge sterile guide cannula for mice (Plastics One) was installed into the mouse brain at 2-mm depth through the burr hole generated during tumor implantation, as described above. Tissue glue was applied around the burr hole to secure the protrusion of the cannula for long-term stable positioning. The scalp was closed with surgical glue around the implantation site. A protection dummy cannula was used to secure the protrusion end during the postop recovery and following observation period.

Spatial weighted Correlation analysis
For spatial weighted correlation analysis, we used the SPATAwrappers package and the runSpatialRegression function. Spatially correlation analysis was performed by either a spatial Lag model or a Canonical Correlation Analysis (CCA).

Cell Type Deconvolution
Cell type deconvolution of each spot was performed by RobusT-cell Type Decomposition (RCTD) a wellvalidated toolbox 79 . The deconvolution was performed by the SPATAwrappers (https://github.com/heilandd-/SPATAwrappers) package using the function runRCTD. Visualization of surface plots or correlation analysis was performed by the SPATA2 toolbox.

Spatial Receptor Ligand Interactions
Recepor-ligand interaction analysis was performed by the NFCN2 package (https://github.com/heilandd/NFCN2). First we identi ed spots with B-cell accumulation (as explained before), neighborhood spots (with a distance of 110 µm) and non B-cell enriched spots. We next quanti ed the co expression of receptor and ligand pairs using the NFCN2 databases. Co expression visualization was performed by the SPATAwrappers package using the plotColorOverlap function.

Single-cell and single-nuclei RNA Sequencing and Analysis
For human GBM samples, public single-cell RNAseq of GBM brains, as well as RNAseq from two paired tumor and PBMC patient samples, were used for the analysis. Single-cell preparation was done using the adult brain dissociation kit for mouse and rat (Miltenyi Biotec). After ensuring cell viability of greater than 70%, cells were given to the RNA sequencing core at Northwestern University for further processing and sequencing.
For murine single-cell transcriptomic analysis, intracranial tumors from CT2A-bearing C57BL/6 mice were dissected from the brains of mice 14 days after CT-2A tumor implantation. Three tumors were pooled for this analysis. Single-cell preparation was done using the adult brain dissociation kit for mouse and rat (Miltenyi Biotec). After ensuring cell viability of greater than 70%, cells were given to the RNA sequencing core at Northwestern University for further processing and sequencing. To visualize the single-cell RNA-seq results, the normalized gene barcode matrix was used to compute a neighborhood graph of cells, then Uniform Manifold Approximation and Projection (UMAP) was performed with default parameters. The whole pipeline was implemented using Scanpy 84 . Cell type annotation was performed using R package SingleR 85 . Differential expression analysis was performed by Wilcoxon rank-sum test between groups. GSEA of selected pathways was performed using the Python package GSEApy (https://gseapy.readthedocs.io/en/latest/index.html).
Bulk RNA Sequencing and Analysis 14 days after implantation, tumors were isolated from 10 mice for each n included in this study and enriched for CD19 using biotinylated CD19 antibody (Biolegend) and anti-biotin microbeads (Miltenyi Biotec). RNA from these samples was isolated using TRIzol (Thermo Fisher Scienti c) to purify. Chloroform (0.2ml) was added to the TRIzol samples. The top layer containing the RNA was precipitated with 70% isopropanol. The resulting pellets were dried, resuspended in sterile water, and sent to Novogene for further processing and analysis. Novogene assessed the RNA for quality and provided all data as total counts and fragments per kilobase per million reads (fpkm).

Multiplex immuno uorescence staining
Sections of 5µm thickness were obtained from FFPE embedded tumor tissue. Depara nization of the slides was achieved using xylene followed by rehydration in histological grade ethanol and xed with 3% hydrogen peroxide in methanol before heat-induced epitope retrieval using BOND epitope retrieval solution (pH6) or pH9 EDTA buffer for 20 minutes. 3,3'-diaminobenzidine chromogen staining was initially performed to determine the optimal concentrations of each antibody in human GBM tissues.
Primary antibodies were diluted with 1x Opal Antibody diluent/block solution and were used in the following order coupled with the indicated Opal dyes Multiplex staining was performed in multiple cycles involving a heat-induced epitope retrieval step, protein blocking, epitope labeling, and signal ampli cation. Once all markers were stained, spectral DAPI was used to counterstain the slides and were mounted using Prolong Diamond Antifade Mountant.

Imaging and analysis of multispectral images
Multispectral imaging (MSI) was performed using the Vectra 3 Automated Quantitative Pathology Imaging System from Akoya Biosciences. First, whole slide images were acquired after auto-adjusting focus and signal intensity. Then, MSI was acquired from the tumor regions delineated by a certi ed neuropathologist at 20x of the original magni cation. For analysis of MSI, we created a spectral library for all Opal dyes to subject acquired multispectral images to spectral unmixing that enabled the identi cation and separation of weakly expressing and overlapping signals from background to visualize the signal of each marker (SOX2, CD20cy, TMEM119, CD163, TFG-b, CD8, DAPI) in inForm Tissue Finder software (inForm 2.6, Akoya Biosciences). Using InForm, the adaptive cell segmentation feature was used to identify the analyzed cells' nuclei and determine each cell's nuclear and cytoplasmic compartments. A machine-learning algorithm within inForm was used in which cells were automatically assigned to a speci c phenotype (SOX2+, TMEM119+, CD163+, CD20+, CD8+, TFG-b+). Batch analysis was used to analyze all tumor samples under the same segmentation and phenotype settings. The processing and analysis of images from all tumor samples were exported to cell segmentation tables. Exported les from inForm were processed in R using R packages Phenoptr and PhenoptrReports to merge and create consolidated single les for each tumor sample. Consolidated les had triple cell phenotypes as outputs that we employed for further quanti cation and spatial analyses using the Phenoptr R addin.
Quanti cation of cell types and spatial analysis of GBM samples.
Merged and consolidated les were analyzed using Phenoptr to quantify the cell density of SOX2 + TGFβ-1+, TMEM119 + TGFβ-1+, CD163 + TGFβ-1+, TMEM119 + CD163-TGFβ-1+, and CD163 + TMEM119-TGFβ-1+, CD20+, CD8+. For the spatial analysis, mean cell counts within a speci ed radius of 15 mm from a given cell type to another were calculated using Phenoptr as an R addin. Then, mean distances between the nearest neighbors were calculated from particular cell types to other cell types. The spatial map viewer addin within R allowed the visualization of the nearesT-cell neighbor between selected phenotypes in a single eld. Cartoons were created using Adobe Illustrator version 22.1.0.

Isolation of GBM in ltrating Immune cells and PBMCs
Freshly resected GBM tumor samples were collected and transported in complete RPMI media (RPMI + 10% heat-inactivated FBS, 10µM HEPES-sodium pyruvate, 1 mM sodium pyruvate, 0.01% 2mercaptoethanol, 2mM L-glutamine, penicillin (100 U/mL), and streptomycin (100ug/mL); all reagents from Thermo Fisher). Samples were manually diced with a razor blade and enzymatically digested using the Adult Brain Digestion Kit (Miltenyi Biotec) according to manufacturer instructions.
Peripheral blood samples from GBM patients were collected in EDTA tubes. Peripheral blood mononuclear cells (PBMC) were isolated using the Ficoll (GE Healthcare) gradient. Tumor cells and PBMCs were immediately put in complete RPMI media after isolation.

In-vitro B-cell Culturing and Phenotypic Analysis
Human B-cells were isolated from healthy donor PBMCs using the Easysep B-cell isolation kit (Stemcell Technologies). Isolated B-cells were cultured in ImmunoCult-XF T-cell Expansion Medium supplemented with ImmunoCult-ACF Human B-cell Expansion Supplement (Stemcell Technologies) for up to 7 days. For cells cultured with TGFβ-1, recombinant human TGFβ-1 (Peprotech, 10ng/mL) was added every day for the duration of the culture. For cells also cultured with SB431542 TGFβ receptor inhibitor, SB431542 (10uM) was 30 minutes before the addition of TGFβ-1.

Western Blots
Cells were removed from the medium, washed once with ice-cold PBS, and lysed in ice -cold mammalian protein extraction reagent (M-PER; Thermo Scienti c, Rockford, IL, USA) supplemented with protease and phosphatase inhibitor cocktail (PPI; Thermo Scienti c, Rockford, IL, USA). Cell lysates were sonicated in a water bath in 3x30-second increments, with 30 seconds of rest between each sonication. The resulting lysates were centrifuged at 14,800 rpm for 10 min at 4°C. Following supernatant collection, the protein concentration for each western blot sample was determined via BSA assay (Thermo Scienti c, Rockford, IL, USA). Lysate concentrations were normalized,, and 4x laemmli sodium dodecyl sulfate buffer (SDS sample buffer; Boston BioProducts, Boston, MA, USA) was added. Samples were boiled at 95°C for 10 minutes in preparation for gel loading. Samples were then loaded and run through 4-15% SDSpolyacrylamide gels (BioRad, Hercules, CA, USA) at 45V initially and then 100V until completion. Proteins were transferred onto 0.2µm polyvinylidene di uoride (PVDF) membranes (Millipore, Darmstadt, Germany) at 14V constant in a Trans-Blot semi-dry transfer machine. Following the transfer, the membranes were rinsed in distilled water brie y and transferred to a blocking solution(5% powdered milk made in TBS-T) for one hour at room temperature. Following block, membranes were washed in TBS-T 3x10 minutes and incubated in primary antibody solution overnight at 4°C. Primary solutions were made 1:1000 in 5% bovine serum albumin solution supplemented with sodium azide 0.02% (w/v) (Cell Signaling, TRAF6 (D21G3) Rabbit mAb #8028, SHIP1 (P290) Antibody #2726, SHIP2 Antibody #2730).
The next morning, membranes were washed in TBS-T 3x10 minutes and incubated at room temp in secondary antibody (Cell Signaling, Anti-rabbit IgG, HRP-linked Antibody #7074) diluted 1:4000 in 5% milk for 1 hour. Following secondary another 3x10 minutes wash cycle in TBS-T, membranes were coated with an enhanced chemiluminescence solution (ECL; Clarity ECL, BioRad), and images were developed in a developer machine (BioRad, Hercules, CA, USA). Data and Code Availability UCLA GBM scRNAseq data was downloaded from GEO GSE154795. We also used another public dataset from GEO GSE182109. All the custom code will be made available upon request.

Statistical Analysis
Data are shown as mean ± standard deviation for continuous variables and numbers for categorical variables. Differences between 2 groups were analyzed using Student's t test. Multiple groups were analyzed using one-way ANOVA with a post hoc Tukey's multiple comparisons test. Survival curves were analyzed and generated using the Kaplan-Meier method and compared by log-rank test. Categorical variables were analyzed using Fisher's exact tests or X 2 tests as appropriate. All tests are two-sided with p values or Benjamini-Hochberg adjusted false discovery rates of < 0.05 were considered signi cant. Statistical analyses were performed using GraphPad Prism 9.4.1. ns = p > 0.05, * = p < 0.05, ** = p < 0.01, *** = p < 0.001, **** = p < 0.0001.    -cell suppressive   cytokines IL10, IL4, IL7, TGFB1, TGFB2, TGFB3, and BMP7. TGFB1 is highly expressed in the melanoma  metastases TME by TAMs, T-