Introduction

Survival for adults with the malignant brain tumor, glioblastoma multiforme (GBM), remains poor despite decades of advancements in surgery, radiation, and chemotherapy. One underexplored strategy for treating these cancers is the targeting of stromal cell types in the tumor microenvironment. In this regard, microglia and macrophages may serve as tractable targets for stroma-directed therapy, as they comprise 30-50% of the cells in both benign and malignant gliomas [1,2]. In previous genomic studies, glioma outcome and progression were shown to correlate with macrophage and microglia gene expression [3], while polymorphisms in the microglial CX3CR1 chemokine receptor locus were associated with improved patient survival [4]. Furthermore, pharmacologic or genetic inhibition of microglial function reduces tumor growth in experimental rodent glioma models [2,5–9].

One of the barriers to developing glioma stromal therapies is the paucity of suitable reagents to characterize the spectrum of macrophage populations in health and disease. Although recent reports have established that the tissue origins for mouse brain (resident) microglia and bone-marrow derived monocytes (BMDM) are distinct [10], mouse and human brain tumors harbor potentially distinct and functionally important subpopulations of infiltrating monocytes and resident microglia. To identify new macrophage markers relevant to high-grade glioma, we sought to discover BMDM- and brain microglia-specific transcripts to enable an analysis of the role of these mononuclear cell populations in GBM.

In this study, we leveraged four converging analysis methods across two complementary platforms to identify a series of differentially-expressed BMDM and brain microglia transcripts. Following validation by real-time quantitative RT-PCR and flow cytometry, we selected two representative differentially-expressed BMDM and microglia surface markers (SELL and F11R) to demonstrate that infiltrating BMDM in induced murine malignant glioma acquire F11R expression, which was verified using allogeneic bone marrow transplantation. To establish the clinical relevance of F11R to human GBM, we show that F11R expression correlates positively with glioma malignancy grade as well as correlates negatively with patient survival independent of GBM molecular subtype.

Materials and Methods

Results

To identify transcripts that potentially distinguish infiltrating monocytes from brain microglia, we isolated and examined gene expression of BMDM (CD11b⁺ CD45^high CD115⁺ Ly6G^- cells) and brainstem microglia (BSM) (CD11b⁺ CD45^low CD115^low Ly6G^- cells) pooled from 6-week-old C57BL/6 male mice (ten mice/set) by FACS. Despite low RNA yields (0.4-100ng of total RNA), overall good quality RNA was obtained (Agilent RNA Integrity Number, RIN=7.6-9.0) [24], and the resulting cDNA yields (4.9-5.8µg) were used to perform parallel Affymetrix Mouse Exon 1.0ST microarray and Illumina RNA-sequencing (Figure S2, Table S3 and S4).

Complementary analysis pipelines Cufflinks and ALEXA-Seq for RNA-Seq [18–20,25] and Aroma and Partek for microarray data [26–28] were employed to analyze each platform. Fold changes and q-values were calculated for each gene, and a mean fold-change rank was generated by merging the four independent analyses. High-confidence gene lists were subsequently generated that included only differentially-expressed genes identified to have q-values <0.05 by at least three of the four analysis methods (Table S5). A global unsupervised analysis, excluding genes not expressed or not variably expressed by Cufflinks analysis, demonstrated clear separation of BMDM from BSM samples (Figure 1A).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Differentially-expressed brainstem microglia (BSM) and bone marrow monocyte (BMDM) transcripts.

(A) A heat map was generated from unsupervised hierarchical clustering analysis that included all genes expressed (FPKM >1 in ≥1/6 samples). (B) Candidate genes selected for further validation for comparison to CX3CR1 and CCR2.

https://doi.org/10.1371/journal.pone.0077571.g001

We next prioritized candidate transcripts based on their expression of protein products potentially amenable to flow cytometry analysis (cell surface markers), and selected ten transcripts of predicted membrane-associated proteins (5 BSM and 5 BMDM; Table 1-2) for real-time quantitative PCR (qPCR) verification (Figure 1B and Figure 2A). We secondarily validated two representative transcripts for BSM (F11R and CD81) and BMDM (SELL and CLEC12A) by flow cytometry (Figure 2B-F). Since CD81 resides inside the cell membrane of unstimulated microglia, and CLEC12A is expressed at relatively low levels in monocytes, we chose F11R and SELL as representative differentially-expressed flow cytometry markers to study monocyte infiltration in murine glioblastoma.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Transcript and protein validation of genes differentially expressed between BMDM and BSM.

(A) BMDM-enriched transcripts, including Sell (p=0.0019), Met (p=0.0074), Cd93 (p=0.0104), Kit (p=0.0377), and Clec12a (p=0.0049), were more highly expressed in independently-generated BMDM (n=5) relative to BSM (n=6) using TaqMan and SYBR Green qPCR. BSM-enriched transcripts, including Mertk (p=0.0001), F11r (p<0.0001), P2ry13 (p=0.0002), Cadm1 (p=0.0002), and Cd81 (p=0.0001), were more highly expressed in BSM. (B) Flow cytometry analysis of BMDM (CD11b⁺ and CD115⁺ cells; R1, left panel) and BSM (CD11b⁺ CD45^low cells; R1, right panel) verified that SELL (C) and CLEC12A (D) were detected on BMDM, and not on BSM, while F11R (E) and CD81 (F) were detected on BSM and not BMDM. *, p<0.05; **, p<0.01; ***, p<0.001.

https://doi.org/10.1371/journal.pone.0077571.g002

Since F11r and Sell are expressed in <2% of the total cells isolated from bone marrow and brain, respectively, we first sought to determine whether F11r and Sell would retain their distinctive expression between infiltrative BMDM and resident microglia populations in the setting of induced murine malignant glioma using Ntv-a Ink4a-Arf-/-;Gli-luc mice injected with RCAS-PDGFB (Figure 3A) [11]. Whereas control naïve mice do not form gliomas nor contain significant numbers of CD11b⁺ CD45^high BMDM cells (R3; <2% cells) (Figure S3, Figure 3B), murine gliomas exhibit increased numbers of lymphocytes (R1; 27% cells) as well as CD11b⁺ CD45^low cells (R2, microglia; 39% cells) and CD11b⁺ CD45^high cells (R3, macrophages; 34% cells). While CD11b⁺ CD45^low microglia were >99% F11r⁺, the population of macrophages was not exclusively Sell⁺. Surprisingly, the majority of these glioma-associated CD11b⁺ CD45^high macrophages were also F11r⁺ (94%, Q1) (Figure 3B). Because fewer than 6% of these glioma-associated F11r⁺ cells were dendritic cells, B cells, or NK cells as determined by flow cytometry using CD3, CD4, CD11c, CD8a, CD19, CD103, and NK1.1 antibodies (data not shown), we hypothesized that Sell⁺ BMDM acquire F11r expression following brain infiltration in GBM.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. High-grade murine gliomas contain F11r⁺ microglia and macrophages.

(A) Ntv-a Ink4a-Arf-/-;Gli-luc mice develop tumors following intracranial RCAS-PDGFB injection (n=4). Sell and F11r expression was examined by flow cytometry in lymphocytes (R1, CD11b^- CD45⁺ cells), microglia (R2, CD11b⁺ CD45^low cells), and macrophages (R3, CD11b⁺ CD45^high cells) within the tumor and in control naïve brains (n=4) in four separate experiments. (B) More F11r⁺ microglia and macrophages were identified in the gliomas relative to the control brains, most notably in the expansion of the R3 population (34% of positively labeled cells), which is typically a very small percentage in control brains (<2%). The majority of the labeled macrophages in the glioma are positive for F11r only (Q1; 94%), with few cells infiltrating the glioma positive for Sell only (Q3; 5%) or double positive for both F11r and Sell (Q2; 1.4%). (C) Bar graphs illustrate the mean (SEM) percentage and SEM for each immune cell population as well as their corresponding F11r and Sell surface expressions. White bars = control, black bars = glioma.

https://doi.org/10.1371/journal.pone.0077571.g003

To demonstrate that Sell⁺ monocytes acquire F11r expression as a consequence of brain infiltration, we employed allogeneic BMT to track infiltrating monocytes and resident microglia by virtue of unique CD45 antigen expression (Figure 4A). In this model, chimeric mice with GVHD have CD45.2-expressing resident microglia and CD45.1-expressing donor BMDM and exhibit brain mononuclear infiltration following DLI, whereas control animals (BMT-only) have negligible monocyte infiltration (Figure 4B). Resident CD45.2 microglia from both control and GVHD mice are F11r⁺ and remain F11r⁺ throughout the course of the disease (3 weeks) without increasing CD45.2 surface expression (Figure S4, Table S6) or expressing CD45.1 (Figure 4C, CD11b⁺ CD45.1^- population below R2). As observed in induced murine GBM, infiltrating CD45.1⁺ BMDM (Figure 4C, Table 3) become F11r⁺ as a function of time after entry into the brain: in the first week post-DLI, the majority of the CD45.1⁺ infiltrating macrophages (R3) express surface Sell (53% F11r⁺ Sell⁺, 25% F11r^- Sell⁺); however, by the third week post-DLI, 90% of the positively-labeled R3 cells express only F11r. No other population of infiltrating cells changed their F11r and Sell expression (data not shown). Collectively, these data strongly suggest that F11r is a marker of brain macrophages in the context of murine GBM, regardless of tissue origin, prompting us to examine its prognostic value in patients with high-grade glioma.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 4. BMDM acquire F11r expression following brain infiltration.

(A) GVHD was induced in recipient BALB/cJ mice following total body irradiation (TBI) and bone marrow transplantation (BMT) from T cell depleted (TCD) B6.SJL-Ptprc^a Pepc^b/BoyJ donors. Following C57BL/6J mouse donor lymphocyte infusions (DLI), immune cell infiltration was assessed by flow cytometry at 1 week, 2 weeks, and 3 weeks post-DLI (n=6 GVHD and n=6 BMT-only control per time point). (B) Control BMT-only mice do not exhibit GVHD and lack substantial macrophage (R3) or lymphocyte (R1, and R4 in Figure S4) infiltration. There is a very small CD11b⁺, CD45.1^high monocyte population that expresses F11r in the control brain (<2%; R3, right panel). (C) Chimeric mice with GVHD have donor monocyte infiltration (CD11b⁺, CD45.1^high cells; R3) with negligible CD11b⁺, CD45.1^low cells (R2). F11r and Sell expression (right column) of positively-labeled infiltrating donor monocytes (R3) over the course of 3 weeks of GVHD demonstrates a shift from F11r⁺ Sell⁺ cells to F11r⁺ only cells. (D) Bar graphs illustrate the mean and SEM for each population of immune system cells over the course of the three weeks following DLI.

https://doi.org/10.1371/journal.pone.0077571.g004

In human gliomas, 30-50% of the cells are CD68⁺ or IBA1⁺ tissue macrophages; however, we previously showed that the percentage of CD68⁺ or IBA1⁺ cells in these human tumors did not correlate with glioma malignancy grade [2]. To determine whether F11R expression was associated with increasing glioma malignancy grade, we first confirmed the differential protein expression of F11R in histological sections of human bone marrow and brain (p=0.0016) (Figure 5A). Second, we quantified the percent of F11R⁺ cells (both resident and infiltrating macrophages) in a series of glioma tissue microarrays containing all glioma malignancy grades. In contrast to our prior findings with CD68 and IBA1, we found a positive correlation between the percent of F11R⁺ cells and high-grade glioma (p<0.0001) (Figure 5B).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 5. F11R expression correlates with GBM malignancy grade and survival.

(A) Immunohistochemistry demonstrates that human bone marrow sections (n=3) have few F11R⁺ cells (arrows), while neurologically-normal post-mortem human frontal cortex brain sections contain numerous F11R⁺ mononuclear cells (n=3) (p= 0.0016). Endothelial cell labeling in the upper left quadrant of the brain section represent a positive control for staining. Scale bar = 50µm. Insets depict representative positively-labeled mononuclear cells. (B) A representative high-grade glioma, GBM, contains many F11R+ cells. Increased percentages of F11R⁺ cells (Kruskal-Wallis test/Dunn's Multiple Comparison Test, p<0.0001) are observed in high-grade glioma (AA, anaplastic astrocytoma, n=23; GBM, n=52) relative to low-grade tumors (PA, pilocytic astrocytoma, n=73) or normal brain (NB, n=23). (C) Kaplan-Meier curves and log rank test demonstrate that increased F11R expression negatively correlated with patient survival (GEO database: GSE16011, n=159, p=0.0037). (D) F11R was highly expressed in GSE16011 GBM samples assigned to Classical and Mesenchymal TCGA subtypes relative to the Proneural subtype (p=3.94E-12).

https://doi.org/10.1371/journal.pone.0077571.g005

Because the percent of F11R⁺ cells was greatest in the high-grade gliomas, we next asked whether F11R expression had prognostic value in predicting patient survival. Using GBM data containing 159 specimens (GSE16011) [23] dichotomized by median F11R expression, high F11R expression was associated with reduced patient survival (HR=1.33, log rank test p=0.0037) (Figure 5C). Previous studies have demonstrated that differences in survival relate to specific GBM molecular subtypes [29], and that expression of macrophage/microglia-related genes are associated with the Mesenchymal subtype [3]. In this regard, we also found that IBA1 (AIF1) and CD68 expression was increased in Mesenchymal subtype tumors and was associated with decreased survival (Table S7 and S8). However, using a multivariate Cox model to account for molecular subtype assignments, survival outcomes using AIF1 and CD68 were strongly influenced by the molecular subtype (continuous expression analysis: Subtype overall p=0.0065 and p= 0.0188, respectively; dichotomized expression analysis: Subtype overall p=0.0072 and p= 0.0128, respectively; Table S9). As a result, CD68 was no longer prognostic for patient survival when we account for these robust subtype contributions (continuous expression analysis: HR=1.16 p=0.1222, dichotomous expression analysis: HR=1.36 p=0.0886). AIF1 retained overall prognostic value (continuous expression analysis: HR=1.32 p=0.0155, dichotomous expression analysis: HR=1.42 p=0.0496); however, it is strongly dependent on the survival difference between the Proneural vs. Mesenchymal subtypes. While F11R was also more highly expressed in the Mesenchymal subtypes relative to the Proneural subtype (p = 5.74E-11) (Figure 5D), F11R expression had the greatest independent prognostic value regardless of molecular subtype (Dichotomized gene expression HR=1.57, p=0.0189) (Table 4). In contrast, the expression levels of two frequently-employed macrophage and microglia markers, CCR2 and CX3CR1, were not predictive of patient survival in GBM (Table S8 and S9). Together, these data establish F11R as a novel monocyte predictor of patient outcome in GBM.

Discussion

In the current study, we applied converging digital genomic platforms and analysis methodologies to study microglia and macrophage infiltration in GBM, yielding several important findings. First, we have created an enabling resource by generating a comprehensive catalog of differentially-expressed monocyte/microglia transcripts that may be used to investigate macrophage populations in the other CNS disease states. In addition to identifying known macrophages and microglia markers, including CCR2, LY6C, SELL, SERPINE2, SPARC, CCL4, CX3CR1, and TREM2 [30–32], we have also discovered and validated a number of novel differentially-expressed microglia and monocyte transcripts (P2RY13, CADM1, MET, CD93, and KIT). Importantly, we identified several new markers amenable to flow cytometry analysis not previously reported to be differentially expressed between microglia and BMDM (F11R, CD81, and CLEC12A).

Second, we leveraged two representative transcripts to demonstrate that brain macrophages in murine GBM express F11R regardless of tissue origin (bone marrow versus brain). Specifically, CD11b⁺ CD45^low brain macrophages (microglia) and CD11b⁺ CD45^high infiltrating macrophages (macrophages) express F11R, rather than the SELL BMDM marker. Furthermore, using allogeneic bone marrow chimeras, we establish that BMDM entering the brain in the setting of GVHD convert from a Sell⁺ macrophage population to an F11r⁺ macrophage population as a function of time following CNS infiltration. While it is possible that Sell⁺ macrophages are eliminated and a rare population of F11r⁺ macrophages preferentially expanded, we favor a model in which F11R conversion identifies a subset of GBM-associated macrophages. In either case, the local brain environment is critical for dictating the profile change of infiltrating monocytes. The importance of monocyte F11R expression to glioma biology is further supported by the finding that more established markers of monocytes (e.g., IBA1, CD68), infiltrating macrophages (e.g., CCR2), and microglia (e.g., CX3CR1) do not provide prognostic information for patients with GBM by both immunohistological and gene expression analyses across tumor grade and molecular subtypes.

While it is tempting to postulate that F11R expression confers new biological properties for brain macrophages, the function of F11R in monocytes has not been investigated. Previous studies have shown that F11R is expressed in the basal processes of endothelial tight junctions [33] where it influences epithelial morphology and matrix adhesion [34,35], as well as immune system trafficking [36–38], and may additionally act as a leukocyte adhesion molecule to facilitate leukocyte transendothelial migration under inflammatory conditions [39]. Further studies will be required to determine whether F11R⁺ macrophages in the setting of CNS malignancy represent a unique subset of monocytes that elaborate specific chemokines and cytokines critical for GBM pathogenesis and progression.

Third, we show that one of these differentially-expressed transcripts, F11R, serves as a unique predictive biomarker for malignant glioma. Our previous studies demonstrated that the percentage of CD68⁺ and IBA1⁺ cells were increased in gliomas relative to non-neoplastic brain, with both low-grade and high-grade gliomas harboring equivalent percentages of macrophages. In contrast, we now show that the percentage of F11R⁺ macrophages are increased in high-grade relative to low-grade glioma and that F11R tumor expression is an independent predictor of patient outcome, regardless of GBM molecular subtype. Taken together with reports revealing that microglia are essential drivers of murine glioma formation and maintenance [1,2,5,8,9], along with studies in other cancer types highlighting the predictive value of stromal gene expression patterns in predicting patient outcome [40,41], the identification of F11R as a marker of a subset of brain macrophages may facilitate the detection of critical microenvironmental factors suitable for future stroma-directed glioma therapy. Similarly, potential derivative microglia/macrophage gene signatures may enable clinically-useful deconvolution of existing complex TCGA datasets for prognostic analysis.

Supporting Information

Acknowledgments

We thank the Washington University/Alvin J. Siteman Cancer Center Flow Cytometry Core, Tumor Tissue Repository, Center for Biomedical Informatics, and Multiplex Gene Analysis Genechip Core facilities. We additionally thank Julie Ritchey, Amy Ly, Scott M Gianino, Corina Anastasaki, Sonika Dahiya, and Patrick J Cimino for technical expertise, and Gregory F. Wu, Sean D. McGrath, and Edward Dela Ziga for advice.