Abstract

   While the accumulation of tumor-associated macrophages (TAMs) in
   glioblastoma (GBM) has been well documented, targeting TAMs has thus
   far yielded limited clinical success in slowing GBM progression due, in
   part, to an incomplete understanding of TAM function. Using an
   engineered 3D hydrogel–based model of the brain tumor microenvironment
   (TME), we show that M2-polarized macrophages stimulate transcriptional
   and phenotypic changes in GBM stem cells (GSCs) closely associated with
   the highly aggressive and invasive mesenchymal subtype. By combining
   proteomics with GBM patient single-cell transcriptomics, we identify
   multiple TAM-secreted proteins with putative proinvasive functions and
   validate TGF-β induced (TGFBI, also known as BIGH3) as a targetable
   TAM-secreted tumorigenic factor. Our work highlights the utility of
   coupling multiomics analyses with engineered TME models to investigate
   TAM–cancer cell crosstalk and offers insights into TAM function to
   guide TAM-targeting therapies.

   Keywords: Immunology, Oncology

   Keywords: Brain cancer, Extracellular matrix, Macrophages
     __________________________________________________________________

   [39]graphic file with name jciinsight-10-181903-g046.jpg
     __________________________________________________________________

   Multi-omics analysis of 3D cell culture and patient biopsies identifies
   candidate glioblastoma therapeutic targets among M2 macrophage-secreted
   factors.

Introduction

   Glioblastoma (GBM) is the most common, lethal, and aggressive form of
   primary brain cancer, with a medium survival time of approximately 15
   months ([40]1). The highly invasive nature of GBM complicates surgical
   resection and promotes recurrence, yet it has been challenging to
   target invasion therapeutically due to an incomplete understanding of
   the underlying cellular and molecular mechanisms. Cell invasion is
   profoundly influenced by interactions between cancer cells and the
   tumor microenvironment (TME), which consists of non-neoplastic cells as
   well as extracellular matrix (ECM) ([41]2). Immune cells have emerged
   as pivotal TME players in tumor development, making them highly
   attractive targets for therapeutic intervention ([42]3, [43]4). GBMs
   exhibit a unique immune landscape dominated by tumor-associated
   macrophages (TAMs), which are composed of multiple subpopulations,
   including bone marrow–derived (BMD) macrophages and brain-resident
   microglia ([44]5, [45]6). TAMs can comprise up to 40% of the total
   cells in gliomas, and their accumulation is highest in patients with
   the most aggressive GBM subtype ([46]6–[47]9). TAMs are proposed to
   actively regulate multiple cellular properties relevant to tumor
   progression, including stemness, proliferation, survival, and
   migration. Investigation into the mechanisms regulating these complex
   TAM-GBM interactions is ongoing. Several TAM-secreted factors,
   including TGF-β family members, interleukins, and C-C chemokine
   ligands, activate downstream oncogenic signaling pathways, including
   β-catenin signaling, AKT, NF-κB, and JAK/STAT3 ([48]10–[49]13). The
   clinical impact of TAM-tumor interactions, however, is strongly
   influenced by patient-specific characteristics of the tumor (location
   and stage) as well as macrophage cell phenotypes ([50]14–[51]19).

   Macrophage phenotype and function are defined by a polarization state,
   which has traditionally been described using the M1/M2 polarization
   model ([52]20, [53]21). TAMs are often associated with the
   antiinflammatory M2-like polarization state due to their high
   expression of antiinflammatory cytokines, scavenger receptors,
   proangiogenetic factors, and proteins involved in ECM organization and
   remodeling ([54]22). The proinflammatory M1-like state is more commonly
   associated with antitumor functions such as antigen presentation and
   costimulation. Although the M1/M2 classification system has proven to
   be valuable for in vitro studies, TAM polarization is likely more
   nuanced than the binary M1/M2 classification would suggest, and human
   GBM TAMs can display both M1 and M2 gene signatures ([55]23–[56]26).
   Thus, M1 and M2 TAMs are more accurately described as antitumor
   (M1-like) and protumor (M2-like) macrophages. We nonetheless use the
   M1/M2 nomenclature here for the sake of conciseness and consistency
   with the literature.

   An important and ongoing barrier to understanding TAM regulation of GBM
   invasion is the lack of in vitro platforms capturing key geometric and
   mechanical aspects of the GBM TME. Advances in biomaterials and tissue
   engineering have facilitated the development of 3D TME models ([57]27,
   [58]28). Hydrogel-based systems are frequently used as matrices for 3D
   cell culture, and among these systems, hyaluronic acid (HA) is a
   valuable biomaterial for modeling the brain TME ([59]29). We and others
   have used HA-based hydrogel platforms to identify and mechanistically
   dissect key pathways driving GBM invasion, metabolism, and therapeutic
   resistance ([60]30–[61]36). Despite the recent expansion of 3D brain
   TME models, integration of macrophages and other immune cells remains
   limited ([62]37, [63]38).

   Here we utilize a 3D HA-based hydrogel platform to test the hypothesis
   that TAMs establish a proinvasive microenvironment that facilitates GBM
   invasion. We first investigate the influence of macrophage polarization
   state on GBM spheroid invasion across a panel of patient-derived GBM
   cells. We find that factors secreted by M2-polarized macrophages induce
   GBM invasion. Using a microscale invasion platform that enables
   regional dissection and characterization of invasive cells, we identify
   macrophage-induced transcriptional changes in invasive GBM cells. We
   then characterize the macrophage secretome across polarization states
   using mass spectrometry, perform a GBM-TAM interaction analysis to
   identify receptor-ligand pairs driving invasion, and prioritize
   putative proinvasive TAM-secreted ligands using single-cell RNA
   sequencing (scRNA-seq) datasets of human GBM TAMs. We identify 2
   TAM-derived secreted factors, BIGH3 (also known as TGF-β induced or
   TGFBI) and S100A9, that stimulate GBM invasion. We demonstrate that
   targeting BIGH3 and downstream mammalian target of rapamycin (mTOR)
   signaling reduces invasion. Overall, our study highlights the utility
   of engineered 3D platforms for investigating TAM–cancer cell
   interactions and uncovering molecular targets that may guide the
   development of future therapeutic intervention.

Results

Establishment of a brain-inspired hydrogel platform to study TAM-GBM
interactions.

   We engineered a 3D in vitro brain tumor model leveraging our previously
   published hydrogel consisting of HA conjugated to the integrin-binding
   peptide GCGYGRGDSPG (HA-RGD) ([64]39) ([65]Figure 1A and
   [66]Supplemental Figure 1A; supplemental material available online with
   this article; [67]https://doi.org/10.1172/jci.insight.181903DS1). To
   study the influence of macrophages on GBM invasion, we performed
   coculture tumorsphere invasion assays by encapsulating patient-derived
   GBM stem cells (GSCs) in HA-RGD hydrogels with THP-1–derived
   M2-polarized macrophages distributed throughout the hydrogel
   ([68]Figure 1, B and C, and [69]Supplemental Figure 1B). Macrophage
   polarization state was confirmed via quantitative PCR (qPCR) and flow
   cytometry ([70]Supplemental Figure 1, C–G). As monocultures, GSC-295
   spheroids were not invasive; however, adding M2-polarized macrophages
   resulted in a proinvasive GSC phenotype characterized by long, thin
   protrusions, quantified by invasive cell area. To test whether M2
   macrophages influence invasion through direct or paracrine mechanisms,
   we performed indirect coculture assays using M2-polarized
   macrophage–conditioned media (M2 CM). M2 CM treatment recapitulated the
   invasive phenotype of direct coculture, suggesting GSC invasion did not
   require macrophage-GSC physical contact ([71]Figure 1, D and E, and
   [72]Supplemental Figure 2A).

Figure 1. An engineered biomaterials platform to enable investigation of
GBM-macrophage crosstalk.

   [73]Figure 1
   [74]Open in a new tab

   (A) Schematic of HA-based hydrogel. (B) Schematic of mono- or coculture
   platforms to study GBM invasion. (C) Schematic of THP-1–derived
   monocyte-to-macrophage differentiation and M1/M2 polarization protocol.
   (D and E) GSC-295 invasion assay with direct M2 macrophage coculture
   and M2 macrophage–conditioned media (CM) (n = 19 spheres) showing (D)
   quantification and (E) representative phase images. (F and G) GSC-295
   invasion assay with CM from monocytes (mono.), M1 macrophages, and M2
   macrophages (n = 30 spheres) showing (F) quantification and (G)
   representative phase images. (H and I) GSC-268 invasion assay with M2
   CM (n = 31 spheres) showing (H) quantification and (I) representative
   phase images. (J and K) GSC-11 invasion assay with M2 CM (n = 39
   spheres) showing (J) quantification and (K) representative phase
   images. (L and M) GSC-268 invasion assays with GBM patient–derived
   macrophage CM (n = 18-59 spheres) showing (L) quantification and (M)
   representative phase images for Patient A. Spheroid invasion results
   were pooled across at least 2–3 independent replicates. Statistical
   significance was analyzed using 1-way ANOVA followed by Tukey’s
   multiple-comparison test (D, F, and L) or an unpaired, 2-sided
   Student’s t test (H and J). *P < 0.05; **P < 0.01; ***P < 0.001; ****P
   < 0.0001. Scale bars: 200 μm.

   To examine the effect of polarization state on macrophage-GSC
   interactions, we treated GSC-295 spheroids with CM collected from
   monocytes and M1-polarized macrophages. GSC-295 spheroids cultured in
   monocyte CM (mono. CM) and M1 macrophage CM (M1 CM) were less invasive
   than GSC-295 spheroids cultured in M2 CM ([75]Figure 1, F and G). These
   results suggest that in our system, M2 macrophages are the predominant
   source of proinvasive factors.

GBM-macrophage interaction varies across a panel of patient-derived GBMs.

   To determine whether M2 CM stimulated invasion of additional
   patient-derived GSCs, we performed tumorsphere invasion assays using a
   previously characterized panel of patient-derived GSCs ([76]40).
   Although the influence of M2 CM on spheroids appeared to be cell line
   specific, the majority of GSC lines (5 out of 6) were highly invasive
   when treated with M2 CM ([77]Figure 1, H and I, and [78]Supplemental
   Figure 2, B–G). Interestingly, GSC-11 spheroids decreased in area when
   treated with M2 CM ([79]Figure 1, J and K). We confirmed the
   polarization-specific response and found that M2 CM drives GSC-268
   invasion greater than monocyte CM or M1 CM ([80]Supplemental Figure 2,
   H and I).

   To explore the clinical relevance of these findings, we obtained
   blood-derived macrophages from multiple patients with GBM (n = 5),
   polarized the macrophages to M1 (with LPS and IFN-γ) or M2 (with IL-4),
   and collected CM for GSC-268 tumorsphere invasion assays. Similarly to
   THP-1–derived macrophages, patient-derived macrophage CM produced a
   polarization-specific response in GSC invasion, with M2 CM driving a
   more invasive phenotype relative to M1 CM ([81]Figure 1, L and M).

   To test whether M2 CM stimulates invasion in patient-derived cells that
   have not been enriched for GSCs, we performed tumorsphere invasion
   assays with 2 GBM patient–derived xenograft lines (GBM-43 and GBM-6)
   ([82]41). Interestingly, GBM-43 spheroids decreased invasion when
   cultured with M2-polarized macrophages or in M2 CM ([83]Supplemental
   Figure 2, J and K). A similar trend was seen in GBM-6 spheroids,
   although it was not statistically significant ([84]Supplemental Figure
   2, L and M).

   Using GSC-268 spheroids and varied concentrations of M2 CM, we found
   that M2 CM stimulates invasion in a dose-dependent fashion
   ([85]Supplemental Figure 2N). To test whether the increased invasion
   was related to proliferation, we quantified the percentage of KI-67^+
   cells. There was no difference in the percentage of KI-67^+ cells
   between GSCs cultured in control media and M2 CM ([86]Supplemental
   Figure 2, O and P). To further investigate the influence of M2 CM on
   phenotypes beyond invasion, we tested whether M2 CM impacted stemness.
   Immunostaining showed decreased expression of SOX2, a common stemness
   marker, in M2 CM–treated GSC 268s compared with controls
   ([87]Supplemental Figure 2, Q and R). We performed a sphere formation
   assay and found reduced sphere size and circularity in M2 CM–treated
   GSC-268s ([88]Supplemental Figure 2, S–U).

M2-polarized macrophage–secreted factors shift GSC transcriptional profiles
toward invasion- and mesenchymal-associated signatures.

   To investigate how M2 CM alters the transcriptional profile of GSCs, we
   performed bulk RNA-seq on GSCs in hydrogels with or without M2 CM. We
   utilized our previously developed microchannel invasion platform
   consisting of a cylindrical cell reservoir channel embedded within a 3D
   hydrogel that enables the physical separation of highly invasive and
   noninvasive (core) cell fractions from a single device ([89]Figure 2A)
   ([90]32, [91]36). As expected, GSC-20s in M2 CM–treated devices were
   more invasive than GSC-20s in control devices ([92]Figure 2, B and C).
   Following the invasion assay, we microdissected the M2 CM–treated
   devices to isolate the highly invasive GSCs. We performed differential
   gene expression analysis between cells isolated from control devices,
   invasive cells isolated from M2 CM–treated devices, and core cells
   isolated from M2 CM–treated devices. The invasive cell populations
   isolated from devices cultured in GSC maintenance media (control
   devices) did not contain enough RNA to submit for bulk RNA-seq
   independently and, therefore, were pooled with the core cell fractions
   from the same devices for transcriptomic analysis. We found invasive
   and core populations separated along the first principal component
   (PC1), indicating maximum variance along the invasive phenotype. The
   second principal component (PC2) generally separated GSCs based on
   media conditions (M2 CM vs. control media) ([93]Supplemental Figure
   3A). We identified approximately 1,000 differentially expressed genes
   (DEGs) between each sample comparison illustrated in the heatmap and
   volcano plots ([94]Supplemental Figure 3, B–F, and [95]Supplemental
   Dataset 1).

Figure 2. GSC transcriptional changes induced by M2 CM–mediated invasion.

   [96]Figure 2
   [97]Open in a new tab

   (A) Schematic of HA-based invasion device and microdissection borders
   (dashed lines). (B and C) GSC-20 invasion assays with M2 CM (n = 9
   devices) showing (B) representative phase images and (C)
   quantification. Scale bar: 400 μm. (D–F) Cellular pathway (MSigDB)
   enrichment analysis of differentially expressed genes. Pathways, odds
   ratio, and statistical significance (*Padj < 0.05) calculated from
   Enrichr between (D) GSC-20 invasive fraction of M2 CM devices and
   entire channel of GSC-20 control devices, (E) GSC-20 core fraction of
   M2 CM devices and entire channel of GSC-20 control devices, and (F)
   GSC-20 invasive fraction of M2 CM devices and GSC-20 core fraction of
   M2 CM devices. (G) Heatmap showing GSC-20 relative gene expression
   levels of mesenchymal (MES+) and non-mesenchymal (MES–) markers
   obtained by bulk RNA-seq of cells isolated from invasion devices.
   Expression levels normalized to GSC-20 devices in control media (entire
   channel). (H and I) Heatmaps showing relative gene expression levels of
   MES+ and MES– markers obtained by qPCR of GSC-11 (H) and GSC-268 (I)
   cells isolated from spheroid invasion assays. Expression levels are
   normalized to GSCs in control media. (J) Relative gene expression
   levels of PDGFRA, SNAI2, and CD44 across GSC lines and culture
   conditions (n = 4 biological replicates). Expression levels for each
   gene are normalized to GSC-11 in control media. Statistical
   significance was analyzed using 1-way ANOVA followed by Šídák’s
   multiple-comparisons test (C) or Tukey’s multiple-comparison test (J).
   *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001.

   To identify pathways associated with the invasive phenotypes in
   response to M2 CM, DEGs were used as inputs for pathway enrichment
   analysis ([98]Figure 2, D–F). M2 CM treatment led to downregulation of
   pathways related to Interferon Gamma Response, Interferon Alpha
   Response, Inflammatory Response, and IL-6/JAK/STAT3 Signaling,
   reflective of an immunosuppressive state marked by decreased CXCL10,
   CXCL11, IFIT2, TLR2, and TLR3. Treatment with M2 CM led to upregulation
   of mesenchymal (MES) pathways, including TGF-β Signaling, Apical
   Junction, and Epithelial-Mesenchymal Transition (EMT) ([99]Figure 2, D
   and E). Invasive GSCs were associated with enrichment pathways related
   to Cholesterol Homeostasis, MYC Targets V1/V2, and Apical Junction
   ([100]Figure 2, D and F).

   Interestingly, MES pathways (TGF-β Signaling, Apical Junction, and EMT)
   were upregulated in both the invasive and core populations of M2
   CM–treated devices, suggesting that M2 CM treatment may induce global
   upregulation of MES-related genes in GSCs. We further examined bulk
   RNA-seq expression levels of MES-associated (MES+) genes across device
   fractions and culture conditions ([101]Figure 2G). We also included
   DEGs associated with less aggressive non-MES subtypes, designated as
   MES–, which have been shown to have more favorable overall patient
   survival compared with the MES subtype ([102]8). Compared with GSCs
   isolated from control devices, GSCs cultured in M2 CM had elevated
   expression of MES+ genes and downregulated expression of MES– genes.
   The gene expression changes were seen in both M2 CM–treated core and
   invasive cells but were larger in the M2 CM–treated invasive cell
   population. Treatment with M2 CM also led to differential expression of
   cadherins, including the characteristic MES-associated downregulation
   of E-cadherin (CDH1) ([103]Supplemental Figure 3G) ([104]42).

   We next explored whether M2 CM induced upregulation of MES genes across
   multiple patient-derived GSCs. GSC-11 spheroids, which do not invade
   when cultured in M2 CM, displayed an upregulation of genes associated
   with the MES+ phenotype (CD44, CEBPB, SERPINE, and SNAI1) and a
   downregulation of MES– genes (OLIG2 and PDGFRA) relative to GSC-11
   spheroids in control media ([105]Figure 2H). GSC-268 spheroids, which
   do invade when cultured in M2 CM, also displayed an upregulation of
   MES+ genes ([106]Figure 2I). To better understand the association of
   MES+ and MES– gene signatures with invasion, we compared gene
   expression levels for PDGFRA (MES–), SNAI2 (MES+), and CD44 (MES+)
   between GSC-11 and GSC-268. The relative expression of MES+ genes was
   generally lower in GSC-11 spheroids than GSC-268 spheroids regardless
   of media condition ([107]Figure 2J). Additionally, GSC-11 spheroids
   exhibited higher gene expression levels of the MES– marker PDGFRA
   relative to GSC-268 spheroids, independent of media condition
   ([108]Figure 2J).

Identification of macrophage-derived proinvasive proteins using mass
spectrometry.

   To directly identify soluble factors secreted by TAMs that may induce
   MES+ signaling and drive GSC invasion, we performed global proteomics
   on macrophage CM. Using size-based exclusion filtration and heat
   treatment of M2 CM, we first determined that the proinvasive factor in
   M2 CM was greater than 10 kDa ([109]Supplemental Figure 4A) and
   heat-sensitive ([110]Supplemental Figure 4B). From these data, we
   hypothesized that the proinvasive factor was a protein and performed
   mass spectrometry to identify secreted proteins that were more abundant
   in M2 CM than M1 CM. Mass spectrometric analysis identified 1,093 total
   proteins, with 207 proteins detected only in M1 CM, 63 proteins
   detected only in M2 CM, and 823 proteins detected in both M1 and M2 CM
   ([111]Figure 3, A and B, [112]Supplemental Figure 4D, and
   [113]Supplemental Dataset 2).

Figure 3. Mass spectrometry–based identification of macrophage-secreted
proteins.

   [114]Figure 3
   [115]Open in a new tab

   (A) Venn diagram illustrating proteins identified by mass spectrometry
   in M1 and M2 CM. (B) Volcano plot of proteins identified by mass
   spectrometry in both M1 and M2 CM. (C) Nanoparticle tracking analysis
   (NTA) of extracellular vesicles (EVs) isolated by ultracentrifugation
   from M2 CM. (D and E) GSC-268 invasion assay with M2 CM, M2 soluble
   proteins (SP), and M2 EVs (n = 30 spheres) showing (D) quantification
   and (E) representative phase images. Scale bar: 200 μm. (F) Volcano
   plot of conventionally secreted proteins identified by mass
   spectrometry in both M1 and M2 CM. Spheroid invasion results were
   pooled across at least 2–3 independent replicates. Statistical
   significance was analyzed using 1-way ANOVA followed by Tukey’s
   multiple-comparison test (D). *P < 0.05; ****P < 0.0001.

   As expected, M1 CM had higher abundance of proinflammatory proteins
   (CXCL8, CCL3, CXCL10, CCL5, and IFI30), while M2 CM had higher
   abundance of antiinflammatory and ECM remodeling proteins (TIMP2,
   BIGH3, ECM1, ADAM10, and CSTHB) ([116]Supplemental Figure 4E). Beyond
   these conventionally secreted proteins, mass spectrometric analysis
   identified various unconventionally secreted proteins such as OTUB1,
   PDIA3, and PGK2 ([117]Figure 3B). We hypothesized that these detected
   proteins could have been secreted through extracellular vesicles (EVs).
   We isolated and purified EVs from M2 CM and performed a tumorsphere
   invasion assay with full M2 CM, M2 EVs, and EV-free M2 CM, referred to
   as M2 soluble protein fraction (M2 SP). GSC-268 spheroids invaded when
   treated with M2 SP and M2 CM, but were not invasive when treated with
   M2 EVs ([118]Figure 3, C–E). A similar trend was observed with GSC-295
   spheroids ([119]Supplemental Figure 4, F and G). From these results, we
   hypothesized that the proinvasive factor was not EV bound and,
   therefore, computationally filtered our list of identified proteins to
   include only conventionally secreted proteins ([120]Figure 3F)
   ([121]43).

GSC-macrophage interaction analysis identifies putative receptor-ligand pairs
driving invasion.

   To identify specific receptor-ligand pairs driving invasion, we created
   a resource of inferred paracrine crosstalk by mapping the expression of
   GSC receptors to that of their cognate ligands detected in macrophage
   CM ([122]Figure 4A). We identified 97 receptor-ligand pairs of which
   the receptor is expressed by GSC-20s and the ligand is present in M2 CM
   ([123]Figure 4B and [124]Supplemental Dataset 3). The 97 pairs included
   65 unique receptors and 22 unique ligands ([125]Figure 4C) and although
   all ligands were detected at higher abundance in M2 CM than M1 CM
   ([126]Figure 4D), their relative expression in M2 CM varied
   ([127]Figure 4E).

Figure 4. Paracrine receptor-ligand interaction analysis of GSCs and M2
macrophages.

   [128]Figure 4
   [129]Open in a new tab

   (A) Schematic of receptor-ligand interaction analysis workflow. (B)
   Quantification of receptor-ligand pair according to ligand expression
   in M2 CM (x axis) and receptor expression in GSCs (y axis). (C) Heatmap
   of receptor expression in GSCs and ligand expression in M2 CM for each
   receptor-ligand pair. (D and E) Bar graphs of ligands detected in M2 CM
   emerging from receptor-ligand analysis graphed to illustrate (D)
   average protein abundance in M2 CM compared to M1 CM and (E) average
   relative protein abundance obtained from mass spectrometry.

Candidate M2 macrophage–secreted factors are abundant in human GBM TME and
preferentially expressed by TAMs.

   We next investigated the relevance of candidate ligands to human GBM
   tumors using a publicly available scRNA-seq dataset ([130]44). We
   focused our analysis on ligands that were predominantly expressed by
   TAM clusters: ECM1, A2M, APOE, GAS6, BIGH3, S100A9, MMP9, GRN, APOC2,
   and LYZ ([131]Figure 5A). To predict whether these ligands were
   specific to microglia or BMD TAMs, we probed a subclustered version of
   the same dataset. Every ligand was identified in at least 1 TAM cluster
   but some ligands were preferentially expressed by either microglia or
   BMD TAMs ([132]Figure 5B). Since most TAMs in GBM are of monocytic
   origin, and CM from polarized human microglia did not drive GSC
   invasion ([133]Supplemental Figure 6, A and B), we prioritized ligands
   predominantly expressed by BMD TAMs (BIGH3, S100A9, and LYZ). We
   validated these findings using a second publicly available human GBM
   scRNA-seq dataset ([134]45) and found that the top 3 DEGs upregulated
   in BMD TAMs were again BIGH3, S100A9, and LYZ ([135]Figure 5C).

Figure 5. Integration of M2 macrophage–secreted ligands with published
transcriptomic datasets.

   [136]Figure 5
   [137]Open in a new tab

   (A and B) Dot plots depicting gene expression profiles of M2
   macrophage–secreted ligands identified by receptor-ligand analysis.
   Plots include (A) gene expression of all predicted M2 macrophage
   ligands across all GBM tumor cell type clusters and (B) gene expression
   of top 10 predicted M2 macrophage ligands across only GBM TAM clusters.
   (C) Differential gene expression analysis between human GBM bone
   marrow–derived TAMs and microglia TAMs. Genes with log[2]FC > 1 are
   expressed higher in microglia TAMs. (D) Relative BIGH3 mRNA expression
   obtained from qPCR across a panel of cell lines (n = 3 biological
   replicates). (E and F) Gene expression data from TCGA database queried
   and downloaded using GlioVis data portal for visualization and analysis
   of brain tumor expression datasets. (E) Simple linear regression plot
   and analysis of CD14 and BIGH3 gene expression in GBM tumors. (F) BIGH3
   mRNA expression (log[2]) across glioma tumor stage. (G–I) Analysis of
   scRNA-seq dataset using CellChat. (G) Interactions between macrophages
   and GBM cells in the core and periphery. (H) BIGH3 expression in
   macrophage populations in the core and periphery. (I) BIGH3 receptor
   expression in GBM core and periphery. Statistical significance was
   analyzed using 1-way ANOVA followed by Tukey’s multiple-comparison test
   (F). ***P < 0.001; ****P < 0.0001; *****P < 0.00001.

   We focused further mechanistic investigation on BIGH3 (which encodes
   the protein BIGH3, also known as TGFBI), as this secreted extracellular
   protein was one of the top ligands emerging from the multiomics
   receptor-ligand analysis. We measured BIGH3 across a panel of human
   cells and found that M2-polarized macrophages had the highest BIGH3
   expression compared with GBM cells and microglia ([138]Figure 5D).

   To determine whether BIGH3 expression varied by glioma grade, we used
   the GlioVis data portal for visualization and analysis of brain tumor
   expression datasets ([139]46). In patients with GBM, we found a
   positive linear correlation between expression levels of BIGH3 and the
   monocytic marker CD14 ([140]Figure 5E). Additionally, BIGH3 expression
   was highest in GBM tumors compared with lower-grade gliomas and
   nontumor regions ([141]Figure 5F).

   To further understand regional variations in BIGH3 in human tumors, we
   probed an additional human scRNA-seq dataset ([142]26). We used the
   CellChat algorithm to investigate interactions between GBM cells and
   macrophages located in the core and periphery. M2 macrophages were
   found to have a greater number of interactions with GBM cells in the
   core when compared with the periphery ([143]Figure 5G). Furthermore, M2
   macrophages both in the core and periphery expressed BIGH3 more highly
   than M1 macrophages in either location ([144]Figure 5H). GBM cells in
   the core were found to express several known integrin receptors for
   BIGH3, including ITGA5, ITGB1, ITGB3, and ITGB5 ([145]Figure 5I).

TAM-secreted BIGH3 promotes GSC invasion.

   We next tested whether recombinant human BIGH3 (rhBIGH3) was sufficient
   to drive GSC invasion. We found that while rhBIGH3 induced GSC-268
   spheroid invasion ([146]Figure 6, A and B), rhBIGH3 interestingly had
   no effect on GSC-11 spheroids, which also did not invade with M2 CM
   ([147]Figure 6, C and D). To test the functional contributions of BIGH3
   in M2 CM–mediated invasion, we performed invasion assays using a
   blocking antibody targeting BIGH3 (anti-BIGH3). GSC-268 spheroids
   cultured in M2 CM and anti-BIGH3 were less invasive than spheroids
   cultured in M2 CM with IgG isotype control antibody ([148]Figure 6, E
   and F).

Figure 6. M2 macrophage–induced GSC invasion is mediated by secreted BIGH3
and is associated with active mTOR pathway.

   [149]Figure 6
   [150]Open in a new tab

   (A and B) GSC-268 invasion assay with 20 ng/mL rhBIGH3 (n = 30 spheres)
   showing (A) quantification and (B) representative phase images. (C and
   D) GSC-11 invasion assay with 20 ng/mL rhBIGH3 (n = 33 spheres) showing
   (C) quantification and (D) representative phase images. (E and F)
   GSC-268 invasion assay with M2 CM neutralized with 10 μg/mL IgG
   (rabbit) or 10 μg/mL anti-BIGH3 (n = 18 spheres) showing (E)
   quantification and (F) representative phase images. (G and H) GSC-268
   invasion assay with 10 μM mTOR activator MHY1845 (n = 24 spheres)
   showing (G) quantification and (H) representative phase images. (I and
   J) GSC-268 invasion assay with 20 ng/mL rhBIGH3 and 10 μM mTOR
   inhibitor temsirolimus (n = 30 spheres) showing (I) quantification and
   (J) representative phase images. (K and L) GSC-268 invasion assay with
   M2 CM and 10 μM mTOR inhibitor temsirolimus (n = 21 spheres) showing
   (K) quantification and (L) representative phase images. Spheroid
   invasion results were pooled across at least 2–3 independent
   replicates. Statistical significance was analyzed using an unpaired,
   2-sided Student’s t test (A, C, and G) or a 1-way ANOVA followed by
   Tukey’s multiple-comparison test (E, J, and K). ****P < 0.0001. Scale
   bars: 200 μm.

   We next explored another top hit from our multiomics analysis: S100
   calcium binding protein A9 (S100A9). GSC-268 spheroids treated with
   recombinant human S100A9 (rhS100A9) exhibited increased invasion
   relative to spheroids cultured in control media and this interaction
   was inhibited using the small molecule S100A9 inhibitor, ABR-238907
   (ABR) ([151]Supplemental Figure 5, C and D). Interestingly, ABR did not
   change the invasive capacity of GSC-268 spheroids cultured in M2 CM, so
   although S100A9 stimulates invasion in our system, it does not appear
   to be a targetable proinvasive secreted factor in M2 CM
   ([152]Supplemental Figure 5, E and F).

   We explored the intracellular signaling pathways activated by BIGH3
   that could be driving invasion. BIGH3 is a secreted ECM protein, and
   its expression is regulated by TGF-β signaling ([153]47). According to
   our bulk RNA-seq analysis, TGF-β signaling and TGFB1 are both
   upregulated in highly invasive GSCs ([154]Supplemental Figure 5G). To
   test the possibility that BIGH3 and M2 CM promote invasion through
   TGF-β signaling, we directly induced TGF-β signaling with recombinant
   human TGF-β1 (rhTGF-β1). The addition of rhTGF-β1 did not change the
   invasive capacity of GSC-268 spheroids, suggesting that the induction
   of TGF-β signaling with rhTGF-β1 is not sufficient to drive GSC
   invasion ([155]Supplemental Figure 5, H and I).

   Another pathway upregulated in highly invasive GSCs was the
   P13K/AKT/mTOR signaling pathway, which is abnormally activated in
   tumors to promote growth, metastasis, and invasion ([156]48). We first
   asked whether mTOR activation was sufficient to drive invasion and
   found that inducing mTOR signaling with an mTOR activator (MHY1845)
   induced GSC-268 spheroid invasion ([157]Figure 6, G and H). We next
   tested whether BIGH3-promoted GSC invasion was mediated by mTOR
   signaling. Indeed, inhibiting mTOR (temsirolimus) decreased
   rhBIGH3-mediated GSC invasion, as GSC-268 spheroids treated with
   rhBIGH3 and temsirolimus were significantly less invasive than
   spheroids treated with rhBIGH3 and vehicle control (DMSO) ([158]Figure
   6, I and J). We then determined what portion of M2-mediated GSC
   invasion was driven by mTOR by testing the ability of temsirolimus to
   block M2 CM–mediated GSC invasion. We found that GSC-268 spheroids
   treated with M2 CM and temsirolimus were significantly less invasive
   than spheroids treated with M2 CM and vehicle control (DMSO)
   ([159]Figure 6, K and L). Inhibition of mTOR signaling through
   temsirolimus blocked BIGH3- and M2 CM–mediated GSC invasion, suggesting
   that in our system, M2 CM and BIGH3 stimulate invasion through the mTOR
   pathway.

Discussion

   Tumor–immune cell interactions within the GBM TME represent a promising
   and largely untapped set of targets for limiting disease progression
   and improving patient outcomes. Progress in identifying these targets
   is limited by a lack of physiologically mimetic culture platforms that
   support both tumor cell invasion and incorporation of TAMs. Our 3D HA
   models address this need by enabling the measurement of GBM invasion in
   response to TAM-secreted factors. Using this model, we showed that
   THP-1–derived M2 CM promotes invasion across multiple patient-derived
   GSCs, and we saw a similar trend using CM from GBM patient-derived
   macrophages. Unbiased multiomics analysis of GBM and TAM populations
   identified the proinvasive TAM-secreted factors BIGH3 and S100A9. BIGH3
   has recently been recognized as a key component of the tumor ECM, with
   both tumor-suppressing and tumor-promoting functions, depending on
   tumor type and stage ([160]49). BIGH3 has also been proposed as a
   potential MES subtype signature gene based in part on its ability to
   promote growth and motility of continuous GBM cell lines ([161]50,
   [162]51). Additionally, macrophage-derived BIGH3 has been shown to
   promote stemness and growth in GSCs ([163]52). Similarly, S100A9 is
   overexpressed in various human cancers and linked to TME
   immunosuppression and therapeutic resistance ([164]53–[165]57). Our
   work provides additional evidence that BIGH3 and S100A9 are active
   drivers of GBM invasion and suggests that they are primarily derived
   from the BMD TAM population.

   Our findings begin to deconstruct complex TAM–GBM cell interactions and
   support ongoing therapeutic efforts to inhibit M2 function or
   repolarize M2 macrophages toward M1-like polarization states
   ([166]58–[167]61). While there have been reports of high M2 TAM
   accumulation in MES tumors, the mechanism behind this correlation
   remains unclear, although it has been proposed that MES tumors attract
   M2 TAMs ([168]8, [169]62). Our work expands on prior studies
   investigating the relationship between TAM abundance and GBM
   heterogeneity by proposing the possibility that M2 TAM–secreted factors
   directly induce MES transition in tumor cells. Several GSC-secreted
   factors, including IL-6, CCL2, VEGFA, and periostin, have been shown to
   recruit and polarize TAMs; however, further research is needed to
   determine to what extent TAM recruitment is influenced by MES
   classification ([170]10). TAMs and GSCs may potentially form a
   bidirectional feedback loop, where M2 TAMs induce a highly invasive MES
   state in GSCs, leading to further M2 polarization of TAMs. This notion
   is further supported by our finding that high expression of MES gene
   signatures correlates with highly invasive behavior, a connection
   previously demonstrated in breast cancer but not fully elucidated in
   brain tumors ([171]63–[172]65).

   This study reveals notable patient-to-patient heterogeneities in
   GSC-TAM interactions. For example, the majority of patient-derived GSC
   lines invaded when stimulated with M2 CM, except for GSC-11. While we
   did not investigate the reasons for this lack of response, prior
   characterization of GSC-11 ([173]40) revealed several differences from
   GSC-20 (highly invasive with M2 CM), including greater hypermethylation
   and substantial enrichment of glioma–CpG island methylator phenotype
   (G-CIMP) signature genes ([174]40). Positive G-CIMP status has
   previously been correlated with lower immune infiltration ([175]66).
   GSC-11 has lower expression of CD44 than GSC-20 ([176]40), which could
   explain differences in invasion given the critical role of CD44 in
   adhesion to and migration through HA ([177]67). An additional example
   of heterogeneity is the differential response of GSC-268 invasion to
   patient TAM CM, which likely reflects the inherent intertumoral
   heterogeneity in GBM. It would be interesting to repeat these studies
   with GSCs and TAMs derived from the same patient.

   In this study, the mTOR pathway emerged as a targetable regulator of
   GSC-TAM interactions and GSC invasion. Although the exact mechanisms
   connecting BIGH3, mTOR signaling, and MES transition remain unclear,
   BIGH3 has several known integrin-binding motifs ([178]68, [179]69) and
   mTOR activation can occur through integrin-dependent mechanisms
   ([180]70, [181]71). Given the importance of mTOR signaling in cell
   growth, proliferation, metabolism, and survival, this pathway is a
   potential therapeutic target, and our data support ongoing exploration
   of mTOR pathway inhibition as combination therapy for patients with GBM
   ([182]48).

   In addition to invasion, TAMs have been shown to support other protumor
   functions, including stemness and regulating immune cell activation.
   Bulk RNA-seq analysis showed downregulation of inflammatory response
   pathways in GSCs exposed to M2 CM, which may indicate TAMs are
   contributing to immunosuppression. Interestingly, we saw that M2 CM
   treatment decreased expression of the stemness marker SOX2 and led to
   reduced sphere area and circularity during sphere formation assays,
   indicating a reduction in stemness. While macrophages have promoted
   stemness in other systems, in this study, M2 CM appeared to support
   differentiation toward a MES subtype characterized by decreased
   cell-cell contacts resulting in smaller spheres with lower circularity
   and increased cell-ECM adhesion, indicated by a highly invasive
   phenotype in HA-based matrices.

   By exploiting engineered 3D TME models to investigate functional
   contributions of the TAM secretome on GBM invasion, we show that the
   effects of TAM secretome on GBM are context specific and vary by
   macrophage ontogeny and polarization state, as well as GBM stemness and
   mesenchymal characterization. BIGH3 and other TAM-secreted factors
   merit further mechanistic and therapeutic study, including the roles
   these factors may play in priming the TME for invasion. It is also
   important to elucidate specific TAM subsets or polarization states
   beyond the traditional M1/M2 model responsible for the secretion of
   proinvasive factors, as well as identify the TME features driving the
   presence of these key TAM polarization states and subtypes. Regional
   variations in TAM-GBM interactions will also be important to understand
   given the microenvironmental differences between the tumor core and
   edge. Finally, understanding how TAM infiltration and function relate
   to GBM transcriptional heterogeneity and tumor recurrence would focus
   efforts to target TAMs in patients who would most benefit.

Methods

Sex as a biological variable.

   Cells used in this study were sourced from both male and female
   patients. Patient sex of all cell lines used is summarized in
   [183]Supplemental Table 1.

Cell lines.

   GBM-43 and GBM-6 (Mayo Clinic) were cultured in Dulbecco’s modified
   Eagle medium (DMEM) (Thermo Fisher Scientific) supplemented with 10%
   (vol/vol) FBS (Corning, MT 35-010-CV), 1% (vol/vol)
   penicillin-streptomycin (Thermo Fisher Scientific), and 1% (vol/vol)
   GlutaMax (Thermo Fisher Scientific, 35-050-061). Bulk GBM cell lines
   (GBM-43, GBM-6) were harvested using 0.25% Trypsin-EDTA (Thermo Fisher
   Scientific, 25200-072) and passaged less than 30 times.

   GSC-11, GSC-268, GSC-28, GSC-262, GSC-20, and GSC-295 lines were
   provided by the University of Texas M.D. Anderson Department of
   Neurosurgery. The GSC lines were propagated as neurospheres in DMEM/F12
   50:50 1× (Corning, 10-090-CV) supplemented with 2% (vol/vol) B-27
   supplement (Gibco), 20 ng/mL EGF (R&D Systems, 236-EG), and 20 ng/mL
   FGF (R&D Systems, 233-FB). All GSCs were harvested using Accutase cell
   detachment solution (Innovative Cell Technologies, 490007-741) and
   passaged less than 20 times.

   THP-1 cells (ATTC, TIB-202) were cultured in RPMI-1640 medium
   (Sigma-Aldrich, R8758-500ML) supplemented with 10% (vol/vol) FBS
   (Corning, MT 35-010-CV), 1% (vol/vol) penicillin-streptomycin (Gibco,
   15140122), 1% (vol/vol) MEM nonessential amino acids (Thermo Fisher
   Scientific, 11140050), 1% (vol/vol) sodium pyruvate (Gibco, 11360070),
   1% (vol/vol) GlutaMax (Thermo Fisher Scientific, 35-050-061), 1%
   (vol/vol) HEPES (1 M, Gibco, 15630080), and 0.05 mM 2-mercaptoethanol
   (Sigma-Aldrich, M3148-25ML). THP-1 cells were cultured at 3 × 10^5
   cells/mL, split when they reached 8 × 10^5 cells/mL, and passaged less
   than 30 times.

   THP-1 cells were differentiated and polarized following previously
   reported protocols ([184]72). In brief, cells were differentiated at a
   concentration of 5 × 10^5 cells/mL with 100 nM phorbol 12-myristate
   13-acetate (PMA; PeproTech, 1652981) in complete RPMI-1640 media for 24
   hours followed by 3 days of resting. Macrophages were polarized to M1
   with 50 ng/mL human IFN-γ (Bio Basic, RC217-17) and 50 ng/mL LPS
   (Sigma-Aldrich, L2630-10MG) for 48 hours. Macrophages were polarized to
   M2 with 50 ng/mL IL-4 (Bio Basic, RC212-15-5) for 48 hours
   ([185]Supplemental Figure 1B). M1/M2 polarization was confirmed by
   measuring mRNA expression of established M1 and M2 markers with qPCR
   ([186]Supplemental Figure 1, C and D), noting morphological changes
   associated with differentiation and polarization ([187]Supplemental
   Figure 1E), and measuring surface M1/M2 marker expression via flow
   cytometry ([188]Supplemental Figure 1F).

   Patient peripheral blood mononuclear cells (PBMCs) were collected from
   multiple patients with GBM (n = 5) at Memorial Sloan Kettering Cancer
   Center (New York) and isolated via magnetic-activated cell sorting for
   CD14^+ cells. CD14^+ cells were plated in RPMI-1640 plus 10% FBS for 24
   hours before differentiating with 50 ng/mL M-CSF for 48 hours. The
   postdifferentiation and polarization protocol mirrored that of the
   THP-1s. Briefly, the differentiated macrophages were left to rest for
   24 hours before treating with polarization stimulus for 48 hours (M1:
   50 ng/mL LPS and 50 ng/mL IFN-γ; M2: 50 ng/mL IL-4). Polarization of
   macrophages was confirmed via qPCR of M1 (TNFA and IL1B) and M2 (CD209)
   markers and imaging of morphological changes ([189]Supplemental Figure
   1, H and I).

   HMC3 cells (ATCC, CRL-3304) were cultured in DMEM/F12 50:50 1×
   (Corning, 10-090-CV) supplemented with 10% (vol/vol) FBS (Corning, MT
   35-010-CV) and 1% (vol/vol) penicillin-streptomycin (Thermo Fisher
   Scientific). Microglia were polarized following the same THP-1 protocol
   described above. Cells were seeded at a density of 3 × 10^4 cells/cm^2,
   split when confluence reached 90%, and passaged less than 20 times.

   All cells were screened for mycoplasma and validated by short tandem
   repeat (STR) analysis every 6 months at the University of California
   Cell Culture Facility. For experiments in hydrogels, 0.5% penicillin
   and streptomycin (Gibco, 15140122) and 0.125 μg/mL Amphotericin B
   solution (Sigma-Aldrich, A2942) was added to cell culture media.
   Additional information on GBM source and classification can be found in
   [190]Supplemental Table 1.

Me-HA synthesis.

   HA hydrogels were synthesized as previously described ([191]39).
   Briefly, methacrylic anhydride (Sigma-Aldrich, 94%, 760-93-0) was used
   to functionalize sodium hyaluronate (Lifecore Biomedical, Research
   Grade, 66–99 kDa, HA60K-5) with methacrylate groups. The extent of
   methacrylation per disaccharide was quantified by ^1H-NMR spectroscopy
   and found to be approximately 85% for materials used in this study. To
   add integrin-adhesive functionality, Me-HA was conjugated via Michael
   addition with the cysteine-containing RGD peptide Ac-GCGYGRGDSPG-NH[2]
   (Anaspec, AS-62349) at a concentration of 0.5 mmol/L.

Hydrogel rheological characterization.

   Hydrogel stiffness was characterized by shear rheology via a Physica
   MCR 301 rheometer (Anton Paar) with an 8-mm parallel plate geometry for
   g = 0.5% and f = 1 Hz. Frequency was controlled to be between 50 and 1
   Hz for the frequency sweep at a constant strain (g = 0.5%), and the
   modulus saturation curve with time was obtained under oscillation with
   constant strain (g = 0.5%) and frequency (f = 1 Hz). The temperature of
   the gel solution was controlled (T = 37°C) with a Peltier element
   (Anton Paar) and the sample remained humidified throughout the
   experiment.

HA hydrogel crosslinking.

   To form hydrogels, 6% (w/w) Me-HA conjugated with 0.5 mM
   integrin-binding peptide (RGD) was crosslinked with a
   protease-cleavable peptide (KKCG-GPQGIWGQ-GCKK, Genscript) in phenol
   red–free and serum-free DMEM (Thermo Fisher Scientific, 21-063-029)
   containing GBM cell spheroids. Unless otherwise stated, all experiments
   in this study utilized a final 1.5% (w/w) Me-HA hydrogel crosslinked
   with 3.064 mM peptide to yield a shear modulus of approximately 200 Pa,
   which is within the range of values typically reported for brain tissue
   ([192]73–[193]75).

Generation of spheroids.

   Tumorspheres were fabricated using Aggrewell 400 24-well plates
   (Stemcell Technologies, 34415). Briefly, 1.2 × 10^5 cells were seeded
   into a single well of the Aggrewell plate to form spheroids consisting
   of 100 cells after a 48-hour incubation at 37°C and 5% CO[2].

3D spheroid invasion assay.

   Spheroids were resuspended in phenol red–free and serum-free DMEM at a
   density of 1.5 spheroids/μL and used as solvent for HA hydrogel
   crosslinking. For direct coculture experiments, THP-1–derived
   macrophages were added to DMEM for a final concentration of 3,500
   cells/μL hydrogel. For fluorescently labeled coculture experiments,
   THP-1s were fluorescently labeled after differentiation and
   polarization with CellTracker dye (Thermo Fisher Scientific, C2925)
   following the manufacturer’s protocol prior to encapsulation with GSC
   spheroids.

CM collection.

   To prepare CM, THP-1s were cultured in DMEM/F12 50:50 1× (Corning,
   10-090-CV) at a density of 1 × 10^6 cells/mL. For
   PBMC-derived-macrophage CM, cells were cultured in Macrophage-SFM media
   (Gibco, 12065074). After 48 hours, CM was collected and centrifuged at
   130g for 5 minutes to remove dead cells and debris and passed through a
   25 mm, 0.45 μm sterile cellulose acetate filter (VWR, 76479-040). CM
   was used immediately or stored at –20°C prior to use. For indirect
   coculture experiments, cell culture supplements (B-27 or FBS) were
   added to CM immediately before use. Multiple freeze/thaw cycles were
   avoided. For invasion assays with patient macrophage CM, invasive area
   was normalized to spheroids treated with CM from unpolarized
   macrophages (control) to determine whether the observed differences in
   GSC invasion were directly due to macrophage polarization states rather
   than heterogeneity among patient samples.

Heat treatment of CM.

   CM was harvested as described above. Samples were placed in a dry heat
   bath at 100°C. Samples were incubated for 1 hour and cooled to room
   temperature before use. If applicable, cell culture supplements (B-27
   or FBS) were added to CM immediately before use.

EV purification.

   We used ultracentrifugation and nanoparticle tracking analysis to
   isolate and characterize M2-polarized macrophage–derived EVs. CM was
   harvested as described above. Microvesicles were removed by
   centrifugation at 5,000g for 15 minutes. Supernatant was collected and
   ultracentrifuged at 140,000g (28,000 rpm) for 1.5 hours in an SW-28
   rotor (Beckman Coulter). The supernatant was collected (soluble protein
   fraction) and the crude EV pellet was resuspended in 1× PBS and
   ultracentrifuged at 160,000g for 2 hours. The supernatant was
   discarded, and the crude EV pellet was resuspended in 0.02-μm-filtered
   PBS.

Nanoparticle tracking analysis.

   EV sizes and quantities were estimated using the NanoSight NS300
   instrument equipped with a 405-nm laser (Malvern Instruments), analyzed
   in the scatter mode. Silica 100-nm microspheres (Polysciences) served
   as a control to check instrument performance. Vesicles collected as
   described above were diluted 1,000-fold with 0.02-μm-filtered PBS. The
   samples were introduced into the chamber automatically, at a constant
   flow rate of 50 for 5 repeats of 60-second captures at camera level 13
   in scatter mode with NanoSight NTA 3.1 software (Malvern Instruments).
   The size was estimated at detection threshold 5 using NanoSight NTA
   3.1, after which “experiment summary” and “particle data” were
   exported. Particle numbers in each size category were calculated from
   the particle data, in which “true” particles with track length greater
   than 3 were pooled, binned, and counted with Excel (Microsoft). M2 CM
   contained EVs at a concentration of 1.67 × 10^8 ± 7.47 × 10^6
   particles/mL with 100–400 nm diameters (mean: 229 ± 89.3 nm, mode:
   178.5 nm) according to nanoparticle tracking analysis of concentrated
   EVs ([194]Figure 3C). For the invasion assay, EVs were reconstituted in
   GSC maintenance media at a concentration of 1.67 × 10^8 particles/mL,
   the mean concentration of EVs in M2 CM.

Invasion device fabrication.

   Invasion devices were fabricated following a modified version of our
   previously published protocol ([195]32). In brief, an acrylic mold was
   made by laser cutting and stacking 1.5-mm-thick acrylic posts and
   acrylic slide (CLAREX Precision Thin Sheet, 1.5 mm, Astra Products) on
   top of a glass microscope slide (Fisherbrand, 12-550-A3). Next, 0.00695
   mm outer diameter cleaning wires (Hamilton, 18302) were inserted into
   the double channel posts to serve as wire and syringe guides during
   hydrogel casting and cell seeding. Polydimethylsiloxane (PDMS) was
   fabricated by mixing a 10:1 mass to mass ratio of Sylgard 184 elastomer
   with the initiator (Dow Corning) and the mixture was pipetted into the
   acrylic mold and wire assembly and cured at 80°C for 2 hours. After
   curing, the wires, acrylic slide, and acrylic posts were carefully
   removed. A razor blade with an acrylic guide was used to cut a 5 mm by
   75 mm rectangle in the center of the PDMS and the acrylic laser cut lid
   was attached to the PDMS and glass slide with epoxy. New 0.00695 mm
   outer diameter cleaning wires were reinserted into the PDMS guides and
   the entire device was UV treated for 10 minutes and stored in a cold
   room prior to use. Laser cutter designs (Adobe Illustrator files) are
   available in [196]Supplemental Figure 6.

   On the day of experiment, devices were brought to room temperature and
   the HA hydrogel solution was cast within the acrylic molds around the
   wire and incubated for 1 hour in a humidified 37°C chamber. After
   crosslinking, devices with hydrogels and wire were submerged in cell
   culture media for at least 10 minutes, before removal of the wire which
   left an open channel. Afterwards, 100,000 single cells were seeded into
   each open channel and the wires were reinserted into the device to plug
   each end of the cell reservoir. The entire device was placed into a
   4-well plate and bathed in 10 mL of medium. Medium was changed every 3
   days.

Invasion quantification.

   For invasion analysis in tumorsphere invasion assays, spheroids were
   imaged every 2 days using a Nikon Eclipse TE2000 microscope with a Plan
   Fluor Ph1 ×10 objective. Images were acquired using NIS-Elements
   Software. For each spheroid, total spheroid area was outlined in ImageJ
   (NIH) and normalized to total day 0 spheroid area. For experiments with
   direct coculture of GBM and THP-1 cells, GSCs were transduced with
   CAG-GFP lentivirus (Cellomics, PLV-10057-50) and selected with 1 μg/mL
   puromycin (Invitrogen, A1113803) prior to experiments.

   For quantification of invasion in devices, cells in devices were imaged
   every 7 days using a Nikon Eclipse TE2000 microscope with a Plan Fluor
   Ph1 ×10 objective. Images were acquired and stitched using NIS-Elements
   Software. For each device, total cell area was outlined in ImageJ and
   normalized to total day 0 cell area.

RNA extraction.

   For RNA extraction of cells on tissue culture plastic, phenol-free
   total RNA was extracted from cells directly in cell culture wells using
   an RNeasy Plus Micro Kit with gDNA eliminator columns (Qiagen, 74034)
   following the manufacturer’s protocol. For RNA extraction of cells in
   invasion devices, devices were carefully disassembled, and if
   applicable, the invasive cells were physically separated from the
   noninvasive “core” cells based on distance from channel using a
   scalpel. The invasive cell populations isolated from devices cultured
   in GSC maintenance media were not large enough to submit for bulk
   RNA-seq independently and were pooled with the core cell fractions from
   the same devices. For RNA extraction of tumorsphere invasion assays,
   hydrogels and cells were placed in Eppendorf tubes. All samples were
   treated with 10,000 U/mL hyaluronidase from bovine testes, type IV-S
   (Sigma-Aldrich, H3884) for 15 minutes with agitation until the hydrogel
   was fully degraded. Afterwards, RNA was extracted using TRIzol Reagent
   (Invitrogen, 15596018) according to the manufacturer’s recommendations.
   In brief, 1 × 10^6 cells were mixed with 100 μL TRIzol Reagent and the
   solution was vortexed for 30 seconds. Chloroform (20 μL; Sigma-Aldrich,
   c2432) was added to the mixture and the solution was placed on ice for
   15 minutes. Samples were centrifuged at 18,000g for 10 minutes and the
   clear RNA layer was carefully transferred to a new Eppendorf tube. Each
   RNA sample was mixed with a 1:1 volume of 2-propanol (Sigma-Aldrich,
   190764-4L) and 0.5 μL glycogen, RNA grade (Thermo Fisher Scientific,
   R0551). After an overnight incubation at –20°C, samples were
   centrifuged at 18,000g for 10 minutes at 4°C and the RNA pellet was
   rinsed twice by aspirating the supernatant, adding 400 μL 75% ethanol,
   and centrifuging at 18,000g for 1 minute. Following the second rinse,
   the supernatant was carefully and completely removed, and RNA pellets
   were incubated at room temperature until visibly dry. RNA was dissolved
   in 30 μL RNase-free DNase-free distilled H[2]O and RNA concentration
   and purity were measured with a Nanodrop.

cDNA synthesis.

   cDNA was synthesized from extracted RNA using an iScript cDNA Synthesis
   Kit (Bio-Rad, 1708891) following the manufacturer’s protocol.

RNA-seq and differential gene expression analysis.

   Isolated RNA was sent to Novogene Corporation Inc. for library
   construction, quality control, and sequencing. In brief, mRNA was
   purified from total RNA using poly-T oligo–attached magnetic beads.
   After fragmentation, the first-strand cDNA was synthesized using random
   hexamer primers followed by the second-strand cDNA synthesis. The
   library was ready after end repair, A-tailing, adapter ligation, size
   selection, amplification, and purification. The library was checked
   with a Qubit and real-time PCR for quantification and bioanalyzer for
   size distribution detection. Quantified libraries were be pooled and
   sequenced on Illumina platforms, according to effective library
   concentration and data amount. Data were filtered to remove low-quality
   reads and sequences containing adapters and DEGs were identified using
   DESeq2 ([197]https://github.com/thelovelab/DESeq2). Statistically
   significant DEGs were chosen based on fold change (FC) using a
   statistical cutoff of |log[2]FC| of 1 or greater and adjusted P value
   of less than 0.05.

Pathway enrichment analysis.

   We used Enrichr ([198]76–[199]78) to perform enrichment analysis of
   gene lists using the Molecular Signatures Database (MSigDB) ([200]79)
   hallmark gene set collection. MSigDB Pathways were ranked by odds
   ratio.

Receptor-ligand interaction analysis.

   To infer receptor-ligand interactions between GSCs and M1/M2 THP-1
   macrophages, we compared our RNA-seq data of GSC-20s and our global
   proteomics data of M1/M2 macrophage CM to a database of 3,631
   receptor-ligand interactions ([201]80, [202]81), which was downloaded
   and analyzed in RStudio (RStudio, Inc., 2022.12.0+353;
   [203]https://dailies.rstudio.com/version/2022.12.0+353/). We first
   filtered the M1/M2 macrophage CM dataset to identify conventionally
   secreted proteins using The Human Protein Atlas database of predicted
   secreted proteins ([204]43). We then annotated cognate pairs
   coexpressed by GSC-20s for which the base mean transcript expression of
   the receptor in the GSC-20 + M2 CM (invasive fraction) sample is 1 or
   greater. This produced 255 receptor-ligand pairs for which the receptor
   is expressed by GSC-20s and the ligand is expressed by M1/M2
   macrophages. To identify the receptor-ligand interactions dependent on
   macrophage polarization state, we grouped our inferred receptor-ligand
   pairs into M2- and M1-specific interactions using FC = (M2 normalized
   intensity)/(M1 normalized intensity). FC ≥ 2 indicated M2-specific
   interactions, 2 < FC > 0.5 indicated pan-macrophage interactions, and
   FC ≤ 0.5 indicated M1-specific interactions. This analysis yielded 100
   M1-specific interactions, 97 M2-specific interactions, and 58
   pan-macrophage interactions ([205]Supplemental Dataset 3).

Acquisition and analysis of scRNA-seq datasets.

   DEGs from scRNA-seq analysis of 9 patient GBMs were downloaded from Cui
   et al. ([206]44), who had obtained 10× Genomics scRNA-seq data from the
   NCBI’s Gene Expression Omnibus (GEO) database under [207]GSE131928
   ([208]82). The sequencing analyses were conducted on samples from 9
   fresh GBM tumors from adult patients in Massachusetts General Hospital.
   Dimensionality reduction, unsupervised clustering, and identification
   of cell types and marker genes in different clusters were performed as
   previously described ([209]44). In brief, single-cell transcriptomes
   for a total of 16,201 cells were retained after initial quality
   controls and were differentiated into 22 different clusters, which were
   further sorted into 3 main cell types by comparing their gene
   expression characteristics using the following marker genes: malignant
   cells (GAP43, GPM6B, SEC61G, and PTN), TAMs (C1QA, C1QB, C1QC, TYROBP,
   and CD68), and T lymphocytes (CD3G, GZMH, IL2RB, PR1F, and ICOS). Pct.1
   represents the percentage of cells where the gene is detected in
   specific cluster and pct.2 represents the percentage of cells on
   average in all the other clusters where the gene is detected.
   Enrichment score was calculated by dividing pct.1 by pct.2. Clusters 2,
   3, 6, 7, 9, and 19 represent TAMs, cluster 17 represents T lymphocytes,
   and clusters 0, 1, 4, 5, 8, 10–16, 18, 20, and 21 represent malignant
   cells. To comprehensively characterize the signatures of TAMs in GBM,
   the cells in the macrophage and microglia groups were subclustered and
   reanalyzed. In total, 5,896 cells were sorted and hierarchically sorted
   into 14 clusters, as described in ref. ([210]44). Clusters 4–7, 10, and
   11 represent resident microglia TAMs and clusters 0–3, 8, 9 and 13
   represent BMD TAMs.

   DEGs from human blood–derived and microglial TAMs were downloaded from
   published data by Müller et al., who performed scRNA-seq on a cohort of
   19 total patients comprising 5,455 TAMs (DESeq-adjusted P < 1 × 10^–3)
   ([211]45). The scRNA-seq datasets were mined for genes encoding
   proteins that were detected by mass spectrometry in M2 CM and
   identified in our GSC-macrophage interaction analysis. Data were
   downloaded and analyzed in RStudio.

   To investigate regional differences in BIGH3 expression and GBM
   receptor expression, we utilized the scRNA-seq dataset from Darmanis et
   al. ([212]26). The R package Seurat was used to further define the
   immune subpopulations and isolate M1 and M2 macrophages using
   previously defined canonical macrophage markers as well as markers that
   were suggestive of M1 and M2 subtypes ([213]83). Once these cells were
   defined from the preexisting immune cluster, this updated RNA-seq
   dataset was used to create a CellChat object, which was analyzed using
   the standard CellChat analysis workflow ([214]84). Briefly, the
   expression of ligand and receptor markers was used to elucidate
   communication between cell clusters to construct a signaling network.
   The analysis was all conducted using R version 4.2.2, Seurat 5.0.3, and
   CellChat 1.6.1.

Acquisition of The Cancer Genome Atlas gene expression.

   The Cancer Genome Atlas (TCGA) data were queried and downloaded using
   the GlioVis ([215]46) data portal for visualization and analysis of
   brain tumor expression datasets. Data were graphed and analyzed in
   GraphPad Prism.

qPCR.

   qPCR was carried out using Applied Biosystems PowerUp SYBR Green Master
   Mix (Thermo Fisher Scientific, A25918) and primers following the
   manufacturer’s recommended protocol in a Bio-Rad CFX Connect Real-Time
   PCR cycler with a 5 μM final forward and reverse primer concentration.
   Ct values were calculated using CFX Maestro software accompanying the
   real-time cycler and relative gene expression was calculated using the
   2^–ΔΔCt method. Each gene expression was internally normalized by the
   expression level of housekeeping gene run on the same qPCR batch.
   Primer sequences were designed using either Primer-BLAST ([216]85) or
   PrimerBank ([217]86). Primer sequences are listed in [218]Supplemental
   Table 2.

Flow cytometry.

   THP-1s were differentiated and polarized as described above at 1 × 10^6
   cells/well in 6-well plates. Cells were lifted with a cell scraper,
   centrifuged at 400g for 4 minutes, and washed with PBS. Cells were
   incubated with Live/Dead Zombie Aqua Dye (BioLegend, 77143) and Fc
   blocker (Human TruStain FcX, BioLegend, 422301) for 15 minutes on ice
   before centrifuging and washing with PBS. Cells were then stained for
   M1 and M2 markers for 45 minutes on ice before 2 additional PBS washes.
   Finally, cells were resuspended in 500 μL of FACS buffer (1% BSA, 0.1%
   sodium azide, in 1× PBS), transferred to flow tubes, and analyzed by
   flow cytometry with a FACSymphony A3 (BD Biosciences). To prepare
   compensation controls, 1 drop of beads (~50 μL) was mixed with 0.5 μL
   of antibody and incubated for 15 minutes before adding 450 μL of PBS.
   Compensation and flow data analysis was performed in FlowJo v10
   (FlowJo). Antibodies and flow reagents are listed in [219]Supplemental
   Table 3.

Immunostaining.

   GSC268s were seeded in HA hydrogels either as single cells or
   spheroids, depending on the experiments. At the end of the experiment,
   gels were washed with PBS and fixed with 4% paraformaldehyde in PBS for
   30 minutes. Next gels were washed with PBS and then permeabilized and
   blocked with 0.3% Triton X-100 and 5% goat serum in PBS for 30 minutes.
   Gels were incubated with the primary antibody in 1% goat serum in PBS
   for 72 hours on a shaker at 4°C. The gels were then washed with PBS and
   then incubated with secondary antibodies and DAPI for an additional 24
   hours at 4°C. Hydrogels were then imaged using confocal microscopy
   (Zeiss LSM 880).

Sphere formation assay.

   GSC268s were seeded at 1 × 10^5 cells/mL as single cells in 24-well
   plates in either control media or M2 CM media. After 3 days, the wells
   were imaged. Sphere area and circularity were measured using ImageJ.

Mass spectrometry sample preparation.

   To prepare CM for mass spectrometric analysis, M1- and M2-polarized
   macrophages were cultured in DMEM/F12 50:50 1× (Corning, 10-090-CV) at
   a density of 1 × 10^6 cells/mL. After 48 hours, CM from M1- and
   M2-polarized macrophages was collected and centrifuged at 130g for 5
   minutes to remove dead cells and debris. The CM was then added to
   Amicon Ultra-4 Centrifugal Filter Unit tubes (Millipore, UFC8010-24)
   and concentrated by centrifugation at 10,000g for 15 minutes, resulting
   in a 20-fold reduction in volume (V[i] = 4.0 mL; V[f] = 0.2 mL).

   In preparation for mass spectrometric analysis, samples underwent a
   trypsin in-solution protein digest followed by sample desalting.
   Peptides were analyzed in biological triplicate on a Thermo Fisher
   Scientific Orbitrap Fusion Lumos Tribid mass spectrometry system
   equipped with an Easy nLC 1200 ultrahigh-pressure liquid chromatography
   system interfaced via a Nanospray Flex nanoelectrospray source.

   Samples were injected on a C18 reverse phase column (25 cm × 75 μm
   packed with ReprosilPur C18 AQ 1.9 μm particles). Peptides were
   separated by an organic gradient from 5% to 30% acetonitrile in 0.02%
   heptafluorobutyric acid over 180 minutes at a flow rate of 300 nL/min.
   Spectra were continuously acquired in a data-dependent manner
   throughout the gradient, acquiring a full scan in the Orbitrap (at
   120,000 resolution with an AGC target of 400,000 and a maximum
   injection time of 50 ms) followed by as many MS/MS scans as could be
   acquired on the most abundant ions in 3 seconds in the dual linear ion
   trap (rapid scan type with an intensity threshold of 5000, HCD
   collision energy of 32%, AGC target of 10,000, maximum injection time
   of 30 ms, and isolation width of 0.7 m/z). Singly and unassigned charge
   states were rejected. Dynamic exclusion was enabled with a repeat count
   of 2, an exclusion duration of 20 seconds, and an exclusion mass width
   of ±10 ppm.

Mass spectrometry analysis.

   Protein identification and quantification were done with Integrated
   Proteomics Pipeline (IP2, Integrated Proteomics Applications, Inc.)
   using ProLuCID/Sequest, DTASelect2 and Census ([220]87–[221]90).
   Identified proteins that were detected in only 1 replicate per sample
   were filtered out from dataset. Protein FC was calculated by dividing
   the average normalized M2 intensity by the average normalized M1
   intensity for each protein.

Quantification and statistics.

   Unless otherwise specified, statistical analyses were performed on
   GraphPad Prism v.9.0.2. Statistical tests used 2-tailed 1-way ANOVA and
   post hoc tests for multiple comparisons, or Student’s t test or
   Wilcoxon’s test for comparison between 2 groups. In the box-and-whisker
   plots, the lower bound of the box indicates the first quartile while
   the upper bound of the box indicates the third quartile. The middle
   line represents the median value. The whiskers represent the minimum
   and maximum values. Error bars represent standard deviation. A P value
   of less than 0.05 was considered significant.

Additional reagents.

   A complete list of reagents and resources utilized in this study can be
   found in [222]Supplemental Tables 3–6.

Data availability.

   All data reported in figures can be found in the [223]Supporting Data
   Values file in the supplemental materials. This study did not generate
   new unique reagents or report original code. The raw and processed bulk
   RNA-seq data of invasive GSCs treated with M2 CM are deposited in the
   NCBI (GEO [224]GSE251777). [225]Supplemental Datasets 1–3 are available
   for download.

Author contributions

   EAA, SK, and MKA conceived the project. EAA wrote the manuscript,
   performed most of the experiments, and prepared the figures. DW
   contributed to the experimental work and manuscript preparation. ZAM,
   KH, and RCO contributed to the experimental work. EAA, DW, ZAM, KH,
   RCO, KKHY, MKA, and SK contributed to editing and revising the
   manuscript.

Supplementary Material

   Supplemental data
   [226]jciinsight-10-181903-s040.pdf^ (3.3MB, pdf)
   Supplemental data set 1
   [227]jciinsight-10-181903-s041.xlsx^ (1.2MB, xlsx)
   Supplemental data set 2
   [228]jciinsight-10-181903-s042.xlsx^ (176.6KB, xlsx)
   Supplemental data set 3
   [229]jciinsight-10-181903-s043.xlsx^ (105.1KB, xlsx)
   Supporting data values
   [230]jciinsight-10-181903-s044.xlsx^ (212.8KB, xlsx)

Acknowledgments