Professional Documents
Culture Documents
Frank 2016
Frank 2016
Human Glioblastoma-associated
Microglia/Monocytes Express a Distinct
RNA Profile Compared to Human Control
and Murine Samples
Frank Szulzewsky,1,2 Sonali Arora,2 Lot de Witte,3 Thomas Ulas,4 Darko Markovic,1,5
Joachim L. Schultze,4,6 Eric C. Holland,2 Michael Synowitz,1,7 Susanne A. Wolf,1 and
Helmut Kettenmann1
Glioblastoma (GBM) is the most aggressive brain tumor in adults. It is strongly infiltrated by microglia and peripheral monocytes
that support tumor growth. In the present study we used RNA sequencing to compare the expression profile of CD11b1 human
glioblastoma-associated microglia/monocytes (hGAMs) to CD11b1 microglia isolated from non-tumor samples. Hierarchical clus-
tering and principal component analysis showed a clear separation of the two sample groups and we identified 334 significantly
regulated genes in hGAMs. In comparison to human control microglia hGAMs upregulated genes associated with mitotic cell
cycle, cell migration, cell adhesion, and extracellular matrix organization. We validated the expression of several genes associated
with extracellular matrix organization in samples of human control microglia, hGAMs, and the hGAMs-depleted fraction via
qPCR. The comparison to murine GAMs (mGAMs) showed that both cell populations share a significant fraction of upregulated
transcripts compared with their respective controls. These genes were mostly related to mitotic cell cycle. However, in contrast to
murine cells, human GAMs did not upregulate genes associated to immune activation. Comparison of human and murine GAMs
expression data to several data sets of in vitro-activated human macrophages and murine microglia showed that, in contrast to
mGAMs, hGAMs share a smaller overlap to these data sets in general and in particular to cells activated by proinflammatory stim-
ulation with LPS 1 INFc or TNFa. Our findings provide new insights into the biology of human glioblastoma-associated microglia/
monocytes and give detailed information about the validity of murine experimental models.
GLIA 2016;64:1416–1437
Key words: glioma, glioblastoma, microglia, macrophage, monocyte, RNA sequencing
Address correspondence to: Cellular Neurosciences, Max-Delbrueck-Center for Molecular Medicine in the Helmholtz Society, Berlin, Germany, Robert Roessle Str.
10 13125, Berlin.
E-mail: kettenmann@mdc-berlin.de
From the 1Department of Cellular Neurosciences, Max-Delbrueck-Center for Molecular Medicine in the Helmholtz Society, Berlin, Germany; 2Department of Human
Biology, Fred Hutchinson Cancer Research Center, Seattle, WA, USA; 3Department of Psychiatry, Brain Center Rudolf Magnus, University Medical Center Utrecht,
Utrecht, the Netherlands; 4Department of Genomics and Immunoregulation, Life and Medical Sciences Institute University of Bonn, Bonn, Germany; 5Department
of Neurosurgery, Helios Clinics, Berlin, Germany; 6Platform for Single Cell Genomics and Epigenomics at the German Center for Neurodegenerative Diseases and
the University of Bonn, Bonn, Germany; 7Department of Neurosurgery, University Medical Center Schleswig-Holstein (UKSH), Kiel, Germany
Michael Synowitz, Susanne A. Wolf and Helmut Kettenmann contributed equally to this work.
Additional Supporting Information may be found in the online version of this article.
1416 V
C 2016 Wiley Periodicals, Inc.
Szulzewsky et al.: Human GAMs RNA Sequencing
et al., 2014; Kim et al., 2015; Meyer et al., 2015; Patel et al., several proinflammatory factors, such as IFNc, TNFa, and
2014; Phillips et al., 2006; Sottoriva et al., 2013). IL1b (Hattermann et al., 2014; Hussain et al., 2006a,b).
Similar to the situation in other solid tumors, tissue- We have previously published a limited transcriptome
resident macrophages—microglia in the brain—and infil- experiment based on microarrays that investigated the expres-
trating monocytes are attracted toward GBM in large num- sion pattern of CD11b1 mGAMs isolated from experimental
bers and can amount up to 30% of the cells in the tumor GL261 mouse glioma compared with naive brain-resident
tissue (Charles et al., 2011; Kerber et al., 2008; Watters microglia and peripheral macrophages (Szulzewsky et al.,
et al., 2005). These glioblastoma-associated microglia/ 2015). We could show that these cells exhibit an activation state
monocytes (GAMs) are exposed to tumor-derived factors that only partially overlaps with the expression pattern of classi-
and are programmed into a tumor specific phenotype (Hu cal pro- or anti-inflammatory activated cultured macrophages.
et al., 2015; Vinnakota et al., 2013). GAMs actively sup- In addition we identified several factors produced by mGAMs
port tumor growth, e.g. by secreting matrix-degrading that might support glioma progression. It is of high importance
enzymes, cytokines, and immunosuppressive factors (Car- to validate rodent models to determine relevant pathways, mol-
valho da Fonseca et al., 2014; Coniglio et al., 2012; Feng ecules and therapeutic targets for potential human treatment
et al., 2015; Hu et al., 2014; Li and Graeber, 2012; Mar- strategies. So far a direct comparison of murine and human
kovic et al., 2009). In addition, recent studies using glioma expression profiles of CD11b1 microglial cells is lacking.
mouse models have shown that GAMs-directed therapeutic In this study, we isolated CD11b1 cells from human
approaches can be beneficial for disease intervention (Kees GBM samples and non-tumor control samples and performed
et al., 2012; Lisi et al., 2014; Markovic et al., 2011; Pyon- a genome-wide RNA sequencing analysis. We compared the
teck et al., 2013; Sarkar et al., 2014; Xu et al., 2014). hGAMs expression profile to our previous mGAMs data set
However, it is unclear if these findings also hold true for and in vitro-activated macrophages and microglia and could
human GAMs (hGAMs) and if approaches that were devel- show that, in contrast to mGAMs, hGAMs did not upregu-
oped in mouse models can be successfully applied for the late genes associated with immune activation.
treatment of human GBMs (Butowski et al., 2015). Thus, a
Materials and Methods
better understanding of hGAMs is crucial for the develop-
Ethics Statement
ment of future therapies.
Freshly resected patient samples were provided by the Department
The bipolar axis of in vitro macrophage activation, rang-
for Neurosurgery Charite University Hospital and the Helios Clinics
ing from proinflammatory activation after stimulation with (both Berlin, Germany). Handling and analysis of these tissues was
LPS or IFNc (previously termed M1 activation) to anti- performed in accordance with the 1964 Helsinki declaration and its
inflammatory activation after stimulation with IL4 (previously later amendments and according to the rules and with the approval
termed M2 activation) has been used to classify GAMs in the of the Ethical Committee and with patient’s written consents (Ethics
past. However, this simplistic model of macrophage activation Committee: “Ethikkommission der Charite—Universit€atsmedizin
has been challenged by many studies and the M1/M2 model Berlin,” application number: EA4/098/11).
of macrophage polarization has been replaced by a multi- Postmortem tissue was obtained from the Netherlands Brain
dimensional model of macrophage activation (Edin et al., Bank (NBB; Amsterdam, the Netherlands, www.brainbank.nl).
Informed consent was obtained by the NBB for brain autopsy and
2012; Ginhoux et al., 2015; Murray et al., 2014; Reinartz
for the use of material and clinical data for research purposes, in
et al., 2014; Xue et al., 2014). To define a glioma-associated
compliance with institutional and national ethical guidelines.
phenotype it seems more promising to compare GAMs with
the phenotypes of a broader activation spectrum than with Cell Isolation
M1 and M2 polarized macrophages only. Several studies have Human tissue samples were dissociated using the Neural Tissue Dis-
investigated the cytokine expression levels of GAMs isolated sociation Kit (Miltenyi Biotec, Bergisch-Gladbach, Germany) accord-
from human and murine samples. Studies on murine GAMs ing to the manufacturer’s instructions. To remove the myelin we
(mGAMs) have shown that these cells produce several anti- followed a protocol published elsewhere (Olah et al., 2012). In brief,
the brain cell suspension was mixed with a 22% Percoll (Geyer, Ren-
inflammatory molecules (e.g., TGFb1, ARG1, and IL10),
ningen, Germany) solution and a layer of cold PBS (Gibco-Invitro-
and molecules supporting tissue remodeling and angiogenesis
gen, Darmstadt, Germany) was added on top. Centrifugation at
(e.g., VEGF, MMP2, MMP9, and MT1-MMP), but in turn 950g with slow acceleration and without breaks created a gradient
also produce pro-inflammatory molecules (e.g., IFNc, TNFa, that separated the cell pellet on the bottom of the tube from the
IL1b, IL-6, and CXCL10) (Gabrusiewicz et al., 2011; Sica myelin which was carefully aspirated. Cells were collected, washed
et al., 2006; Umemura et al., 2008; Ye et al., 2012). In con- once with PBS, and centrifuged again at 300g for 10 min. Subse-
trast, studies on hGAMs could not detect the expression of quently, the cell pellet was resuspended in PBS, containing 0.5%
Primers for target genes were designed to recognize all vali- mortem and hippocampal and epilepsy tissues. The eight
dated mRNA splice variants of the gene, relying on the RefSeq tumor-derived samples cluster together on the left side of the
sequences of the UCSC genome browser (http://genome.ucsc.edu/). plot, whereas the five postmortem nontumor samples cluster
The primer sequences were generated using the Primer-Blast tool together on the right side. The three non-tumor samples iso-
(http://www.ncbi.nlm.nih.gov/tools/primer-blast/). Amplification of lated from two epilepsy patients and one patient suffering
unintended targets was excluded by BLAST search (http://blast.ncbi.
from hippocampal sclerosis cluster between these two groups
nlm.nih.gov/Blast.cgi). Calculation of unfavorable secondary struc-
but align toward the postmortem nontumor samples (Fig.
tures and primer-dimer amplification was done with the IDT oligo
analyzer (http://eu.idtdna.com/analyzer/applications/oligoanalyzer/). 1A, Supporting Information Fig. S2). In addition, unsuper-
ACTB Fwd: GAGCTACGAGCTGCCTGAC; ACTB Rev: vised hierarchical clustering based on the variance of all
GACTCCATGCCCAGGAAGG; ADAMDEC1 Fwd: ATCACT- 11,777 genes showed a clear separation of the eight hGAMs
CAAAAGCCCAGAGAAAG; ADAMDEC1 Rev: TAGGTTCATCA- samples from the eight control samples (Fig. 1B). The cluster-
CATCAAACACAAAG; ADAMTS2 Fwd: ATCACGCCATCTTCCT ing indicated no separation of hGAMs isolated from primary
CACA; ADAMTS2 Rev: GCCACCACAAACGCTGA; and recurrent glioblastoma samples.
AIF1 Fwd: CGAATGCTGGAGAAACTTGGA; AIF1 Rev: We performed a one-way ANOVA to compare the gene
GAAAGTCAGGGTAGCTGAACG; CTSL Fwd: GACATCCC- expression profiles of the eight hGAMs samples, the three
TAAGCAGGAGAAGG; CTSL Rev: AATCCGTAGCCAACCACC microglia samples isolated from epilepsy/hippocampal biop-
AG; sies, and the five microglia samples isolated from postmortem
FN1 Fwd: ACAGAACTATGATGCCGACCA; FN1 Rev:
tissues (Fig. 1C, Supporting Information Fig. S3 and Sup-
CACGACCATTCCCAACACAC;
porting Information dataset S1). Nearly 200 genes were sig-
GFAP Fwd: CGCACGCAGTATGAGGCAA; GFAP Rev:
nificantly upregulated in hGAMs when compared with the
GACTCCAGGTCGCAGGTCAA;
epilepsy and hippocampal control samples, whereas only 12
HARS Fwd: GGTGAAGAATGAGATGGTGGGA; HARS
Rev: GCCTGCTTGTTTTGGGATAGT; genes were downregulated. When compared with the post-
IBSP Fwd: ATCCGAAGAAAATGGAGATGACAG; mortem control samples 384 genes were upregulated and 180
IBSP Rev: GTGTTGTAGCAGAAAGTGTGGTATT; genes were downregulated in hGAMs. However, we found no
RND3 Fwd: GATGCTGTGCTGATTTGCTTTGA; RND3 genes being upregulated and only 15 genes downregulated in
Rev: CGTCTGCCTGTGATTGGAGAG; SDC3 Fwd: GATG the samples isolated from epilepsy and hippocampal tissues
ACTCCTTTCCCGATGATG; SDC3 Rev: TGTCTCAATGCCCG when compared with the samples isolated from postmortem
ACTCC. tissues, indicating that these two control groups were very
similar.
Results As the two control group data sets clustered similarly in
Gene Expression Profile of Human the unsupervised hierarchical clustering and based on the
Glioblastoma-associated Microglia/Monocytes results of the ANOVA we combined both control groups and
To identify tumor-regulated transcripts in hGAMs, we used compared the eight hGAMs samples to all eight control sam-
MACS to isolate CD11b1 cells from human glioblastoma ples for all subsequent analysis. We identified 292 and 42
samples (primary and recurrent tumors) and from non-tumor genes that were significantly (Padj 0.05) two-fold up- or
brain samples (samples from epilepsy, hippocampal biopsies, downregulated, respectively (Fig. 1D,E; Tables 1 and 2; Sup-
and postmortem cortex samples) (Supporting Information porting Information Tables S2 and S3; Supporting Informa-
Table S1). FACS analysis of sorted samples revealed a high tion dataset S2).
purity of CD11b1 sorted cells. We isolated total RNA from
these samples and performed RNA sequencing on an Illumina hGAMs Upregulate Genes Associated with “Mitotic
HiSeq 1500. We validated the enrichment of CD11b1 cells Cell Cycle” and “Extracellular Matrix Organization”
in our samples using qPCR analyzing the expression of Iba1 We performed gene ontology (GO) analysis to functionally
and GFAP in the CD11b positive and negative fractions and link the 292 up- and 42 downregulated genes in hGAMs.
found a high enrichment of Iba1 in the positive fractions Terms related to “mitotic cell cycle”, “extracellular matrix
(Supporting Information Fig. S1). organization”, “organ development”, “cell adhesion”, “cell rec-
During data processing we implemented a nonspecific ognition”, “positive regulation of cell migration”/“leukocyte
filter to select only genes that had at least three reads in each migration”, “protein phosphorylation”, “collagen catabolic”/
hGAMs sample which resulted in 11,777 remaining genes. “metabolic process” were significantly overrepresented in the
Principal component analysis (PCA) and multidimensional 292 upregulated genes in hGAMs (Table 3, Supporting Infor-
scaling (MDS) plots of the 16 samples clearly distinguished mation dataset S3). Interestingly also the terms for “blood
the eight hGAMs samples from microglia-derived from post- vessel development”/“angiogenesis” and “response to
wounding”/“wound healing” were significantly overrepre- family members, FOXM1, TFDP1, FOSL2, SINA3, and
sented, functions that are typical features of other solid- JUND (Supporting Information data set S5). In accordance
tumor-associated macrophages. In contrast, there were no sig- with this, the expression of E2F1, E2F7, E2F8, and FOXM1
nificantly overrepresented terms for the 42 downregulated was significantly upregulated in hGAMs. The expression of
genes in hGAMs compared with the eight control samples. E2F4 was 1.4 fold downregulated. E2F transcription factors
We used Webgestalt to analyze the 292 upregulated are key regulators of cell cycle progression, and an upregula-
genes in hGAMs for known transcription factor binding sites tion of these genes might facilitate the high abundance of
and found a highly significant enrichment for E2F family genes related to mitosis and cell cycle regulation/progression
members, as well as E2F1 and E2F4 in particular (Supporting in the upregulated genes.
Information dataset S4). We performed an additional analysis We used ingenuity pathway analysis (IPA) to further
using iRegulon and found an enrichment for several E2F analyze our data set. In accordance with the GO analysis we
found a highly significant enrichment for the pathways “cell sorted using CD11b as a marker. We calculated the signifi-
cycle control of chromosomal replication” and “mitotic roles cantly regulated genes (fold change 2, Padj 0.05), excluded
of polo-like kinases” (Fig. 2). genes that are not present in human and found 940 genes
that were upregulated and 1005 genes that were
Human and Mouse GAMs Share Transcripts downregulated in mGAMs in comparison to na€ıve mouse
Associated with “Mitotic Cell Cycle” and microglia (Supporting Information dataset S6).
“Extracellular Matrix Organization” To investigate similarities between human and murine
We have previously shown that mGAMs isolated from experi- GAMs we compared both data sets for overlap using gene set
mental GL261 mouse gliomas exhibit a gene expression pat- enrichment analysis. We found that the 292 upregulated genes
tern that is different from na€ıve microglia or in vitro-activated in hGAMs were significantly enriched in the 940 upregulated
macrophages (Szulzewsky et al., 2015). Both human and genes in mGAMs (P 5 2.2 3 10216) and 111 genes (38% of
murine GAMs were isolated with the same method and the upregulated genes in hGAMs) were upregulated in both
hGAMs and mGAMs in comparison to the corresponding naive human and murine GAMs. Using Webgestalt we analyzed
control microglia (Fig. 3A,B, Supporting Information dataset these shared genes for known transcription factor binding
S7). We found no significant overlap between the sites and found a highly significant enrichment for E2F fam-
downregulated genes in hGAMs and mGAMs. ily members. In fact, both E2F1 and E2F8 were upregulated
We performed a GO analysis with the 111 genes that in both human and murine GAMs.
were upregulated in both human and murine GAMs. Most of In contrast, when looking at the 181 genes that were
the significantly overrepresented terms were related to significantly upregulated in human GAMs but not in murine
“mitotic cell cycle”; however, although less abundant, the GAMs, we only found very few significantly overrepresented
terms “cell adhesion”, “response to stimulus”, and GO terms. These terms were related to “system/organ devel-
“extracellular matrix organization” were also significantly opment”, “extracellular matrix organization”, “cell recog-
overrepresented (Fig. 3C,D, Supporting Information dataset nition”, “cell motility”, and “collagen metabolic/catabolic
S8). This indicates that the upregulation of transcripts process”. However, these 181 genes were not significantly
involved in cell cycle control is a common feature of both enriched for terms related to “mitotic cell cycle” (Fig. 3C,
August 2016
represented genes category
P value
GO:0000278 Mitotic 1.80E-20 73 854 CDKN3, NDC80, CENPE, CENPF, NES, UBE2C, CHEK1, CIT, ZWINT,
cell cycle CDCA5, E2F7, KIF18B, ESCO2, DHFR, E2F1, SKA1, SKA3, TPX2,
FOXM1, NCAPH, KIF4A, ASPM, BIRC5, ID4, KIF11, KIFC1, MAD2L1,
MCM2, MCM4, MYBL2, NEK2, ORC1, NUSAP1, GTSE1, GINS2,
NCAPG2, CDCA8, CEP55, FANCI, MCM10, PRKAR2B, CASC5, RRM1,
RRM2, CLSPN, CENPK, NCAPG, BUB1, BUB1B, TOP2A, TTK, TYMS,
WEE1, CENPM, E2F8, FZD3, CDT1, CDC45, NUF2, CCNB1, TICRR,
PRC1, PKMYT1, AURKB, CD28, KIF23, ESPL1, DLGAP5, CDK1,
MELK, GINS1, CDC6, KIF14
GO:0030198 Extracellular 5.45E-10 26 233 FBLN5, COL1A2, COL5A2, COL6A1, COL8A1, CTSL, AGT, EFEMP1,
matrix organization FN1, ANXA2, HAS2, TNC, IBSP, ITGA3, ITGA4, LAMC1, LGALS3,
LOX, MMP14, PLOD2, SDC1, SDC4, SPARC, TGFBI, ADAMTS2,
SDC3
GO:0048513 Organ development 1.92E-09 85 1717 GPNMB, CENPF, NES, CCR7, COL1A2, COL5A2, COL8A1, CD109,
E2F7, ESCO2, ACE, AEBP1, AGT, E2F1, APLNR, ECM1, EDNRB, AHR,
AK4, EFEMP1, PALLD, FOXM1, FLNB, CLEC5A, CADM1, FJX1,
ASPM, GCNT1, GPC1, ANXA2, HAS2, HMGB3, HOXB6, TNC, IBSP,
ID4, IL2RA, IL6, IL7R, AQP1, ITGA3, ITGA4, AREG, CCDC103,
LGALS1, LOX, SMAD1, KITLG, MKI67, MMP14, MT1G, NNMT,
NT5E, PLAU, PRRX1, ACP5, GNG12, ANKH, C8orf4, VANGL2,
PTPRZ1, RFX4, RRM1, SDC1, SDC4, SPARC, TGFBI, TGM2, TK1,
TOP2A, TYMS, VIM, E2F8, FZD3, C11orf63, KCNAB2, NRP1, APLN,
CCNB1, PAPSS2, CD28, ADAMTS2, CELSR1, MELK, KIF14
GO:0009611 Response to 4.88E-08 42 651 CENPE, CHL1, CLU, CCR7, CNR1, COL1A2, CD109, DHFR, DPYSL3,
wounding AGT, EDNRB, F3, FABP5, FN1, KIF4A, ANXA2, TNC, IL2RA, IL6,
AQP1, ITGA3, ITGA4, KIF11, KIFC1, LGALS1, LOX, NT5E, PLAU,
ACP5, PRKAR2B, VANGL2, CCL18, CCL20, SDC1, SDC4, SPARC,
WEE1, CCNB1, PAPSS2, CD28, KIF23, CELSR1
GO:0007155 Cell adhesion 4.65E-07 47 831 CDH2, GPNMB, FBLN5, CHL1, CCR7, COL6A1, COL8A1, PLEKHA7,
AGT, FN1, CADM1, GCNT1, TNFRSF21, ADAMDEC1, HAS2, TNC,
IBSP, IGFBP2, IL2RA, IL6, IL7R, ITGA3, ITGA4, RND3, LAMC1,
LGALS1, LGALS3, LY9, KITLG, MMP14, NRCAM, NT5E, PCDHGC3,
PLAU, EPDR1, ACKR3, CADM3, S100A10, SDC4, SIGLEC1, TGFBI,
TGM2, PPAP2B, CD28, CELSR1, SDC3, KIF14
1423
Szulzewsky et al.: Human GAMs RNA Sequencing
Supporting Information datasets S9, S10, and S11). This
CDH2, COL1A2, COL8A1, E2F7, AGT, ECM1, AHR, F3, FOXM1, FN1,
indicates that most of the genes that were upregulated in
233
62
associated Genes
27
18
3.81E-05
4.88E-05
P value
GO:0008037
GO:0030335
FIGURE 2: Human glioblastoma-associated microglia/monocytes upregulate cell cycle-associated genes Schematics of the pathways “cell
cycle control of chromosomal replication” and “mitotic roles of polo-like kinases”. Proteins highlighted in red are encoded by genes
that were significantly upregulated (Fold change 2, Padj £ 0.05) in hGAMs. Images were created through the use of IPA.
Human GAMs Lack Transcripts Associated with human control microglia. Eighteen additional genes showed a
Immune Activation non-significant >2-fold higher expression in hGAMs com-
Gene ontology analysis revealed no significant overrepresenta- pared with the human control microglia. In turn, six of the
tion of terms related to “immune activation/response” in the 112 genes were non-significantly lower expressed >2-fold in
upregulated genes in hGAMs. In contrast, these terms were hGAMs compared with human control microglia.
significantly overrepresented in the significantly upregulated We used Webgestalt and iRegulon to search for pre-
genes in mGAMs isolated from GL261 tumors (Fig. 3C). In dicted transcription factor binding sites in the 112 genes asso-
addition, we found no significant activation or inhibition of ciated with the GO term “immune response” and found
general immune pathways in hGAMs using IPA (Supporting enrichment for the factors IRF1, IRF2, IRF7, and NFjB in
Information Fig. S5), whereas, in mGAMs we found a signifi- both analyses. From these factors only Irf7 was significantly
cant activation of the pathways “interferon signaling”, upregulated 18-fold in mGAMs in comparison to murine
“complement system”, and “pattern recognition receptors”. control microglia, whereas Irf1, Irf2, Rela, Relb, Nfkb1, and
We analyzed how genes that were upregulated in Nfkb2 were unregulated. In contrast, IRF7 was nonsignifi-
mGAMs and were part of the GO term “immune response” cantly 1.7-fold lower expressed in hGAMs in comparison to
were regulated in hGAMs (Supporting Information dataset human control microglia. Similar to mGAMs IRF1, IRF2,
S13). There was no correlation of the fold change of immune RELA, RELB, NFKB1, and NFKB2 were also not regulated in
genes between the hGAMs experiment and the mGAMs hGAMs.
experiment (Spearman’s correlation coefficient 5 0.00797), We next compared the expression of different immune
indicating that these genes are regulated differently in human markers between human and mouse GAMs (Table 4, Sup-
versus mouse. Out of these 112 genes only 10 genes were sig- porting Information Table S4). Several genes were similarly
nificantly >2-fold upregulated in hGAMs in comparison to regulated in both species; however hGAMs displayed a lower
expression of several proinflammatory marker genes. Although was highly upregulated in mGAMs), and although signifi-
still highly expressed, IL1B expression was not significantly cantly upregulated, both ARG1 and MMP13 were expressed
upregulated in hGAMs. In contrast, Il1b expression was 2.46 only at very low levels. This highlights the fundamental dif-
fold upregulated in mGAMs. In addition, although almost ference between GAMs isolated from human GBM samples
three-fold upregulated in mGAMs, TNF expression was not and from experimental GL261 mouse gliomas, as the latter
significantly regulated in hGAMs. In addition, although cell type possess features that are related to “immune
upregulated in mGAMs, IFNG and NOS2 (coding for iNOS) activation” whereas hGAMs lack those features.
expression were both not regulated in hGAMs and only
expressed at very low levels. However, also several marker Comparison of Human and Murine GAMs to
genes associated with anti-inflammatory immune activation In Vitro-Stimulated Human Macrophages
showed a different regulation in hGAMs when compared To further investigate the immune activation states of these
with mGAMs. CD163 expression was strongly upregulated in cells we next compared the expression pattern of hGAMs and
hGAMs, whereas the expression of TGFB3, CCL22, CCL17, mGAMs to in vitro-activated human macrophages and
and CCL1 expression was downregulated. Furthermore, in murine microglia. We have previously shown that mouse
hGAMs ARG2 expression was not regulated (Arg2 expression GAMs share only a part of their up- and downregulated
FIGURE 4: Significantly enriched Reactome pathways in upregulated genes in hGAMs and mGAMs and in genes that are shared between
both species Significantly enriched pathways for genes A: upregulated in hGAMs versus human control microglia B: upregulated in
mGAMs versus murine control microglia C: genes that were upregulated in hGAMs but not upregulated in mGAMs D: genes upregu-
lated in mGAMs but not upregulated in hGAMs E: genes upregulated in both hGAMs and mGAMs. Listed pathways were selected to
avoid redundant listing.
genes with in vitro-activated mouse macrophages (Szulzewsky each of the nine major classes and calculated the up- and
et al., 2015). In addition, we have previously compared the downregulated genes in comparison to un-stimulated control
gene expression patterns of human macrophages after stimula- macrophages for these data sets (fold change 2, Padj 0.05).
tion with several factors in vitro (Xue et al., 2014). We could Subsequently we used gene set enrichment analysis to calcu-
show that in addition to the conventional pro- and anti- late the significant overlap between each of the data sets,
inflammatory activation with LPS 1 IFNc (in the old classifi- hGAMs, and mGAMs. We selected data sets for treatment
cation assigned as M1) and IL4 (in the old classification with GC (dexamethasone), IL4, LPS 1 IFNc, oleic acid,
assigned as M2), respectively, macrophages can also acquire PGE2 (prostaglandin E2), TNFa, TNFa 1 PGE2, TPP
activation states diverging from this bipolar axis when stimu- (TNFa 1 PGE2 1 TLR2-ligand), and ultra-pure LPS 1
lated with free fatty acids, high density lipoprotein (HDL), or immune complexes (upLPS 1 IC).
combinations of stimuli associated with chronic inflamma- We first compared the upregulated genes in hGAMs to
tion. Using 28 different substances we could identify nine the upregulated genes in the different in vitro-stimulated mac-
major activation classes. We selected one stimulation from rophage sets. We found that the 292 upregulated genes in
1429
Szulzewsky et al.: Human GAMs RNA Sequencing
hGAMs were most significantly enriched for genes upregu- Comparison of Human and Murine GAMs to Murine
lated in macrophages stimulated with upLPS 1 IC (P 5 4.6 Microglia Stimulated with LPS, IL4, and TGFb
3 1029), oleic acid (P 5 4.4 3 1026), TNFa (P 5 1.2 3 Since—at least in murine experimental gliomas—in addition
1024), and dexamethasone (P 5 1.5 3 1024). A less signifi- to invading monocytes and monocyte-derived macrophages
cant enrichment was found for the upregulated genes of mac- also brain-resident microglia constitute a major cellular frac-
rophages stimulated with TPP (P 5 1.8 3 1023), PGE2 tion of GAMs we used data sets from murine adult microglia
(P 5 4.9 3 1023), LPS 1 IFNc (P 5 0.01), and TNFa 1 that were stimulated with LPS, IL4, and TGFb. We calcu-
PGE2 (P 5 0.019) (Fig. 6A, Supporting Information Fig. lated the significantly regulated genes (fold change 2,
S6). Interestingly, the upregulated hGAMs genes were not sig- P 0.05) in comparison to un-stimulated control microglia
nificantly enriched in the upregulated genes of macrophages and performed gene set enrichment analysis to calculate the
stimulated with IL4. Rather, the upregulated genes in hGAMs significant overlap between these data sets, hGAMs, and
highly significantly overlapped with the downregulated genes mGAMs.
in macrophages stimulated with IL4 (P 5 3 3 1025), indicat- When comparing the upregulated genes in hGAMs we
ing an inverse regulation between hGAMs and macrophages only found a significant overlap to the upregulated genes for
treated with IL4. This is in contrast to previous studies microglia stimulated with TGFb (P 5 7.4 3 1024; Fig.
reporting that GAMs are activated towards an alternative acti- 6C,E). This emphasizes the lack of immune activation in
vation phenotype that resembles the state of macrophages hGAMs. We found no significant overlap between the down-
treated with IL4 (previously termed M2 activation). In addi- regulated genes in hGAMs and any of the in vitro data sets
tion to IL4, the upregulated genes in hGAMs were signifi- (Supporting Information Fig. S7C).
cantly overlapping with the downregulated genes of The mGAMs were highly overlapping with the
macrophages stimulated with TPP (P 5 4.4 3 1024), upregulated genes in microglia stimulated with LPS (P < 2.2
TNFa 1 PGE2 (P 5 3.2 3 1023), TNFa (P 5 0.025), and 3 10216) and the up- and downregulated genes in microglia
LPS 1 IFNc (P 5 0.038). In addition, the downregulated stimulated with IL4 (P 5 8.2 3 1026 and 8.8 3 1024,
genes of macrophages treated with oleic acid (P 5 6.3 3 respectively; Fig. 6D,F). In addition, we found a highly sig-
1023) constituted the only data set that was significantly nificant overlap between the downregulated genes in mGAMs
enriched when comparing the downregulated genes and the downregulated genes in microglia stimulated with
of hGAMs with these data sets (Supporting Information LPS (P 5 6.5 3 10212) and IL4 (P 5 1.89 3 1026; Support-
Fig. S7A). ing Information Fig. S7D).
In contrast, when comparing the upregulated genes in Interestingly, we found a completely inverse overlap
mGAMs to the in vitro-activated human macrophages we between mGAMs and microglia stimulated with TGFb. There
found a significant overlap to all nine conditions, with the was no significant overlap between the upregulated genes or
most significant overlaps to the upregulated genes of macro- the downregulated genes of both cell types, however, we
found a highly significant overlap between the upregulated
phages stimulated with TPP, LPS 1 INFc and TNFa 1 PGE2
genes in mGAMs and the downregulated genes in TGFb-
(all P < 2.2 3 10216, Fig. 6B). In addition we found a highly
stimulated microglia (P 5 2.6 3 10213) as well as between
significant overlap to the upregulated genes of macrophages
the downregulated genes in mGAMs and the upregulated
stimulated with oleic acid (P 5 4.6 3 10212), TNFa
genes in TGFb-treated microglia (1.9 3 1026).
(P 5 6.1 3 10212), and upLPS 1 IC (P 5 1.8 3 1027). This
The genes that were related to the GO term “mitotic
highlights a higher association of mGAMs with proinflamma-
cell cycle” and were upregulated in the human samples were
tory immune activation which is lacking in hGAMs.
not found to be regulated in the cultured macrophage or
Although less strong, we also found a significant overlap to
microglial cells stimulated under different conditions (Sup-
macrophages stimulated with IL4 (P 5 0.015). We also found
porting Information Fig. S7E).
a highly significant overlap when comparing the upregulated
genes in mGAMs to the downregulated genes in macrophages Human GAMs and Immune Cells Isolated from Lung
stimulated with TPP (P 5 1.4 3 1028), upLPS 1 IC Tumors Share Gene Expression Patterns
(P 5 5.7 3 1025), TNFa 1 PGE2 (P 5 2.3 3 1024), and To investigate similarities in gene expression between hGAMs
LPS 1 IFNc (P 5 4.9 3 1024). When comparing the down- and tumor-associated monocytes/macrophages (TAMs) in
regulated genes in mGAMs to the in vitro data sets we found other solid cancers we used a published gene list of significantly
a significant enrichment for the upregulated genes of macro- regulated genes from CD11b1CD331CXCR22 immature
phages treated with dexamethasone (P 5 0.038; Supporting monocytic myeloid cells and CD11b1CD332CXCR21 neu-
Information Fig. S7B). trophils that were isolated from human non-small cell lung
Serres et al., 2014). This further highlights the important role vation in non-tumor control microglia which was isolated
of hGAMs in glioblastoma. from epilepsy and hippocampal tissues, as well as from post-
We could show that there is a significant overlap between mortem cortex tissues. The genes that were related to
the upregulated genes of murine and human CD11b1 GAMs. “immune response” were not expressed at high levels in con-
This overlap includes genes related to “extracellular matrix trol cells; however a direct comparison of the absolute gene
organization” and “cell adhesion”, but was primarily restricted expression between control microglia from human and
to genes related to “mitotic cell cycle regulation”. While several mouse tissues is difficult due to the differences of microar-
genes related to “immune activation” were upregulated in ray and RNA sequencing. A recent study has found that,
mGAMs this was not the case for human GAMs. In line with based on the analysis of several immune related genes using
this, IFNG and NOS2 were unchanged and expressed at only Nanostring technology hGAMs share features of unactivated
very low levels in hGAMs and TNF was expressed at medium M0 macrophages (Gabrusiewicz et al., 2016), which is in
levels, but not upregulated. The lack of genes related to line with the absence of immune activation that we
“immune response/activation” in whole tumor tissues of high detected. Moreover, the immune response of mGAMs in the
grade versus low grade glioma has previously been reported GL261 model could be explained by the nature of the
(Mieczkowski et al., 2015). We have previously compared the model itself. GL261 cells originate from C3H mice and are
expression profiles of CD11b1 mGAMs to in vitro-activated not a syngeneic transplantation model and the immune
murine macrophages and could show that mGAMs express response of mGAMs isolated from these tumors may repre-
markers associated with both pro- and anti-inflammatory acti- sent a mild version of graft versus host response (Akbasak
vation (Szulzewsky et al., 2015). Here we compared the differ- et al., 1991; Ausman et al., 1970; Seligman and Shear,
entially expressed genes of CD11b1 hGAMs and mGAMs to 1939). In addition, the relatively short tumor latency of
those of in vitro-stimulated human macrophages and murine around 3 weeks might be too short for complete reprogram-
microglia. We have recently demonstrated that macrophages ming GAMs. It remains to be shown if different glioma
can acquire activation states that extend beyond the bipolar axis models, such as genetically engineered mouse models or
of pro- and anti-inflammatory activation when stimulated with xenograft models mimic the phenotype of hGAMs more
free fatty acids, high density lipoprotein (HDL), or combina- closely. This might be especially interesting in the light of
tions of stimuli associated with chronic inflammation (Xue an ongoing discussion about the reliability of animal models
et al., 2014). In general we found a lower overlap between the to study human immune activation (Mestas and Hughes,
in vitro data sets to hGAMs than mGAMs. The comparison to 2004; Reynolds and Haniffa, 2015; Shay et al., 2013).
these in vitro data sets showed that hGAMs share a smaller over- Despite these differences, the GL261 model is suitable
lap to cells treated in vitro with pro-inflammatory stimuli, such to investigate distinct aspects that are shared between hGAMs
as LPS 1 IFNc and TNFa, than mGAMs. This underlines the and mGAMs. Our findings provide new insights into the
results from the GO analysis which did not detect a significant biology of human glioblastoma-associated microglia/mono-
enrichment of terms related to “immune activation” in hGAMs, cytes and give detailed information about the validity of
whereas in mGAMs a high abundance of these terms was murine experimental models. These data will also serve as a
detected. It remains to be shown if there is a homogeneously guide to select potential therapeutic targets which are relevant
activated GAMs population throughout the tumor or—which in the human context.
seems more likely—if GAMs activation depends on the location
within or the stage of the tumor.
Acknowledgment
The lack of the GO terms related to “immune
response” in hGAMs versus control GAMs needs to be fur- Grant sponsor: Deutsche Forschungsgemeinschaft; Grant
ther investigated. In contrast to murine samples, which allow numbers: KE 329/30-1, SY 144/4-1, SZ 350/1-1; Grant
the use of truly na€ıve and age-matched samples, this is not sponsor: NeuroCure. Grant sponsor: NIH Grant; Grant
possible for the human setting. The brain tissue we used for number: U01CA160882-01A1; Grant sponsor: Deutsche For-
the isolation of our human control samples originated either schungsgemeinschaft; Grant number: SFB645/SFB704
from postmortem tissues of aged patients (average donor age The authors thank Hans van Mierlo and Manja Litjens
of 80.6 years, however only one out of five donors showed from the Department of Translational Neuroscience at the
signs of cognitive dysfunction) or from epilepsy and hippo- UMC Utrecht.
campal sclerosis tissue (average donor age of 37 years). In
contrast, the average donor age for the glioblastoma samples
References
Akbasak A, Oldfield EH, Saris SC. 1991. Expression and modulation of major
was 59 years. Thus, a possible explanation for the lack of histocompatibility antigens on murine primary brain tumor in vitro.
immune activation in hGAMs may be an already present acti- J Neurosurg 75:922–929.
Barnholtz-Sloan JS, Verhaak RG. 2015. Whole-genome and multisector Aldape K. 2006. Molecular subclasses of high-grade glioma predict progno-
exome sequencing of primary and post-treatment glioblastoma reveals pat- sis, delineate a pattern of disease progression, and resemble stages in neuro-
terns of tumor evolution. Genome Res 25:316–327. genesis. Cancer Cell 9:157–173.
Kruger TE, Miller AH, Godwin AK, Wang J. 2014. Bone sialoprotein and Pietras A, Katz AM, Ekstrom EJ, Wee B, Halliday JJ, Pitter KL, Werbeck JL,
osteopontin in bone metastasis of osteotropic cancers. Crit Rev Oncol Hema- Amankulor NM, Huse JT, Holland EC. 2014. Osteopontin-CD44 signaling in
tol 89:330–341. the glioma perivascular niche enhances cancer stem cell phenotypes and pro-
motes aggressive tumor growth. Cell Stem Cell 14:357–369.
Lawrence M, Huber W, Pages H, Aboyoun P, Carlson M, Gentleman R,
Morgan MT, Carey VJ. 2013. Software for computing and annotating Pong WW, Walker J, Wylie T, Magrini V, Luo J, Emnett RJ, Choi J, Cooper
genomic ranges. PLoS Comput Biol 9:e1003118. ML, Griffith M, Griffith OL, Rubin JB, Fuller GN, Piwnica-Worms D, Feng X,
Hambardzumyan D, DiPersio JF, Mardis ER, Gutmann DH. 2013. F11R is a
Li W, Graeber MB. 2012. The molecular profile of microglia under the influ-
novel monocyte prognostic biomarker for malignant glioma. PLoS One 8:
ence of glioma. Neuro-Oncology 14:958–978.
e77571.
Lisi L, Laudati E, Navarra P, Dello Russo C. 2014. The mTOR kinase inhibitors
Pyonteck SM, Akkari L, Schuhmacher AJ, Bowman RL, Sevenich L, Quail DF,
polarize glioma-activated microglia to express a M1 phenotype.
Olson OC, Quick ML, Huse JT, Teijeiro V, Setty M, Leslie CS, Oei Y, Pedraza
J Neuroinflamm 11:125.
A, Zhang J, Brennan CW, Sutton JC, Holland EC, Daniel D, Joyce JA. 2013.
Love MI, Huber W, Anders S. 2014. Moderated estimation of fold change CSF-1R inhibition alters macrophage polarization and blocks glioma progres-
and dispersion for RNA-seq data with DESeq2. Genome Biol 15:550. sion. Nat Med 19:1264–1272.
Markovic DS, Vinnakota K, Chirasani S, Synowitz M, Raguet H, Stock K, Sliwa R Development Core Team. 2011. R: A language and environment for statis-
M, Lehmann S, Kalin R, van Rooijen N, Holmbeck K, Heppner FL, Kiwit J, tical computing. ISBN: 3-900051-07-0 http://www.R-project.org/.
Matyash V, Lehnardt S, Kaminska B, Glass R, Kettenmann H. 2009. Gliomas
Reinartz S, Schumann T, Finkernagel F, Wortmann A, Jansen JM, Meissner
induce and exploit microglial MT1-MMP expression for tumor expansion.
Proc Natl Acad Sci USA 106:12530–12535. W, Krause M, Schworer AM, Wagner U, Muller-Brusselbach S, Muller R. 2014.
Mixed-polarization phenotype of ascites-associated macrophages in human
Markovic DS, Vinnakota K, van Rooijen N, Kiwit J, Synowitz M, Glass R, ovarian carcinoma: Correlation of CD163 expression, cytokine levels and early
Kettenmann H. 2011. Minocycline reduces glioma expansion and invasion by relapse. Int J Cancer 134:32–42.
attenuating microglial MT1-MMP expression. Brain Behav Immun 25:624–
628. Reynolds G, Haniffa M. 2015. Human and mouse mononuclear phagocyte
networks: A tale of two species? Front Immunol 6:330.
Mestas J, Hughes CC. 2004. Of mice and not men: Differences between
mouse and human immunology. J Immunol 172:2731–2738. Sangaletti S, Di Carlo E, Gariboldi S, Miotti S, Cappetti B, Parenza M, Rumio
C, Brekken RA, Chiodoni C, Colombo MP. 2008. Macrophage-derived SPARC
Meyer M, Reimand J, Lan X, Head R, Zhu X, Kushida M, Bayani J, Pressey bridges tumor cell-extracellular matrix interactions toward metastasis. Cancer
JC, Lionel AC, Clarke ID, Cusimano M, Squire JA, Scherer SW, Bernstein M, Res 68:9050–9059.
Woodin MA, Bader GD, Dirks PB. 2015. Single cell-derived clonal analysis of
human glioblastoma links functional and genomic heterogeneity. Proc Natl Sarkar S, Doring A, Zemp FJ, Silva C, Lun X, Wang X, Kelly J, Hader W,
Acad Sci USA 112:851–856. Hamilton M, Mercier P, Dunn JF, Kinniburgh D, van Rooijen N, Robbins S,
Forsyth P, Cairncross G, Weiss S, Yong VW. 2014. Therapeutic activation of
Mieczkowski J, Kocyk M, Nauman P, Gabrusiewicz K, Sielska M, Przanowski macrophages and microglia to suppress brain tumor-initiating cells. Nat Neu-
P, Maleszewska M, Rajan WD, Pszczolkowska D, Tykocki T, Grajkowska W, rosci 17:46–55.
Kotulska K, Roszkowski M, Kostkiewicz B, Kaminska B. 2015. Down-regulation
of IKKbeta expression in glioma-infiltrating microglia/macrophages is associ- Seligman AM, Shear MJ. 1939. Studies in Carcinogenesis: Experimental pro-
ated with defective inflammatory/immune gene responses in glioblastoma. duction of brain tumors in mice with methylcholanthrene. Am J Cancer 37:
Oncotarget 6:33077–33090. 364–395.
Milacic M, Haw R, Rothfels K, Wu G, Croft D, Hermjakob H, D’Eustachio P, Serres E, Debarbieux F, Stanchi F, Maggiorella L, Grall D, Turchi L, Burel-
Stein L. 2012. Annotating cancer variants and anti-cancer therapeutics in Vandenbos F, Figarella-Branger D, Virolle T, Rougon G, Van Obberghen-
reactome. Cancers (Basel) 4:118021211. Schilling E. 2014. Fibronectin expression in glioblastomas promotes cell
cohesion, collective invasion of basement membrane in vitro and orthotopic
Muller A, Brandenburg S, Turkowski K, Muller S, Vajkoczy P. 2015. Resident tumor growth in mice. Oncogene 33:3451–3462.
microglia, and not peripheral macrophages, are the main source of brain
tumor mononuclear cells. Int J Cancer 137:278–288. Shannon P, Markiel A, Ozier O, Baliga NS, Wang JT, Ramage D, Amin N,
Schwikowski B, Ideker T. 2003. Cytoscape: A software environment for integrated
Murray PJ, Allen JE, Biswas SK, Fisher EA, Gilroy DW, Goerdt S, Gordon S, models of biomolecular interaction networks. Genome Res 13:2498–2504.
Hamilton JA, Ivashkiv LB, Lawrence T, Locati M, Mantovani A, Martinez FO,
Mege JL, Mosser DM, Natoli G, Saeij JP, Schultze JL, Shirey KA, Sica A, Shay T, Jojic V, Zuk O, Rothamel K, Puyraimond-Zemmour D, Feng T,
Suttles J, Udalova I, van Ginderachter JA, Vogel SN, Wynn TA. 2014. Macro- Wakamatsu E, Benoist C, Koller D, Regev A, ImmGen C. 2013. Conservation
phage activation and polarization: nomenclature and experimental guidelines. and divergence in the transcriptional programs of the human and mouse
Immunity 41:14–20. immune systems. Proc Natl Acad Sci USA 110:2946–2951.
Olah M, Raj D, Brouwer N, De Haas AH, Eggen BJ, Den Dunnen WF, Biber Sica A, Schioppa T, Mantovani A, Allavena P. 2006. Tumour-associated
KP, Boddeke HW. 2012. An optimized protocol for the acute isolation of macrophages are a distinct M2 polarised population promoting tumour
human microglia from autopsy brain samples. Glia 60:96–111. progression: Potential targets of anti-cancer therapy. Eur J Cancer 42:717–727.
Pasini FS, Zilberstein B, Snitcovsky I, Roela RA, Mangone FR, Ribeiro U, Jr., Sottoriva A, Spiteri I, Piccirillo SG, Touloumis A, Collins VP, Marioni JC,
Nonogaki S, Brito GC, Callegari GD, Cecconello I, Alves VA, Eluf-Neto J, Curtis C, Watts C, Tavare S. 2013. Intratumor heterogeneity in human glio-
Chammas R, Federico MH. 2014. A gene expression profile related to blastoma reflects cancer evolutionary dynamics. Proc Natl Acad Sci USA 110:
immune dampening in the tumor microenvironment is associated with poor 4009–4014.
prognosis in gastric adenocarcinoma. J Gastroenterol 49:1453–1466.
Szulzewsky F, Pelz A, Feng X, Synowitz M, Markovic D, Langmann T, Holtman
Patel AP, Tirosh I, Trombetta JJ, Shalek AK, Gillespie SM, Wakimoto H, IR, Wang X, Eggen BJ, Boddeke HW, Hambardzumyan D, Wolf SA,
Cahill DP, Nahed BV, Curry WT, Martuza RL, Louis DN, Rozenblatt-Rosen O, Kettenmann H. 2015. Glioma-associated microglia/macrophages display an
Suva ML, Regev A, Bernstein BE. 2014. Single-cell RNA-seq highlights intratu- expression profile different from M1 and M2 polarization and highly express
moral heterogeneity in primary glioblastoma. Science 344:1396–1401. Gpnmb and Spp1. PLoS One 10:e0116644.
Phillips HS, Kharbanda S, Chen R, Forrest WF, Soriano RH, Wu TD, Misra A, Trikha P, Sharma N, Pena C, Reyes A, Pecot T, Khurshid S, Rawahneh M,
Nigro JM, Colman H, Soroceanu L, Williams PM, Modrusan Z, Feuerstein BG, Moffitt J, Stephens JA, Fernandez SA, Ostrowski MC, Leone G. 2015. E2f3 in