Characterization of Distinct Immunophenotypes Across Pediatric Brain Tumor Types

Patient cohort

Surgical tumor samples were obtained from 42 patients who presented between 2009 and 2013 for treatment at Children’s Hospital Colorado (Aurora, CO) who were diagnosed with PA (n=7; median age 8 years), EPN (n=19; 3 years), GBM (n=5; 15 years), or MED (n=6; 5 years) according to World Health Organization guidelines. All tumor samples used in this study were obtained at the time of initial resection and prior to therapy, apart from a single EPN sample that was obtained at second look surgery of the initial presentation thus having received chemotherapy. Control NT tissue was obtained from pediatric epilepsy patients (median age 13 years) at the time of surgical intervention. Five epilepsy samples of temporal lobe were selected that had no neoplasm or frank immune cell infiltrate. Patient collection was conducted in compliance with Institutional Review Board regulations (COMIRB 95-500 and 09-0906).

Disaggregation of tissue samples

Samples were collected in Neurobasal A media (Gibco, Grand Island, NY) and immediately disaggregated as described previously (17). Briefly, resected tumor was disaggregated by finely mincing with a razor and further triturated by vigorous pipetting. A single cell suspension was obtained by passing the sample through a 70μm cell strainer that is of sufficiently large pore size to permit passage of all immune and tumor cells but not clumped tumor cells (Becton Dickinson, Franklin Lakes, NJ). Due to differences in the relative cohesiveness of the tumor types in the study, there may be different levels of tumor cell retention in the 70 um filters, resulting in a semi-quantitative measurement of tumor cells by FACS. Heterogenous distribution of infiltrating immune cells may be exist within the tumor margins, but this cannot be addressed by FACS once tissue samples have been dissagregated. Rather, FACS of disaggregated samples provides a net proportion of cellular subpopulations within the margins of each particular specimen. The majority of tumor samples processed measured at least one cm³ in total volume, providing a significant portion of the total tumor mass. Disaggregated cells were viably frozen in standard freezing media containing 10% DMSO and stored in liquid nitrogen for subsequent analysis by FACS. Different immune cell populations may be more or less succeptible to cell death after freeze/thaw, but comparative analysis of data generated from identically processed samples minimizes the effect of this potential deficiency.

FACS Ab panel design

Myeloid cells, predominantly macrophage/microglial cells in the context of the CNS, were distinguished in disaggregated tumor specimens by co-expression of CD45 and CD11b. HLA-DR and Fcγ-R CD64 were used as markers of an activated myeloid cells, CD64 also being a marker of the activated M1 polarized phenotype (18). These markers have been associated with improved outcome in microarray-based studies of EPN and GBM, providing further rationale for inclusion in this study (1, 2). Scavenger receptor CD163 and mannose receptor CD206 were used in the present study as markers of this suppressed inflammatory or alternatively activated M2 myeloid phenotype (19). CD50 (ICAM-3) is expressed by undifferentiated or immature myeloid cells (20, 21).

CD45 and CD3 coexpression was used to distinguish tumor-infiltrating lymphocytes. To reveal the phenotype of T-cells infiltrating pediatric brain tumors, expression of CD4 and CD8 was used to distinguish Th-cells and CTL respectively. Effector memory T-cells were identified by expression of CD45RO, a tumor infiltrating T-cell population that has been associated with a favorable outcome in a number of cancer types (22). Programmed cell death protein 1 (PD-1, CD279) expression was also measured, being a clinically targetable marker of inhibitory T-cell activity that has previously identified as a mediator of immunosuppression in GBM (23-25). Details of Abs used are described in Supplemental Table I.

FACS analysis

Viably frozen disaggregated cells were gently thawed and suspended in PBS supplemented with 10% human serum (Jackson Laboratories, Bar Harbor, ME) as a blocking agent and DAPI (Fisher, Pittsburgh, PA) for live/dead cell identification. After blocking for 15 minutes at 4°C samples were resuspended in FACS buffer (PBS/10% FBS) and multicolor Ab panels or corresponding isotype controls. Samples were stained for 30 minutes at 4°C and then washed with FACS buffer, fixed in 2.5% paraformaldehyde and immediately analyzed by FACS. PD-1 staining was performed on cells that had been incubated for 4 hours with 50ng/ml of PMA and 0.5μg/ml of ionomycin as previously described (26).

FACS was performed using a Beckman Coulter Gallios (Beckman Coulter, Brea, CA). All Abs were titered for optimal Ab fluorescence intensity and voltages were set using optimized volumes. Compensation was calculated using the autosetup-preprogrammed compensation on Gallios and settings were saved and used as starting settings for subsequent experiments.

FACS files were analyzed using FloJo version 9.5.2. (Tree Star Inc., Ashland, OR). All samples were first gated on size and singularity, followed by dead cell exclusion. Gating strategies for myeloid and T-cell populations and phenotype markers are shown in Supplemental Figure 1. All gates were drawn using IgG controls for each sample (Supplemental Figure 2). Population percentages were compiled for each sample to create a data matrix. Comparative and statistical analyses were performed on this matrix using Microsoft Excel two-tailed student’s t-test and GraphPad Prism (version 5) Pearson’s correlation analyses. Principal components analysis (PCA) and unsupervised clustering of flow data were performed using R (http://cran.r-project.org) using publicly available software packages from Bioconductor (http://www.bioconductor.org).

Gene expression analysis

Microarray processing was performed for pediatric brain tumor and NT samples as described previously (27). The gene expression study sample dataset included 15 PA, 46 EPN, 20 GBM, 22 MED and 13 NT brain samples. Tumors were obtained from pediatric patients at initial presentation and with no prior treatment. NT brain was collected from autopsy or epilepsy surgery. Sample collection was conducted in compliance with Institutional Review Board regulations (COMIRB 95-500 and 09-0906). Briefly, RNA was extracted from snap frozen surgical samples, processed and applied to HG-U133 Plus 2 GeneChip microarrays (Affymetrix). Scanned microarray data were background corrected and normalized using the guanine cytosine Robust Multiarray Average (gcRMA) algorithm resulting in log 2 gene expression values (28). Gene expression data corresponding to myeloid and T-cell markers were then extracted for further analysis. The probeset with the highest expression was selected in cases where multiple probesets for the same gene existed. Expression levels of PD-1 were below the threshold for accurate detection (log2 normalized expression <3) in all samples, and were thus excluded from this analysis. CD45R0 is a splice form of CD45, and could therefore not be distinguished in this analysis. Gene expression levels associated with the more prevalent myeloid cells were accordingly higher, allowing all myeloid markers to be assessed by this gene expression analysis. PCA and unsupervised clustering of gene expression data was performed as described above. The microarray data discussed in this publication have been deposited in the National Center for Biotechnology Information Gene Expression Omnibus (GEO) database and are publicly accessible through GEO Series accession number GSE50161 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE50161).

Statistical analyses

Statistical analyses including those described above were performed using Microsoft Excel, Graphpad Prism and Bioconductor.

Extent of myeloid cell infiltration varies widely dependent on pediatric brain tumor type

The average percentage of endogenous myeloid cells in the different tumor tissues varied widely, the most myeloid cell-rich being PA (31.6%) followed by EPN (27.1%), GBM (7.6%), MED (4.1%) and then NT (0.4%)(Figure 1A). The proportion of immune cells in patient tumor specimens was compared to that seen in control NT specimens to estimate the influence of malignancy on extent of myeloid cell infiltration. A large and significantly increased number of infiltrating myeloid cells was observed in both PA (72-fold, p=0.025) and EPN (61-fold, p=9.12×10⁻⁶) compared to NT. Although modest compared to PA and EPN, more infiltrating myeloid cells were also seen in GBM compared to NT (17.2-fold), showing a trend toward significance (p=0.074). MED exhibited the lowest levels in myeloid infiltration compared to NT (4.1-fold, p=0.154).

The functional phenotype of tumor-infiltrating myeloid cells is skewed differently between different pediatric brain tumor types

PA exhibited the most activated myeloid phenotype of pediatric tumor types in the present study. Myeloid activation markers HLA-DR and CD64 were both significantly higher (P<0.05) in PA compared to NT (3.5- and 2.5-fold respectively). Markers for undifferentiated and immune suppressed myeloid cells, CD50 and CD163 respectively, were significantly lower (3.7- and 3.2-fold respectively) in PA than NT. Myeloid suppression marker CD206 was also lower than NT (2.3-fold) showing a trend toward significance (p=0.053).

As observed in PA, EPN significantly over-expressed both myeloid HLA-DR (3.0-fold) and CD64 (2.3-fold) and under-expressed all three non-activated myeloid markers CD50, CD163 and 206 compared to NT (1.72-fold, 2.4-fold and 2.3-fold less respectively; all trending toward significance (p<0.1)).

GBM showed evidence of a suppressive M2 phenotype. Unlike PA and EPN, non-activated myeloid phenotype markers CD50, CD163 and CD206 in GBM were not significantly different from NT. M2 markers CD163 and CD206 were respectively 5.0- and 2.5-fold higher in GBMs than PA, and similarly overexpressed when compared to EPN. Despite the relatively elevated levels of non-activated myeloid markers, the level of activation markers in GBM was comparable to PA and EPN (HLA-DR 3.7-fold, CD64 2.9-fold) when compared to NT.

Compared to other tumor types, including GBM, MED contained a myeloid population that was more clearly skewed toward a suppressive M2 phenotype. MED were distinct from other brain tumors studied in that CD64, indicative of a classically activated M1 phenotype, showed no increase compared to NT. In addition, alternatively activated M2 marker CD206 was significantly higher in MED than both PA and EPN and showed a trend to significance when compared to NT. CD50 and CD163 were not different from NT.

T-cell infiltration is correlated with, but significantly lower, than myeloid cell infiltration across all tumor types

Total T-cell infiltration was determined using combined expression of CD45, CD3 and CD8 or CD4. The average number of infiltrating T-cells correlated strongly with myeloid cell infiltration in corresponding tumor and NT (Pearson R=0.96, p=0.0088). However, T-cells were consistently less common (~10%) than myeloid cells in each tumor type and NT. The average percentage of endogenous T-cells in the different tumor types varied widely but was in all cases higher than NT (0.02%), the most T-cell rich being PA (4.05%) followed by EPN (2.27%), GBM (0.79%) and MED (0.40%).

CD8+ and CD4+ T-cell infiltration varies significantly between tumor types and non-tumor brain

Compared to NT and other tumor types, PA contained on average the highest number of infiltrating CD8 T-cells (3.28%), approximately 300-fold more frequent than NT (Figure 2A). EPN, GBM and MED exhibited progressively lower CD8 T-cell infiltrations than PA, but all were higher than NT, 125-, 46- and 23-fold higher respectively. CD4 T-cell infiltration in EPN (1.40%) was higher than NT (83-fold) and all other tumor types (Figure 2B). PA, GBM and MED exhibited progressively lower CD4 T-cell infiltration than EPN but that was again higher than NT, 73-, 26- and 13-fold respectively, although GBM and MED were not statistically different from NT.

The ratio of CD8 to CD4 T-cells has been identified as a prognostic factor in FACS studies of certain non-CNS tumors (29). In the present study CD8:CD4 T-cell ratios were higher in all tumor types than NT. The average ratio of CD8 to CD4 cells in PA was 3.91, significantly higher (p=0.0011) than the ratio in NT, where CD8 and CD4 T-cells were present in roughly equal numbers (average ratio CD8:CD4 = 0.83). The average ratio of CD8:CD4 ratio in EPN, GBM and MED was progressively lower (3.59, 2.83 and 2.78 respectively). The higher CD8:CD4 ratios in tumor compared to NT brain suggests a preferential recruitment of CD8 T-cells to the tumor microenvironment rather than a non-specific migration of CD8 and CD4 T-cells from adjacent normal brain. Activated/memory T-cell marker CD45RO was significantly higher (p<0.05) in PA and GBM than NT (1.9 and 2.1 fold respectively) (Figure 2C). EPN also demonstrated CD45RO expression on CD8 T-cells that was elevated above NT and this difference showed a trend toward significance (p=0.077). CD8 T-cells CD45RO was not significantly elevated in MED compared to NT. CD4 T-cells in all tumor types exhibited modestly elevated CD45RO compared to NT (Figure 2D).

Evidence for an active phenotype in the majority of pediatric brain tumor-infiltrating T-cells was presented by analysis of PD-1 expression, a marker of T-cell inactivation (Figure 2E,F). CD8 and CD4 T-cells in NT were predominantly expressed PD-1 (72.4% and 81.9% respectively). Compared to NT, all tumor types exhibited significantly (p<0.05) lower PD-1 expression on both CD8 and CD4 T-cells, apart from GBM, which trended toward significance with average expression of PD-1 on CD8 and CD4 T-cells 1.9- and 1.5-fold lower than NT (p<0.1) (Figure 2 E,F). Again PA exhibited the most activated immune cell phenotype exhibiting the lowest PD-1 expression in both CD8 and CD4 T-cells of any tumor analyzed.

Unbiased clustering analyses of immune markers emphasizes distinct tumor type-specific immunophenotypes

The multiple immune markers measured in this study were combined to create a data matrix that was subjected to unbiased clustering analyses. Principle components analysis (PCA) demonstrated that based on these immune markers, partial grouping of samples occurs depending on tissue type (Figure 3A). The MED group overlapped with the NT group, with GBM, EPN and PA groups moving progressively farther away from the NT group. GBM, EPN and PA clusters showed some overlap with each other, demonstrating varying levels of heterogeneity within groups.

Unbiased hierarchical clustering of the 12 myeloid and T-cell immune markers was performed to identify relationships between expression patterns of immune markers across all samples (Figure 3B). This analysis identified two distinct groups which corresponded to markers of activated and suppressed phenotypes respectively and not according to whether markers were myeloid or T-cell specific. This analysis also identified strong correlation between some markers, most notably HLA-DR and CD64, which showed the closest grouping of any marker used in the present study. The data matrix was further examined to directly measure correlations between myeloid markers HLA-DR, CD64, CD50, CD163 and CD206 across all samples. This analysis showed that HLA-DR and CD64 had the strongest correlation of any myeloid marker (Pearson’s R=0.79, p<0.0001), in accord with the clustering analysis. This analysis also demonstrated an inverse correlation of HLA-DR with CD50 (R= −0.33, p=0.032) and a trend to inverse correlation of HLA-DR with CD206 (R= −0.28, p=0.078). CD50 positively correlated with CD163 (R=0.31, p=0.045) and CD206 (R=0.31, p=0.044). The positive and inverse correlations observed between myeloid markers prompted re-analysis of FACS data to identify whether those markers identified distinct sub-populations of cells within the myeloid cell population. This analysis demonstrated that none of the studied markers was expressed exclusively. HLA-DR and CD64 showed the strongest co-expression, consistent with clustering analysis, however CD50, CD163 and CD206 expressing populations each showed expression of HLA-DR and CD64 (data not shown). This is consistent with previous studies that demonstrated overlapping marker expression across polarized myeloid phenotypes (18).

As observed by PCA, clustering analysis of FACS data (Figure 3B) illustrated the varying activation phenotype between sample types, the distinction between PA (high activation) and non-tumor brain (low activation) being particularly strong. Again, EPN, GBM and MED exhibit an immune phenotype that is progressively less distinct from NT, with GBM expressing markers of both activated and non-activated phenotypes.

Gene expression analysis distinguishes pediatric brain tumor-specific immunophenotypes

Gene expression microarray data was mined to determine whether immunophenotypes identified by FACS were reflected by corresponding immune marker gene expression in microarray data expression profiles of a larger cohort of pediatric brain tumor samples.

PCA analysis of immune marker gene expression correlated with results of the same analysis using FACS data, the biggest difference being that NT samples were less distinct from other brain tumor types (Figure 4A). The strongest separation was between MED and PA, as was seen with FACS immune markers. EPN and GBM clusters exhibited a more spread out pattern that overlapped with each other and to some extent with PA, but was centered between PA and MED. Hierarchical clustering demonstrated that activation (HLA-DR and CD64) and suppression markers (CD50, CD163 and CD206) clustered separately into two groups, as was seen with FACS data (Figure 4B). As was seen with clustering analysis of FACS data, gene expression of activation markers between PA and NT was distinct, with EPN, GBM and MED expressing progressively lower levels of activation marker genes. Thus the tumor type specific immunophenotypes observed by FACS analysis were recapitulated by gene expression analysis of a larger cohort of tumor samples.

To determine if gene expression of specific immune markers can predict immune environment within a tumor, myeloid gene expression from the microarray database was correlated with corresponding population percentages obtained by FACS. Linearized gene expression and FACS data was averaged for each tissue type and correlated using Pearson’s correlation. CD11b gene expression (ITGAM; Affymetrix probeset ID 205786_s_at), was positively correlated with FACS data (R=0.83, p=0.044). With respect to the functional markers, significant correlation was observed with immune activation markers but not those associated with immune suppression. CD64 (FCGR1A/FCGR1C; 216950_s_at) had the best correlation (R= 0.91, p=0.057) for gene expression by microarray and expression level by FACS. Although not exclusively expressed by immune cells HLA-DR (210982_s_at) also had a positive statistically significant correlation with FACS data (R=0.84, p=0.037). In summary, myeloid activation marker gene expression microarray data reiterated FACS results, demonstrating that different brain tumor types confer distinct infiltrating myeloid cell phenotypes (Figure 4C). Based on these results, extent of myeloid cell infiltration and activation phenotype can be estimated by expression of these key markers in independent gene expression microarray datasets. No significant correlation between gene expression and FACS data for T-cell markers CD8or CD4 was observed, likely due to the relative rarity of infiltrating T-cells.

The use of control tissue from lateral temporal and frontal lobes the FACS study described above means that there is no location matched control NT for those tumors arising in the cerebellum, namely MED and approximately 50% of PA, due to the rare surgical availability of such tissue. This potential pitfall was partly addressed by estimation of myeloid infiltration and activation between different compartments of normal brain, specifically cerebrum, cerebellum and brainstem/thalamus using samples available in the gene expression microarray database. No significant difference in expression of immune markers CD11b, HLA-DR or CD64 was observed between cerebrum (n=8), cerebellum (n=2) or brainstem/thalamus (n=3)(Supplemental figure 3).