Skip to main content
Oncoimmunology logoLink to Oncoimmunology
. 2018 May 24;7(9):e1462430. doi: 10.1080/2162402X.2018.1462430

Subgroup-specific immune and stromal microenvironment in medulloblastoma

Michael Bockmayr a,b,c, Malte Mohme d, Frederick Klauschen b, Beate Winkler a, Jan Budczies b, Stefan Rutkowski a, Ulrich Schüller a,c,e,
PMCID: PMC6140816  PMID: 30228931

ABSTRACT

Knowledge on immune and stromal cells in medulloblastoma microenvironment is still limited as previous work was frequently restricted by low sample size and the lack of molecular subgroup information. We characterized 10 microenvironment cell populations as well as PD-L1 from gene expression in 1422 brain tumors and 763 medulloblastomas. All in all, medulloblastomas showed low expression of immune markers. Still, there were substantial differences with a clustering of medulloblastoma subgroups according to their microenvironment profile. Specifically, SHH medulloblastomas displayed strong signatures of fibroblasts, T cells and macrophages, while markers of cytotoxic lymphocytes were enriched in Group 4 tumors. PD-L1 gene expression appeared to be relatively high in single SHH and WNT cases but was undetectable by immunohistochemistry. In addition, two diverse immuno-stromal patterns were identified, indicating distinct types of local tumor immunosuppression, which were primarily controlled by either macrophage and regulatory T cell-mediated mechanisms or immunosuppressive cytokines and checkpoints, respectively. None of the immune cell signatures had an independent prognostic value in the present dataset after multiple testing correction. These results suggest a mild, but subgroup-specific infiltration of immune cells in medulloblastoma.

KEYWORDS: immune system; gene expression; medulloblastoma; microenvironment; PD-L1, stroma

Introduction

Medulloblastoma is the most frequent malignant brain tumor in children. A comprehensive molecular characterization generating large amounts of genomic, epigenomic, and transcriptomic data has permitted a precise profiling of this entity and the definition of four major molecular subgroups.1-5 These subgroups and additional genetic alterations as well as clinical factors allow for a precise risk stratification of medulloblastoma patients.6

The tumor microenvironment is of major importance for tumor growth and progression as well as treatment response in many cancers.7-9 It has further gained interest in the last years with the availability of immune checkpoint inhibitors, which allow for a therapeutic targeting of the interaction between tumor and immune system.10 Tumor-infiltrating immune cells, principally lymphocytes, have been shown to be of prognostic relevance and predictive for response to chemotherapy in various solid tumors.11,12 In particular, extensive research has been performed on tumor immunology in glioblastoma13,14 and the potential for immunotherapy has been discussed since many years in medulloblastoma.15 However, despite a very deep molecular characterization of this entity, immunotherapeutic approaches are still rare, which is likely due to limited knowledge about the tumor microenvironment and the prognostic relevance of the cells contributing to it.

Using multicolor FACS analysis in a cohort of 6 medulloblastomas, it was shown that medulloblastoma exhibited a different immune microenvironment compared to 3 other pediatric brain tumor entities and the findings were validated by gene expression.16 Natural killer cell (NK cell) therapy was among the first immunotherapeutic approaches that have been proposed for medulloblastoma (see17 for review). Further, Margol et al.18 showed that tumor-associated macrophages were more numerous in SHH medulloblastoma than in Group 3 and Group 4 tumors and proposed macrophages as therapeutic targets.

With the avenue of immune checkpoint inhibitors, the PD-1/PD-L1 axis started being studied in medulloblastoma. Pham et al.19 developed murine SHH and Group 3 models and showed that immune infiltrates were more present in SHH tumors. However, there were more PD-1+ CD8+ T cells in the Group 3 animals. Interestingly, treatment with PD-L1 and CTLA4 inhibitors provided a survival benefit in the Group 3 animals only. In a cohort including 16 tumors, Murata et al. reported that high expression of PD-L1 (seen in 56% of the cases) was associated with low CD8+ T cell infiltration and poor prognosis in medulloblastoma.20 In contrast, Majzner et al.21 (40 cases) and Vermeulen et al.22 (29 cases) did not see PD-L1 expression in any of the studied medulloblastomas. None of the last three studies found subgroup-specific patterns of tumor-infiltrating lymphocytes (TILs).

Most of these results are mainly based on preclinical models or on immunohistological profiling of tumors. This yields to limitations in sample size and reduced statistical power as well as small numbers of markers that can be studied simultaneously. Furthermore, some of the immunohistological results on the expression of PD-L1 in medulloblastoma were contradictory. Associations with survival in larger cohorts have not yet been published.

Recently, large gene expression datasets of medulloblastoma have been released, including one dataset of 763 tumors.4,5 Also, computational methods have been proposed to characterize the cellular composition of the tumor microenvironment from gene expression data. One of the main methods, CIBERSORT,23 uses a deconvolution algorithm that allows for the quantification of 22 immune subpopulations. The algorithm was primarily designed to infer the relative proportion of immune cells and successfully established different immune cell populations as prognostic markers in various cancers.24,25 Furthermore, the recently developed Microenvironment Cell Populations-counter26 (MCPcounter) uses characteristic gene signatures that are specifically associated with 8 immune cell populations (T cells, CD8+ T cells, cytotoxic lymphocytes, B lineage cells, NK cells, monocytic lineage cells, myeloid dendritic cells, neutrophils) and 2 stromal cell populations (fibroblasts and endothelial cells) and allows for their absolute quantification.

Hence, we set out to analyze the tumor microenvironment of medulloblastoma using gene expression data. This allows to overcome sample size limitations and to study several immune cell populations simultaneously. We used MCPcounter as it also quantifies two stromal cell populations and yields more consistent results if there are only few infiltrating immune cells, which is frequently the case for pediatric brain tumors. We first compared medulloblastoma to seven other major brain cancers. Further, we used a large published gene expression dataset to infer subgroup-specific differences and associations with survival.

Results

Medulloblastoma shows low expression of PD-L1 and microenvironment cell signatures in comparison with seven other brain tumors

Expression scores for the eight immune cell populations and the two stromal cell populations characterized by MCPcounter as well as PD-L1 were determined in a dataset of 1422 samples including 421 glioblastomas (GBM), 282 medulloblastomas (MB), 272 ependymomas (EPN), 99 meningiomas (MGM), 159 pilocytic astrocytomas (PA), 61 anaplastic oligodendrogliomas (AO), 69 anaplastic astrocytomas (AA), and 59 atypical teratoid/rhabdoid tumors (AT/RT), (see methods). Although a more precise division into existing molecular tumor subgroups was not possible due the lack of respective information, this should give a rough overview on microenvironment signatures in MB compared to other brain tumors.

The expression scores for all cell populations and PD-L1 were highly significantly different between entities (all p <5.4e-8, see supplement). The overall expression of immune and stromal markers in medulloblastoma was low compared to the other entities (Fig. 1A, B). In the clustering analysis, the expression patterns of medulloblastoma were most similar to the embryonal AT/RT and ependymomas (Fig. 1A). The astrocytic tumors built one cluster and were the only ones to show high expression levels of PD-L1 and markers for cells from the monocytic lineage, e.g. macrophages. There were lower levels of cytotoxic lymphocytes but higher levels of neutrophils in GBM than in the lower grade gliomas. MGM showed a distinct expression pattern to the other tumor entities with large numbers of fibroblasts as well as myeloid dendritic cells and NK cells.

Figure 1.

Figure 1.

Association of the expression scores for 10 microenvironment cell populations and PD-L1 in 8 brain tumors. A: Heatmap representing the median expression score for each cell population and each tumor type. Medulloblastoma has rather low levels of immune and stromal cells and the expression patterns are most similar to AT/RT and ependymoma. B: Each dot corresponds to one tumor in the beeswarm plots. The bands indicate the median, the lower and the upper quartile. Medulloblastoma shows rather low expression scores for most cell types and PD-L1 compared to other brain tumors but the variance is large.

Overall, medulloblastomas had the lowest gene expression level of PD-L1, cytotoxic lymphocytes, and endothelial cells of all studied entities (Fig. 1B). However, there was substantial variation among different medulloblastomas for PD-L1 and the microenvironment cell populations. The expression levels were not uniformly low, indeed, single cases of medulloblastoma showed higher expression than the median of all tumors for all MCPcounter populations and PD-L1.

Hierarchical clustering of all samples revealed a clear-cut aggregation of meningeal tumors, but the remaining entities did not form sharply separated groups (Suppl. Fig. 1). In particular, medulloblastomas did not form a distinct cluster but were mixed with the other entities. Hence, although most medulloblastomas showed lower levels of immune infiltration and stromal cells than the remaining brain tumors some tumors exhibited higher expression of certain cell populations similar to brain tumors from other entities.

Expression of PD-L1 and microenvironment signatures is highly subgroup-specific

Next, we studied the microenvironment populations (MCP) and PD-L1 expression in a gene expression dataset of 763 medulloblastomas with known molecular subgroup (70 WNT, 223 SHH, 144 Group 3 and 326 Group 4).5 The clustering analysis of the 10 MCPs separates most of the SHH and WNT tumors from most of the Group 3 and Group 4 tumors (Fig. 2A). There is one cluster composed mostly of SHH and WNT tumor that shows relatively high PD-L1 expression.

Figure 2.

Figure 2.

Association of the microenvironment populations and PD-L1 with molecular subgroups of medulloblastoma. A: Heatmap representing the expression scores for each individual tumor. Only very few tumors express higher levels of PD-L1, belonging mostly to the WNT and SHH subgroup. B: Beeswarm plots reveal strongly subgroup-specific expression patterns for most of the considered microenvironment cell populations. The bands indicate the median, the lower and the upper quartile. Depicted p-values are computed with the Kruskal-Wallis test.

The expression scores were strongly significantly different between subgroups for PD-L1 and all MCPs with the exception of the B lineage and NK cells (all p <3.85e-10, Fig. 2B). Median PD-L1 expression was highest in WNT tumors (4.06, log2-scale) followed by SHH tumors (3.67) and lowest in Group 4 tumors (3.54, fold change WNT/Group 4: 1.4). The distribution of PD-L1 expression was heavy-tailed and few tumors mostly from the SHH and the WNT subgroup showed comparably high expression (Fig. 2A). Indeed, 20% of WNT, 6.7% of SHH, 2.7% of Group 3, and 0.3% of Group 4 tumors had expression of PD-L1 higher than 4.5, while 5.7% of WNT, 2.7% of SHH, 1.4% of Group 3 and 0.3% of Group 4 tumors had expression higher than 5.0. Still, in a series of 10 WNT and 36 SHH medulloblastomas we detected small islets of perivascular PD-L1 positive cells, but none of the samples showed membranous PD-L1 expression in more than 1% of tumors cells by using antibodies against PD-L1 (data not shown).

SHH tumors were associated with the most distinctive pattern of microenvironmental cells (Fig. 2B). In particular, they had larger numbers of T cells, fibroblasts, and cells from the monocytic lineage and lower numbers of neutrophils than the tumors from the other subgroups. Group 3 tumors had low numbers of T cells and macrophages, but the highest number of CD8+ T cells. Group 4 tumors had the lowest number of fibroblasts and largest number of cytotoxic lymphocytes.

Immuno-stromal classification of medulloblastoma

To get a more precise overview of the complex relations between molecular subgroups and immuno-stromal context, we computed a two-dimensional embedding of 98 MCP markers using t-SNE27 yielding an almost complete separation of the molecular subgroups (Fig. 3A, Suppl. Tables, Suppl. Fig. 2). Using unsupervised k-means clustering, we identified 5 major clusters with distinct immuno-stromal microenvironment (Fig. 3B). Furthermore, four previously published signatures28 were used to identify immunosuppressive mechanisms (see methods and supplement). Individual clustering of distinct signature sets revealed two major immuno-stromal patterns (Fig. 3C). Type I pattern represented a primarily monocytic/macrophage supported tumor environment, with a major fibroblastic stromal component and infiltration of regulatory T cells. Type II pattern on the other hand was defined by a stronger infiltration of cytotoxic CD8+ T cells and neutrophils with a simultaneous activation of immunosuppressive cytokine and checkpoint pathways, which include CD274 (PD-L1), CTLA4 and TGFB1. Endothelial-, B-, NK- and myeloid dendritic cells did not group into these two patterns (Fig. 3C, lower panel). The type I pattern, found in cluster 2 and 4 (Fig. 3B), can be assigned to SHH-, WNT- and approximately one quarter of Group 4 medulloblastoma. Type II pattern was representative for Group 3 and the remaining Group 4 tumors. Interestingly, the unsupervised k-means clustering demonstrates that the two distinct immuno-stromal patterns separate Group 4 tumors into two subgroups (cluster 1 and 4), indicating two potentially diverse mechanisms of immunosuppression within the microenvironment of Group 4 medulloblastoma.

Figure 3.

Figure 3.

Immuno-stromal classification of medulloblastoma. A: Two-dimensional embedding using t-SNE of medulloblastomas according to the expression of 98 MCP marker genes reveals subgroup-specific clustering. B: Definition of five distinct immuno-stromal subgroups using k-means clustering. C: Expression values of the 10 MCPcounter populations and 4 immunosuppressive signatures.28 There are two major patterns (type I + II) potentially associated with immunosuppression in medulloblastoma. Endothelial-, B-, NK- and myeloid dendritic cells did not group into these two patterns.

Associations of microenvironment signatures with survival

The dataset was split into subgroups prior to survival analysis since molecular subgroups are associated with a significantly different prognosis. The overall association of immune cells with survival was only weak. There were no significant associations with survival in WNT tumors, which might be explained by the good prognosis associated with a low number of events and low statistical power (Fig. 4). The expression score for fibroblasts in SHH tumors in univariate analysis was the only significant finding after multiple testing correction using the Benjamini-Hochberg (BH) method (p = 0.003, BH-p = 0.033, Fig. 4). However, in multivariate analyses adjusted for age, histology, metastatic status, and gender, it narrowly missed significance (p = 0.078). There were no significant associations between MCP-markers or PD-L1 in Group 3 and Group 4 tumors after multiple testing correction. In univariate analysis of Group 3 and Group 4 tumors, myeloid dendritic cells marker expression was associated with better survival (p = 0.028, BH-p = 0.32 and p = 0.05, BH-p = 0.36). In multivariate analysis, this association also lost significance (p = 0.32 and p = 0.41, respectively). There were no significant associations between survival and the expression of the markers for the monocytic lineage (macrophages) or NK cells in any of the subgroups.

Figure 4.

Figure 4.

Association of the microenvironment populations and PD-L1 with survival in molecular subgroups of medulloblastoma. Hazard ratios (HR) with 95% confidence interval (CI) are shown for univariate survival analysis. *significant after correction for multiple testing (Benjamini-Hochberg method).

Discussion

Medulloblastoma, an embryonal tumor originating from the posterior fossa, is the most common malignant pediatric brain tumor. With multimodal therapeutic concepts including surgery, radiotherapy, and chemotherapy, about two thirds of the patients can be cured. However, the prognosis of very high-risk patients is still poor.6 Further, many patients suffer from important treatment associated neurocognitive sequelae (reviewed in29). Hence, there is a great need for the development of new therapeutic strategies, and immunotherapy, which has been successfully used in the treatment of many adult solid tumors, is a promising treatment option. A precise understanding of the function of the tumor microenvironment appears key to design medulloblastoma specific immunotherapies. However, there is still not much known about the tumor microenvironment of medulloblastoma. Presumably due to the rarity of medulloblastoma compared to major adult cancers, there are no large immunohistochemical studies characterizing the microenvironment of this entity. Previous reports were limited by low sample sizes and/or focused on only one or two cell populations.16,18-22 Hence, to gain a more comprehensive overview, we used a state of the art bioinformatics method to quantify eight immune cell populations and two stromal cell populations as well as PD-L1 in a pancancer dataset of 1422 brain tumors and a medulloblastoma dataset including 763 samples. This allows for the simultaneous characterization of several cell populations and overcomes the described limitations. To our knowledge, this is the largest and most comprehensive study of the microenvironment in medulloblastoma.

The gene signatures proposed by Becht et al. have been carefully selected to be specific for the described immune cell populations and validated in several thousand tumors including a small dataset of medulloblastoma.26 A study of CD8 and Th17 infiltrates in GBM30 showed comparable associations with prognosis of immunohistological data and data based on gene expression signatures in a similar approach to those proposed by Becht et al. Further, intratumoral heterogeneity of gene expression in medulloblastoma is very low,31 which also supports our assumption that transcriptome based analysis is suitable for a global description of the tumor microenvironment. However, it cannot be excluded that tumor cells contribute to the expression score, which might significantly confound the analysis. Indeed, the expression of certain genes might be associated with different cell types in different tissues/organs. Thus, we reviewed all the marker genes manually and excluded genes associated with medulloblastoma tumor biology (see supplement and methods). For example, EOMES is highly expressed in cytotoxic lymphocytes but only weakly in CD4+ T cells,32 making it a suitable marker for this cell population outside of the nervous system. However, it is also associated with embryonal brain development and a specific marker for Group 4 medulloblastoma,33 which would clearly confound the deconvolution approach. This shows that a careful and tissue-specific selection of the markers is key and should also be performed in further studies.

There have been large numbers of immunohistochemical studies of immune infiltrates particularly in glioma and meningioma including T cells, NK cells and macrophages, with partially contradictory results (reviewed in34-36). This makes it difficult to establish a clear reference level for immune infiltration in these entities, however, for most of the microenvironment populations used in this study, the expression levels are compatible with previously published results based on immunohistochemistry. In the pancancer analysis, PD-L1 expression was highest in GBM in line with results based on immunohistology.37 The very distinctive profile of meningiomas with a larger number of fibroblasts and a larger number of NK cells might be linked to the different histological origin of these tumors.

Medulloblastoma had the lowest amount of PD-L1 and cytotoxic lymphocytes of all studied entities and overall a very small amount of infiltrating immune cells. This suggests that the tumor either actively modulates the immune response or simply has little immunogenicity as is suggested by the relatively low mutational burden.38 However, a recent study also demonstrated some form of systemic immune suppression present in medulloblastoma with lower lymphocyte counts and a higher neutrophil-to-lymphocyte count ratio already present at the moment of diagnosis prior to the application of steroids.39

Our study demonstrated highly significant differences in the pattern of immune and stromal cell signatures as well as PD-L1 between the molecular subgroups of medulloblastoma. Margol et al.18 were the first to highlight differences in the microenvironment of medulloblastoma subgroups and showed that there were more macrophages in SHH tumors. They also defined a gene expression signature with 25 tumor genes as well as 6 immune genes able to accurately classify medulloblastoma into subgroups. Here, we showed that these differences apply to all immune and stromal cell populations except NK cells and B cells. Further, our analysis shows that an immune signature alone without genes associated with tumor biology yields a good separation of medulloblastoma subgroups in a clustering analysis (Fig. 3, Suppl. Fig. 2). Recently, it could be shown that WNT medulloblastoma but not the other subgroups lack a functional blood-brain barrier.40 This might explain differences between WNT tumors and the remaining subgroups, but cannot explain the relatively distinct immune profile of SHH tumors.

Since SHH and WNT tumors had the highest expression levels of PD-L1, we also stained WNT and SHH tumors for PD-L1. Although this is the largest series of medulloblastomas stained for PD-L1 currently reported, we did not find membranous PD-L1 expression in more than 1% of tumor cells in any of the studied samples. This confirms previous studies,21,22 which did not see PD-L1 expression in medulloblastoma and contradicts a small study by Murata et al.20 reporting PD-L1 expression in 9/16 samples. Hence, immunohistological staining of PD-L1 cannot be used as a predictive biomarker for checkpoint inhibition. However, results of PD-L1 immunohistochemistry are variable and patients exist without detectable PD-L1 expression in tumor cells showing a substantial response to PD-L1 inhibitory therapy.39 In a murine model of medulloblastoma, disruption of CDK5 was associated with lower PD-L1 expression and tumor rejection mediated by CD4 T cells, indicating biological relevance of the PD-L1 pathway in medulloblastoma.41 High levels of interferon gamma (IFNγ) were present in CDK5 deficient tumors.41 As we previously reported, TH1 predominance with high expression of IFNγ in the peripheral blood was associated with favorable outcome in children with medulloblastoma treated by high-dose chemotherapy followed by autologous stem cell transplantation.42 This further corroborates the relevance of the IFNγ/PD-L1 axis in medulloblastoma although its precise functioning in vivo remains to be elucidated.

SHH tumors exhibited larger numbers of fibroblasts, macrophages and T cells than the other subgroups. The fibroblast score is mostly based on the expression of collagens and other elements from the extracellular matrix, and it cannot be excluded that these molecules are produced by tumor cells. Higher numbers of fibroblasts were associated with significantly better prognosis and might be linked to a better differentiation of these tumors. The larger amount of fibroblasts in SHH tumors may be explained by desmoplastic histology, which is known to be associated with better prognosis43 and the addition of which as a marker reduces the significance of fibroblast expression in multivariate analysis. Nonetheless, the numerical amount of fibroblasts seems to be an important factor.

We identified two distinct immuno-stromal patterns, defined by an either monocytic/macrophage supported pro-tumorigenic microenvironment versus a pattern which suppresses cytotoxic T cell infiltration by immunosuppressive cytokines and checkpoint pathways. Interestingly, as the diverse patterns, with the exception of the divided Group 4 tumors, are accurately distributed to the known medulloblastoma subtypes, a presumable relation to driver pathways has to be considered. Recent studies support such a pathway-specific immunomodulation. For example, an in vivo colitis study showed that a specific activation of the hedgehog pathway resulted in local immunosuppression mediated by stromal and regulatory T cells.44 A similar immunomodulation of the tumor microenvironment was described in WNT/β-catenin pathway activation. Here, β-catenin activation supported tumor-associated macrophage and regulatory T cell survival (summarized in45). Furthermore, segregation of TCGA datasets according to an inflamed T cell signature set showed that WNT/β-catenin pathway activation inversely correlated with T cell infiltration.46 These observations are in line with our finding of a predominantly monocytic/macrophage and Treg supported (type I) immunosuppressive environment within SHH and WNT medulloblastoma. A pathway-specific inhibition in these tumors could therefore result in a pleiotropic effect, simultaneously promoting tumor-specific immune activation. In contrast, according to our data, pattern type II tumors, including most group 3 and 4 medulloblastoma, will presumably benefit to a greater extent from a targeted inhibition of immunosuppressive signaling pathways, which is also supported by preclinical results from murine models.19

Our analysis showed only weak associations between intratumoral immune infiltrates and survival, which all missed significance after multiple testing correction despite a large sample size. This might be explained by the overall very low immune infiltration, with small differences in the absolute levels of immune infiltrating cells. It corroborates findings by Vermeulen et al., who did not see associations between TILs and prognosis.22 In contrast, peripheral levels of TH1 cells were associated with better prognosis already in a small cohort of medulloblastoma.42 Hence, a simultaneous investigation of peripheral and intratumoral immune cells seems to be of key relevance in further studies of medulloblastoma tumor immunology.

In summary, we report that medulloblastoma has the lowest level of PD-L1 and cytotoxic lymphocytes of eight studied brain tumors. There are substantial differences in the microenvironment cell populations between molecular subgroups, which are likely to be associated with different interactions between the tumor and the immune system. In particular, two immuno-stromal patterns (macrophage and regulatory T cell-mediated mechanisms vs. immunosuppressive cytokines and checkpoints) were identified. Both patterns were closely linked to the molecular subgroups. Upcoming immunotherapeutic treatment strategies will have to consider these distinct microenvironmental phenotypes in order to efficiently promote a tumor-specific immune response in the context of different mechanisms of immune escape in medulloblastoma subgroups.

Material and methods

All data analyses were performed using the statistical programming language R47 including the packages survival, gplots, bee-swarm and Rtsne.48-51

Datasets and preprocessing

Raw gene expression data (.CEL files) were downloaded from Gene Expression Omnibus (GEO).52 For the pancancer analysis, 18 datasets covering 8 different tumors entities measured on the Affymetrix Human Genome U133 Plus 2.0 Array were used (see supplement). To allow for comparisons between different series, fRMA normalization, which includes a batch effect correction, was used as implemented in the R-package fRMA53 using the input vectors from the R-package hgu133plus2frmavecs.54 Duplicated samples (defined by a Pearson correlation > 0.995) were removed. This yielded a final dataset of 1422 unique tumor samples (421 glioblastomas (GBM), 282 medulloblastomas (MB), 272 ependymomas (EPN), 99 meningiomas (MGM), 159 pilocytic astrocytomas (PA), 61 anaplastic oligodendrogliomas (AO), 69 anaplastic astrocytomas (AA) and 59 atypical teratoid/rhabdoid tumors (AT/RT). To exclude major remaining batch effects after preprocessing, we compared samples from entities present in multiple gene expression series. Clustering analysis showed that MCPcounter profiles for the same entity were similar with substantial variance between entities (data not shown).

The data published by Cavalli et al. was used for medulloblastoma specific analyses.5 Raw data were normalized using the RMA method as implemented in the R-package affy55 with the custom chip definition file hugene11sthsensgcdf (v22.0.0).56

Computation of immune and stromal signatures

Published signatures of mRNA markers from 10 cell populations (MCPcounter) were used to estimate the content of T cells, CD8+ T cells, cytotoxic lymphocytes, B lineage cells, NK cells, monocytic lineage cells, myeloid dendritic cells, neutrophils, fibroblasts and endothelial cells in tumor tissue.26 All markers were manually reviewed and those associated with medulloblastoma tumor biology were removed to avoid bias (see supplement). MCPcounter scores were defined as the average of the log2-scaled values of the corresponding probes or genes (see supplement) for the U133 Plus 2.0 Array and the 1.1 ST array respectively. Expression of PD-L1 was defined as the mean of the two corresponding probes (223834_at and 227458_at) for the U133 Plus 2.0 Array. The resulting scores are expressed in arbitrary units. Immunosuppressive signatures were obtained from Doucette et al.28 and computed analogously.

Heatmap, differential expression, k-means clustering, t-SNE and survival analysis

Medulloblastoma data was trimmed at Z-scores ± 4 prior to heatmap analysis. Heatmaps were generated using the average linkage method and Pearson correlations as similarity measure. Significance of differential expression between subgroups of medulloblastoma was assessed using the Kruskal-Wallis test (two-sided).

Immuno-stromal clusters were defined using k-means clustering of the MCPcounter marker genes. The number of clusters (5) was selected using the elbow method. For visualization, t-SNE27,51 was used with perplexity 100.

Survival analysis was performed using the proportional hazards model as implemented by the coxph function from the R package survival. For multivariate analysis, age, metastatic status, histology and gender were included. Samples with missing clinical data were excluded. Expression values for the different signatures were transformed to 3 quantiles (tertiles) and included as numerical values. Multiple testing correction for the cell populations and PD-L1 was performed using the Benjamini-Hochberg (BH) method where appropriated.

Immunohistochemistry

Immunohistochemical stains were performed on formalin-fixed, paraffin-embedded tumor material according to standard protocols on an automated Ventana HX IHC system (Ventana-Roche Medical systems, Tucson, AZ, USA) following the manufacturer's instructions. Diaminobenzidine (Sigma) was used as Chromogen and sections were counterstained with hematoxylin. Antibodies against PD-L1 (E1L3N, Cell Signaling) were used in a dilution of 1:200.

Informed consent was obtained from all patients prior to the analysis.

Supplementary Material

KONI_A_1462430_Supple.zip

Funding Statement

This research was supported by the Fördergemeinschaft Kinderkrebs-Zentrum Hamburg.

Abbreviations

AA

anaplastic astrocytoma

AO

anaplastic oligodendroglioma

AT/RT

atypical teratoid rhabdoid tumor

BH

Benjamini-Hochberg

EPN

ependymoma

GBM

glioblastoma

IFNγ

interferon gamma

MB

medulloblastoma

MCP

microenvironment cell population

MGM

meningioma

PA

pilocytic astrocytoma

PD-1

programmed death 1

PD-L1

programmed death-ligand 1

TIL

tumor-infiltrating lymphocyte

Treg

regulatory T cell

Disclosure of potential conflicts of interest

No potential conflicts of interest were disclosed.

Authors' contributions (CRediT)

Conceptualization: MB and US, Methodology: MB, MM, FK, JB and US, Formal Analysis: MB, Investigation: MB, MM, FK, BW, JB, SR and US, Resources: FK and US, Data Curation: MB, Writing – Original Draft: MB and US, Writing – Review & Editing: MB, MM, FK, BW, JB, SR and US, Visualization: MB, Supervision: US, Funding: US.

References

  • 1.Northcott PA, Jones DTW, Kool M, Robinson GW, Gilbertson RJ, Cho Y-J, Pomeroy SL, Korshunov A, Lichter P, Taylor MD, et al.. Medulloblastomics: the end of the beginning. Nat Rev Cancer. 2012;12:818–34. doi: 10.1038/nrc3410. PMID:23175120. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Taylor MD, Northcott PA, Korshunov A, Remke M, Cho Y-J, Clifford SC, Eberhart CG, Parsons DW, Rutkowski S, Gajjar A, et al.. Molecular subgroups of medulloblastoma: the current consensus. Acta Neuropathol (Berl). 2012;123:465–72. doi: 10.1007/s00401-011-0922-z. PMID:22134537. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Northcott PA, Shih DJH, Peacock J, Garzia L, Morrissy AS, Zichner T, Stütz AM, Korshunov A, Reimand J, Schumacher SE, et al.. Subgroup-specific structural variation across 1,000 medulloblastoma genomes. Nature. 2012;488:49–56. doi: 10.1038/nature11327. PMID:22832581. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Northcott PA, Buchhalter I, Morrissy AS, Hovestadt V, Weischenfeldt J, Ehrenberger T, Gröbner S, Segura-Wang M, Zichner T, Rudneva VA, et al.. The whole-genome landscape of medulloblastoma subtypes. Nature. 2017;547:311–7. doi: 10.1038/nature22973. PMID:28726821. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Cavalli FMG, Remke M, Rampasek L, Peacock J, Shih DJH, Luu B, Garzia L, Torchia J, Nor C, Morrissy AS, et al.. Intertumoral Heterogeneity within Medulloblastoma Subgroups. Cancer Cell. 2017;31:737–754.e6. doi: 10.1016/j.ccell.2017.05.005. PMID:28609654. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6.Ramaswamy V, Remke M, Bouffet E, Bailey S, Clifford SC, Doz F, Kool M, Dufour C, Vassal G, Milde T, et al.. Risk stratification of childhood medulloblastoma in the molecular era: the current consensus. Acta Neuropathol (Berl). 2016;131:821–31. doi: 10.1007/s00401-016-1569-6. PMID:27040285. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Hanahan D, Weinberg RA. Hallmarks of cancer: the next generation. Cell. 2011;144:646–74. doi: 10.1016/j.cell.2011.02.013. PMID:21376230. [DOI] [PubMed] [Google Scholar]
  • 8.Chen F, Zhuang X, Lin L, Yu P, Wang Y, Shi Y, Hu G, Sun Y. New horizons in tumor microenvironment biology: challenges and opportunities. BMC Med. 2015;13:45. doi: 10.1186/s12916-015-0278-7. PMID:25857315. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Quail DF, Joyce JA. The Microenvironmental Landscape of Brain Tumors. Cancer Cell. 2017;31:326–41. doi: 10.1016/j.ccell.2017.02.009. PMID:28292436. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Pardoll DM. The blockade of immune checkpoints in cancer immunotherapy. Nat Rev Cancer. 2012;12:252–64. doi: 10.1038/nrc3239. PMID:22437870. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Fridman WH, Pagès F, Sautès-Fridman C, Galon J. The immune contexture in human tumours: impact on clinical outcome. Nat Rev Cancer. 2012;12:298–306. doi: 10.1038/nrc3245. PMID:22419253. [DOI] [PubMed] [Google Scholar]
  • 12.Gooden MJM, de Bock GH, Leffers N, Daemen T, Nijman HW. The prognostic influence of tumour-infiltrating lymphocytes in cancer: a systematic review with meta-analysis. Br J Cancer. 2011;105:93–103. doi: 10.1038/bjc.2011.189. PMID:21629244. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Boussiotis VA, Charest A. Immunotherapies for malignant glioma. Oncogene. 2017:1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Sampson JH, Maus MV, June CH. Immunotherapy for Brain Tumors. J Clin Oncol Off J Am Soc Clin Oncol. 2017;35:2450–6. [DOI] [PubMed] [Google Scholar]
  • 15.Sonabend AM, Ogden AT, Maier LM, Anderson DE, Canoll P, Bruce JN, Anderson RCE. Medulloblasoma: challenges for effective immunotherapy. J Neurooncol. 2012;108:1–10. doi: 10.1007/s11060-011-0776-1. [DOI] [PubMed] [Google Scholar]
  • 16.Griesinger AM, Birks DK, Donson AM, Amani V, Hoffman LM, Waziri A, Wang M, Handler MH, Foreman NK. Characterization of distinct immunophenotypes across pediatric brain tumor types. J Immunol Baltim Md. 1950 2013;191:4880–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Pérez-Martínez A, Fernández L, Díaz MA. The therapeutic potential of natural killer cells to target medulloblastoma. Expert Rev Anticancer Ther. 2016;16:573–6. doi: 10.1080/14737140.2016.1184978. PMID:27144504. [DOI] [PubMed] [Google Scholar]
  • 18.Margol AS, Robison NJ, Gnanachandran J, Hung LT, Kennedy RJ, Vali M, Dhall G, Finlay JL, Erdreich-Epstein A, Krieger MD, et al.. Tumor-associated macrophages in SHH subgroup of medulloblastomas. Clin Cancer Res Off J Am Assoc Cancer Res. 2015;21:1457–65. doi: 10.1158/1078-0432. PMID:25344580. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Pham CD, Flores C, Yang C, Pinheiro EM, Yearley JH, Sayour EJ, Pei Y, Moore C, McLendon RE, Huang J, et al.. Differential Immune Microenvironments and Response to Immune Checkpoint Blockade among Molecular Subtypes of Murine Medulloblastoma. Clin Cancer Res Off J Am Assoc Cancer Res. 2016;22:582–95. doi: 10.1158/1078-0432.CCR-15-0713. PMID:26405194. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Murata D, Mineharu Y, Arakawa Y, Liu B, Tanji M, Yamaguchi M, Fujimoto K-I, Fukui N, Terada Y, Yokogawa R, et al.. High programmed cell death 1 ligand-1 expression: association with CD8+ T-cell infiltration and poor prognosis in human medulloblastoma. J Neurosurg. 2017:1–7. doi: 10.3171/2016.11.JNS16991. PMID:28474991. [DOI] [PubMed] [Google Scholar]
  • 21.Majzner RG, Simon JS, Grosso JF, Martinez D, Pawel BR, Santi M, Merchant MS, Geoerger B, Hezam I, Marty V, et al.. Assessment of programmed death-ligand 1 expression and tumor-associated immune cells in pediatric cancer tissues. Cancer. 2017;123:3807–15. doi: 10.1002/cncr.30724. PMID:28608950. [DOI] [PubMed] [Google Scholar]
  • 22.Vermeulen JF, Hecke WV, Adriaansen EJM, Jansen MK, Bouma RG, Hidalgo JV, Fisch P, Broekhuizen R, Spliet WGM, Kool M, et al.. Prognostic relevance of tumor-infiltrating lymphocytes and immune checkpoints in pediatric medulloblastoma. OncoImmunology. 2017;0:e1398877. doi: 10.1080/2162402X.2017. PMID:29399402. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Newman AM, Liu CL, Green MR, Gentles AJ, Feng W, Xu Y, Hoang CD, Diehn M, Alizadeh AA. Robust enumeration of cell subsets from tissue expression profiles. Nat Methods. 2015;12:453–7. doi: 10.1038/nmeth.3337. PMID:25822800. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Gentles AJ, Newman AM, Liu CL, Bratman SV, Feng W, Kim D, Nair VS, Xu Y, Khuong A, Hoang CD, et al.. The prognostic landscape of genes and infiltrating immune cells across human cancers. Nat Med. 2015;21:938–45. doi: 10.1038/nm.3909. PMID:26193342. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Ali HR, Chlon L, Pharoah PDP, Markowetz F, Caldas C. Patterns of Immune Infiltration in Breast Cancer and Their Clinical Implications: A Gene-Expression-Based Retrospective Study. PLOS Med. 2016;13:e1002194. doi: 10.1371/journal.pmed.1002194. PMID:27959923. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Becht E, Giraldo NA, Lacroix L, Buttard B, Elarouci N, Petitprez F, Selves J, Laurent-Puig P, Sautès-Fridman C, Fridman WH, et al.. Estimating the population abundance of tissue-infiltrating immune and stromal cell populations using gene expression. Genome Biol. 2016;17:218. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Maaten L van der, Hinton G. Visualizing Data using t-SNE. J Mach Learn Res. 2008;9:2579–605. doi: 10.1118/1.3267037. PMID:20175497. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Doucette T, Rao G, Rao A, Shen L, Aldape K, Wei J, Dziurzynski K, Gilbert M, Heimberger AB. Immune Heterogeneity of Glioblastoma Subtypes: Extrapolation from the Cancer Genome Atlas. Cancer Immunol Res. 2013;1:112–22. doi: 10.1158/2326-6066.CIR-13-0028. PMID:24409449. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Palmer SL, Reddick WE, Gajjar A. Understanding the Cognitive Impact on Children Who are Treated for Medulloblastoma. J Pediatr Psychol. 2007;32:1040–9. doi: 10.1093/jpepsy/jsl056. PMID:17329318. [DOI] [PubMed] [Google Scholar]
  • 30.Madkouri R, Kaderbhai CG, Bertaut A, Truntzer C, Vincent J, Aubriot-Lorton MH, Farah W, Limagne E, Ladoire S, Boidot R, et al.. Immune classifications with cytotoxic CD8+ and Th17 infiltrates are predictors of clinical prognosis in glioblastoma. Oncoimmunology. 2017;6:e1321186. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Morrissy AS, Cavalli FMG, Remke M, Ramaswamy V, Shih DJH, Holgado BL, Farooq H, Donovan LK, Garzia L, Agnihotri S, et al.. Spatial heterogeneity in medulloblastoma. Nat Genet. 2017;49:780–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Zhu Y, Ju S, Chen E, Dai S, Li C, Morel P, Liu L, Zhang X, Lu B. T-bet and Eomesodermin Are Required for T Cell-Mediated Antitumor Immune Responses. J Immunol. 2010;185:3174–83. doi: 10.4049/jimmunol.1000749. PMID:20713880. [DOI] [PubMed] [Google Scholar]
  • 33.Jones DTW, Jäger N, Kool M, Zichner T, Hutter B, Sultan M, Cho Y-J, Pugh TJ, Hovestadt V, Stütz AM, et al.. Dissecting the genomic complexity underlying medulloblastoma. Nature. 2012;488:100–5. doi: 10.1038/nature11284. PMID:22832583. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Bieńkowski M, Preusser M. Prognostic role of tumour-infiltrating inflammatory cells in brain tumours: literature review. Curr Opin Neurol. 2015;28:647–58. doi: 10.1097/WCO.0000000000000251. PMID:26402405. [DOI] [PubMed] [Google Scholar]
  • 35.Domingues P, González-Tablas M, Otero Á, Pascual D, Miranda D, Ruiz L, Sousa P, Ciudad J, Gonçalves JM, Lopes MC, et al.. Tumor infiltrating immune cells in gliomas and meningiomas. Brain Behav Immun. 2016;53:1–15. doi: 10.1016/j.bbi.2015.07.019. PMID:26216710. [DOI] [PubMed] [Google Scholar]
  • 36.Kmiecik J, Zimmer J, Chekenya M. Natural killer cells in intracranial neoplasms: presence and therapeutic efficacy against brain tumours. J Neurooncol. 2014;116:1–9. doi: 10.1007/s11060-013-1265-5. PMID:24085644. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Garber ST, Hashimoto Y, Weathers S-P, Xiu J, Gatalica Z, Verhaak RGW, Zhou S, Fuller GN, Khasraw M, de Groot J, et al.. Immune checkpoint blockade as a potential therapeutic target: surveying CNS malignancies. Neuro-Oncol. 2016;18:1357–66. doi: 10.1093/neuonc/now132. PMID:27370400. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Alexandrov LB, Nik-Zainal S, Wedge DC, Aparicio SAJR, Behjati S, Biankin AV, Bignell GR, Bolli N, Borg A, Børresen-Dale A-L, et al.. Signatures of mutational processes in human cancer. Nature. 2013;500:415. doi: 10.1038/nature12477. PMID:23945592. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39.Patel SP, Kurzrock R. PD-L1 Expression as a Predictive Biomarker in Cancer Immunotherapy. Mol Cancer Ther. 2015;14:847–56. doi: 10.1158/1535-7163.MCT-14-0983. PMID:25695955. [DOI] [PubMed] [Google Scholar]
  • 40.Phoenix TN, Patmore DM, Boop S, Boulos N, Jacus MO, Patel YT, Roussel MF, Finkelstein D, Goumnerova L, Perreault S, et al.. Medulloblastoma Genotype Dictates Blood Brain Barrier Phenotype. Cancer Cell. 2016;29:508–22. doi: 10.1016/j.ccell.2016.03.002. PMID:27050100. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.Dorand RD, Nthale J, Myers JT, Barkauskas DS, Avril S, Chirieleison SM, Pareek TK, Abbott DW, Stearns DS, Letterio JJ, et al.. Cdk5 disruption attenuates tumor PD-L1 expression and promotes antitumor immunity. Science. 2016;353:399–403. doi: 10.1016/j.ccell.2016.03.002. PMID:27050100. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42.Wiegering V, Eyrich M, Rutkowski S, Wölfl M, Schlegel PG, Winkler B. TH1 predominance is associated with improved survival in pediatric medulloblastoma patients. Cancer Immunol Immunother CII. 2011;60:693–703. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43.Rutkowski S, Bode U, Deinlein F, Ottensmeier H, Warmuth-Metz M, Soerensen N, Graf N, Emser A, Pietsch T, Wolff JEA, et al.. Treatment of Early Childhood Medulloblastoma by Postoperative Chemotherapy Alone. N Engl J Med. 2005;352:978–86. doi: 10.1056/NEJMoa042176. PMID:15758008. [DOI] [PubMed] [Google Scholar]
  • 44.Lee JJ, Rothenberg ME, Seeley ES, Zimdahl B, Kawano S, Lu W-J, Shin K, Sakata-Kato T, Chen JK, Diehn M, et al.. Control of inflammation by stromal Hedgehog pathway activation restrains colitis. Proc Natl Acad Sci. 2016;113:E7545–53. doi: 10.1073/pnas.1616447113. PMID:27815529. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 45.Pai SG, Carneiro BA, Mota JM, Costa R, Leite CA, Barroso-Sousa R, Kaplan JB, Chae YK, Giles FJ. Wnt/beta-catenin pathway: modulating anticancer immune response. J Hematol OncolJ Hematol Oncol. 2017;10:101. doi: 10.1186/s13045-017-0471-6. PMID:28476164. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46.Luke JJ, Bao R, Spranger S, Sweis RF, Gajewski T. Correlation of WNT/β-catenin pathway activation with immune exclusion across most human cancers. J Clin Oncol. 2016;34:3004. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47.R Core Team R: A language and environment for statistical computing [Internet]. Vienna, Austria: R Foundation for Statistical Computing; 2016. Available from: https://www.R-project.org/ [Google Scholar]
  • 48.Therneau TM. until 2009) TL (original S->R port and maintainer. survival: Survival Analysis [Internet]. 2017. [cited 2017December9]. Available from: https://cran.r-project.org/web/packages/survival/index.html
  • 49.Warnes GR, Bolker B, Bonebakker L, Gentleman R, Liaw WHA, Lumley T, Maechler M, Magnusson A, Moeller S, Schwartz M, et al.. gplots: Various R Programming Tools for Plotting Data [Internet]. 2016. [cited 2017December9]. Available from: https://cran.r-project.org/web/packages/gplots/index.html
  • 50.Eklund A. beeswarm: The Bee Swarm Plot, an Alternative to Stripchart [Internet]. 2016. [cited 2017December9]. Available from: https://cran.r-project.org/web/packages/beeswarm/index.html
  • 51.Krijthe J. code) L van der M (Author of original C. Rtsne: T-Distributed Stochastic Neighbor Embedding using a Barnes-Hut Implementation [Internet]. 2017. [cited 2018March22]. Available from: https://CRAN.R-project.org/package=Rtsne
  • 52.Edgar R, Domrachev M, Lash AE. Gene Expression Omnibus: NCBI gene expression and hybridization array data repository. Nucleic Acids Res. 2002;30:207–10. PMID:11752295. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 53.McCall MN, Bolstad BM, Irizarry RA. Frozen robust multiarray analysis (fRMA). Biostat Oxf Engl. 2010;11:242–53. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 54.hgu133plus2frmavecs [Internet]. Bioconductor [cited 2017December8]; Available from: http://bioconductor.org/packages/hgu133plus2frmavecs/
  • 55.Gautier L, Cope L, Bolstad BM, Irizarry RA. affy—analysis of Affymetrix GeneChip data at the probe level. Bioinformatics. 2004;20:307–15. [DOI] [PubMed] [Google Scholar]
  • 56.Dai M, Wang P, Boyd AD, Kostov G, Athey B, Jones EG, Bunney WE, Myers RM, Speed TP, Akil H, et al.. Evolving gene/transcript definitions significantly alter the interpretation of GeneChip data. Nucleic Acids Res. 2005;33:e175. doi: 10.1093/nar/gni179. PMID:16284200. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

KONI_A_1462430_Supple.zip

Articles from Oncoimmunology are provided here courtesy of Taylor & Francis

RESOURCES