- Open Access
Induction of a common microglia gene expression signature by aging and neurodegenerative conditions: a co-expression meta-analysis
© Holtman et al.; licensee BioMed Central. 2015
Received: 3 April 2015
Accepted: 12 April 2015
Published: 23 May 2015
Microglia are tissue macrophages of the central nervous system that monitor brain homeostasis and react upon neuronal damage and stress. Aging and neurodegeneration induce a hypersensitive, pro-inflammatory phenotype, referred to as primed microglia. To determine the gene expression signature of priming, the transcriptomes of microglia in aging, Alzheimer’s disease (AD), and amyotrophic lateral sclerosis (ALS) mouse models were compared using Weighted Gene Co-expression Network Analysis (WGCNA).
A highly consistent consensus transcriptional profile of up-regulated genes was identified, which prominently differed from the acute inflammatory gene network induced by lipopolysaccharide (LPS). Where the acute inflammatory network was significantly enriched for NF-κB signaling, the primed microglia profile contained key features related to phagosome, lysosome, antigen presentation, and AD signaling. In addition, specific signatures for aging, AD, and ALS were identified.
Microglia priming induces a highly conserved transcriptional signature with aging- and disease-specific aspects.
Neuroinflammation plays an important role in the progression of neurodegenerative diseases, with a prominent role for microglia [1-5]. Microglia are the primary innate immune cells of the brain and the first to respond to a variety of stimuli, like neuronal damage and infections, initially to restore homeostasis . Upon activation, microglia release increased amounts of inflammatory cytokines, phagocytose cellular debris, and support tissue remodeling .
Microglia are versatile cells that, depending on environmental cues, are able to adopt different phenotypes but clear phenotypical identities have not been established. Microglia, like other cultured macrophages, are often classified into inflammatory (M1) and alternatively activated (M2) phenotypes [7,8], in which the M1 phenotype was originally induced using LPS or IFNγ stimulation, and the M2 phenotype using IL-4, IL-13 or IL-10.
In several neurodegenerative disorders and upon aging, chronic activation of microglia has been reported to induce a hypersensitive phenotype, often referred to as primed [9-11]. Primed microglia do not secrete high amounts of cytokines, but when triggered by pro-inflammatory stimuli, they become hyper-reactive, secreting large amounts of cytokines, chemokines, and other reactive molecules associated with neurotoxicity. We recently reported that microglia priming in a mouse model for accelerated aging was induced by an affected neuronal environment and not by intrinsic aging . Although microglia priming is becoming a generally accepted concept , at present priming primarily is a functional definition and it is unclear whether microglia priming is a homogeneous phenotype with a specific transcriptional signature or a heterogenous phenotype with model-system specific transcriptional profiles and what the functional consequences of priming are.
In this study, these aspects were addressed by comparing the gene expression networks in pure cell populations of primed microglia that were isolated from mouse models for neurodegenerative disease and aging. The mouse models included are: 1) aged mice; 2) accelerated aging mice (Ercc1∆/KO), a DNA repair-deficient mouse model that displays features of accelerated aging ; 3) APPswe/PS1dE9 (App-Ps1), a mouse model for Alzheimer’s disease, carrying transgenes for mutated Amyloid Precursor Protein and Presenilin-1 and 4) a mouse model for Amyotrophic Lateral Sclerosis (Sod193A, abbreviated as Sod1), a line carrying a mutation in the SuperOxide Dismutase-1 gene, encoding an enzyme involved in free radical degradation, resulting in motor neuron degeneration in the spinal cord .
In addition, the microglia priming network was also analyzed using (unsorted) brain tissue expression data. The mouse models included are: 1) aged mice; 2) App-Ps1 mice; 3) rTg4510, a mouse line expressing P301L mutant human tau [13,14]; 4) an ME7 model of murine prion disease, associated with neuronal loss and microglial activation [15,16] (for an overview of mouse models and data sets used, see Additional file 1: Table S1).
Transcriptional profiles of microglia isolated from four mouse models of aging and disease and four brain tissue expression data sets were analyzed in parallel and compared using WGCNA . In contrast to traditional differential gene expression analysis, co-expression network analysis does not regard genes as single entities, but incorporates the interrelation of genes to generate structures called modules. WGCNA has been reported to be a useful approach to integrate immunology with bioinformatics , and has been applied to evaluate common denominators in meta-analyses or disease models [1,19-21]. By raising the network to a power function, WGCNA results in a heterogeneous network dominated by a few highly connected nodes (hubs), which link the rest of the less connected nodes to the system . These hub genes are likely control points or key genes that modulate the expression of the network-module and thereby are considered important for the observed phenotype [19,21,22]. In this paper, a WGCNA-based meta-analysis was applied to determine the transcriptional signature and hub genes of different microglia phenotypes: primed, age- and neurodegeneration-associated, and acute inflammatory.
Materials and methods
Microglia and brain tissue expression profiling
Pure ex vivo microglia populations were obtained by FACS sorting and RNA was isolated as recently described in [10,23]. Three microglia expression datasets were generated; 4 and 24 months old DBA/2 J and C57/SJL mice (Harlan, The Netherlands) were used. For acute LPS activated microglia, C57BL/6 mice (4 months, Harlan, The Netherlands) were i.p. injected with LPS (10 mg/kg) or PBS and microglia were isolated after 4 hr. RNA quantity and quality of the RNA samples was checked using the Experion RNA HighSense Analysis kit (BioRad, Cat.no. 700-7105), samples with high integrity (RIN > 7) were used for expression profiling. RNA was amplified with Nugen Ovation PicoSL WTA system (Cat nr. 3310-48), labeled with the Encore BiotinIL Module (Cat nr. 4210-48) and hybridized to Illumina MouseRef8 bead-chip microarrays. Raw data were generated using Illumina Genome studio.
rTg4510 mice carry a human P301L mutant tau transgene downstream of the tetracycline operon-responsive element (TRE), whose expression is driven by a second transgene expressing the tetracycline-controlled transactivator (tTA) under control of the Ca2+/calmodulin-dependent protein kinase II α (CaMKIIα) promoter. tTA constitutively induces tau expression via the TRE, but can be inactivated with doxycycline administration. Transgenic mice were bred at Taconic, Denmark. Mice expressing the tTA activator transgenes were maintained on a 126S6 background strain (Taconic) and mutant tau responder mice were maintained in the FVB/N background strain . rTg4510 mice were perfused and sacrificed at 2, 4 , 6 and 8 months of age. RNA was isolated from brain tissue and hybridized to Illumina MouseWG6 bead-chip microarrays. All experiments were approved by the animal experimentation committees of the University of Groningen and the Royal Netherlands Academy for Arts and Sciences and are in accordance with the European Communities Council Directive #86/609 and the directives of the Danish National Committee on Animal Research Ethics. Previously published transcriptomes from pure microglia, brain tissue, and cultured and stimulated macrophages were included in our analysis, for detailed platform and, experimental design information see Additional file 1: Table S1 [15,24-28].
Pre-processing of transcriptomes
Raw expression values were preprocessed using R and Bioconductor package Limma . Samples with an average inter-sample correlation three standard deviations below the mean inter-sample correlation after normalization were filtered out and this procedure was repeated until all samples met the inclusion criteria. Quantile normalization was applied to the Illumina microarrays. To eliminate batch effects between both physiological aging datasets, the ComBat function was applied . For Agilent array preprocessing, background correction was performed with an offset of 50 followed by Lowess within array normalization and Quantile between array normalization. Relative intensities were converted into expression values. The Affymetrix microarrays were preprocessed using the Expresso-function of R package Affy . The parameters were set to RMA background correction and quantile normalization, with pm correct pmonly and a medianpolish. From the Sod1 RNA-sequencing dataset  the published RPKM-values were used, to which quantile normalization was applied to ensure that all samples have the same distribution in order to generate a more stable network.
Select representative probes
Datasets from different platforms were made comparable at the level of gene symbols. The WGCNA collapseRows function was applied to calculate the representative gene expression for several probes, associated with a single gene . The default method ‘MaxMean’ was used to select the row with the highest mean value. Similarly for the RNA-sequencing data several RefSeq accession numbers, associated on the same gene, were collapsed on gene symbols. Next, all gene symbols from the different platforms were intersected and only those genes that were present on all included platforms were used for further analysis.
Parallel and consensus network formation
In all pure microglia datasets, genes with low variation or low connectivity were filtered out, resulting 7512 genes in the 5 parallel networks for the individual datasets, as described previously . In the combined pure microglia and brain tissue analysis, no further filtering was applied, because less genes were present as more platforms were included, resulting in 9936 genes that were taken into this analysis. Subsequently, the topological overlap (TO) matrices from all five models were scaled such that the 95th quantiles matched. A consensus TO matrix was calculated using the minimal value (pMin) for all gene pairs in any of the scaled TO matrices. From each of these six TO matrices, a dendrogram was generated by average linkage hierarchical clustering. Using the tree cut function, branches of highly co-expressed genes were grouped into modules. Only modules of a minimum size of 100 genes were considered for further analysis. Modules from the five model networks were defined using a hard-clustering approach, meaning that only genes directly clustered in the module were taken, and the module EigenGenes (ME) were calculated. For the consensus network, modules were defined using a soft-clustering approach, in which meta-q values and meta ME correlation thresholds were used to determine which genes were included (min correlation of 0.25 and min meta FDR-corrected q-value of 1E-8).
For each module, a Kruskall Wallis non parametric test, was used to assess differential expression of the ME with respect to aging or disease. Only modules with a p < 0.005 were considered to be differentially expressed, and were used for further analysis. A Fisher’s exact test was used to determine if the modules from the 5 datasets had a significant number of overlapping genes and these results were depicted as an overlap Heatmap. Modules were annotated by using WEB-based GEneSeT AnaLysis Toolkit (WEBGESTALT) to perform KEGG pathway and GO analysis [22,32]. The gene list that resulted from the intersection of the Illumina and Agilent arrays and Illumina Sequencing was used as the background list. To compare our modules to other gene expression studies WGCNA’s userListEnrichment function was used .
Hub gene classification to compare different WGCNA core networks
The importance of a gene in a network module is determined by the strength of the correlation to the ModuleEigene, or module membership (kME) value . The 35 genes with the highest (most significant) kME were taken from the networks to be analyzed. Module membership correlation thresholds were used to determine whether a gene is highly associated (i.e., a “hub” gene; FDR-q < 1.0E-11 for primed and acute FDR-p < 1E-7), moderately associated (below hub-gene association and FDR-p < 1.0E-2) or not associated (FDR-p > 1.0E-2) with a module. This strategy resulted in five clusters of genes: 2 clusters with hub genes significantly correlated with one and not the other network, 2 clusters with hub genes significantly correlated with one network and less significantly with the other dataset and a cluster containing hub genes strongly correlated to both networks.
Gene set enrichment analysis: pre-ranked analysis
Systematic differences between two network modules were determined with gene set enrichment pre-ranked list analysis . The 1000 most significantly module membership associated genes from either the acute and/or the primed networks were taken into the analysis, negative correlations were set to zero, and genes were ranked on strength of module membership to both networks. The difference in rank-values between consensus primed and acute was used as input for the analysis. GSEA pre-ranked list analysis was applied using a 1000 permutations.
Quantitative RT-PCR and immunohistochemistry
Differential gene expression analysis
Differential gene expression was applied to the pure microglia datasets (see Additional file 3: Table S3 for these lists) as well as to datasets related to several in vitro stimulation conditions like LPS, IL-4, and IFNγ [27,34] which were used as genesets for UserListEnrichment. Differential gene expression was done using Limma  for microarray data and EdgeR for RNA-seq data .
Aim and outline of the co-expression network analysis
Generation of co-expression networks and identification of modules related to phenotype
Expression profiles of pure microglia populations from different mouse models were used to generate co-expression networks. In these co-expression networks, we searched for WGCNA modules that were differentially expressed between conditions (i.e. young vs. aged, control vs. App-Ps1 etc.). Two classes of differentially expressed modules were identified; either up-regulated or down-regulated between conditions. The up-regulated modules are the blue modules in aged, Ercc1, App-Ps1, and Sod1 microglia, the red module in LPS-stimulated microglia as well as the Sod1-specific dark-turquoise module. The down-regulated modules are the dark-green modules in aged, Ercc1, App-Ps1, and Sod1 microglia, the magenta module in Ercc1, and the green-yellow module in LPS-stimulated microglia. For an overview of all modules and their relation to different conditions with their associated p-values see Additional file 4: Table S4.
To determine the (dis)similarities between these modules, a pair-wise comparison of all differentially expressed modules was performed. A highly significant overlap was observed between the 4 up-regulated blue modules (p-values ranging from 1.79E-79 to 1.62E-146) (Figure 1, phase 4) and the overlap of these blue modules with the acute LPS-induced red module was much less (p-values ranging from 2.44E-8 to 1.13E-29. These data indicate that aging and neurodegeneration induce a very similar up-regulated gene expression profile in microglia. For a pair-wise comparison between all modules identified in all mouse models, see Additional file 5: Figure S1.
The overlap between the down-regulated dark-green modules in aged, Ercc1, App-Ps1, and Sod1 microglia, as well as the overlap of these modules with the down-regulated green-yellow module of LPS-activated microglia, was less pronounced (p-values ranging from 1.67E-04 to 5.56E-65). No significant overlap of the Ercc1-specific magenta module with any other differentially expressed module was observed.
In order to address the observed overlap between the blue and dark-green modules in aged, Ercc1, App-Ps1, and Sod1 microglia, we generated a consensus network, consisting of co-expressed genes shared between the four individual datasets. This consensus network contained two modules: a blue and a dark-green module. The consensus blue module, consisting of 295 genes, is up-regulated and the consensus dark-green consensus module (205 genes) is down-regulated in aged, Ercc1, App-Ps1, and Sod1 microglia (these gene lists can be found in Additional file 6: Table S5).
GO and KEGG annotation of modules related to phenotype
The priming modules strongly overlap with an independent age-induced microglia expression dataset
The effect of aging on microglia gene expression was recently determined by direct RNA sequencing with a focus on proteins for sensing endogenous ligands and microbes, referred to as the microglia sensome . Using this dataset, genes significantly increased in expression during aging were determined and compared to our up-regulated primed and acute microglia modules. This microglia aging profile significantly overlapped with the primed microglia blue modules (ranging from p = 1.56E-25 to 6.39E-44; Figure 2a), and much less significant with the acute LPS-stimulated red module (p = 8.33E-13). This observation validates our observation that the up-regulated profiles of primed microglia is very similar to the transcriptional profile reported for aged microglia using an independent expression dataset.
M1- and M2-classifications in relation to the blue and red modules
Microglia, in analogy to macrophage activation terminology, are often classified as M1 or M2, with M1 considered as a classical pro-inflammatory activation state and M2 as a tissue supportive, remodeling or anti-inflammatory state . Using the WGCNA function userListEnrichment, the up-regulated primed blue and acute red modules were compared to M1 and M2 macrophage datasets (Figure 2). The acute red microglia module showed a highly significant overlap with LPS-stimulated macrophages (p = 2.45E-45), and this was a much more significant overlap than was observed with primed microglia (p = 1.22E-5 to 3.06E-16). Only the Sod1 up-regulated blue module had a minor overlap with the M2 up-regulated profile (p = 6.92E-5). These results suggested that the primed microglia phenotypes did not resemble a clear M1, M2 or intermediate phenotype.
Microglia activation and priming is associated with a decreased expression of the ‘microglia unique signature’
Recently, a ‘microglia unique’ gene expression signature was reported , but the relationship between this signature and microglial activation is unknown. We compared this profile to our differentially expressed microglia modules. Where no significant overlap with the up-regulated blue module was detected, surprisingly all down-regulated primed dark-green and acute green-yellow modules significantly overlapped with this core microglial signature (Figure 2b; p-values ranging from p = 1.17E-5 to p = 8.11E-24). Hub genes of the down-regulated dark-green and green-yellow module were determined, and many of them were present in the microglia-unique expression signature, including Mertk, Tmem119, P2ry12, P2ry13, SPARC, and Cx3cr1 (Additional file 8: Figure S2). This suggests that upon activation or priming, microglia not only acquire an activation signature, but also decrease their ‘surveilling’ state expression profile. The genes of down-regulated consensus modules are listed in Additional file 6: Table S5.
Priming and acute LPS activation induce distinct transcriptional programs
“Acute” hub genes mark processes specifically activated in the acute microglia response, whereas “primed” hub genes mark processes occurring in primed microglia. “General” hubs mark processes common to both acute and primed microglia and therefore relate to general microglial activation. Genes that belong to the “acute” hub category are not significantly associated with any of the other datasets and examples of “acute”-unique hub genes are Map3k8 and Socs3. The “general” hub category contained genes that were up-regulated (and highly connected) in all five mouse models. This hub included genes like Tlr2, Il-1β, Cxcl10, and Spp1, representing a group of genes consistently up-regulated in activated microglia. Importantly, also a “primed” hub was identified containing genes specifically increased in expression and highly connected in primed microglia, including genes as Apoe, Axl, Clec7a, Itgax (also known as CD11c), and Lgals3 (also known as Galectin-3 and Mac2). Of these genes Lgals3 has been associated with microglia priming during accelerated aging  and microglia activation following axonal injury . Also two intermediate hub categories were defined, containing genes that were primarily highly connective in either primed microglia or acute LPS activation. The “mainly primed” hub contained genes like Cybb and Csf1 (also known as Mcsf) that were highly connected in the priming datasets and were also significantly correlated in the acute data set, but to a lesser extent. In the “mainly acute” hub, genes were significantly but not very strongly associated to any dataset other than the “acute” profile. It contained genes like Nf-kb2 and Irf1 that were highly connected in the acute dataset and were also significantly correlated in the primed data sets, although not in all mouse models or to a lesser degree.
Ranked gene set enrichment analysis was applied to determine potential functional differences between primed and acute networks directly. Gene sets that were significantly enriched in acute activation were NF-κB factor p65 (RelA), toll like receptor, and NOD like receptor signaling (Figure 2c). Gene sets significantly enriched in primed microglia were: Alzheimer’s and Parkinson’s disease signaling, oxidative phosphorylation, mitochondria, and lysosome (Figure 2d; for all annotations see Additional file 7: Table S6). These data indicate that the expression profiles of primed and acute activated microglia differed in several ways, and the most prominent changes were oxidative phosphorylation, and lysosome in primed microglia and NF-κB signaling in acute activation. These results are in agreement with the findings of the Webgestalt-KEGG-pathway analysis, further strengthening the notion that the primed microglia profile substantially differs from the M1 or M2-phenotype observed in acute activated microglia.
Specific expression profiles for aged, Ercc1, App-Ps1, and Sod1 microglia
Validation of the primed blue module and model system-specific gene profiles
Differential expression of several hub genes of the consensus blue module as well as mouse model-specific gene expression was validated using quantitative RT-PCR analysis of Ercc1 and App-Ps1 microglia. Expression levels of primed microglia blue module hub-genes Axl, Cybb, Apoe, Clec7a, and Cox6a were determined (Figure 3b). All these genes were significantly increased in Ercc1 and App-PsS1 microglia compared to controls, confirming the validity of the consensus primed microglia blue module. Lgals3 is a hub gene in the primed microglia consensus module and identified as a marker for primed microglia in accelerated aging Ercc1 mutant mice . Brain sections of aged, Ercc1, and App-Ps1 mice were stained and Iba1/Lgals3 double positive cells were only observed in aged, Ercc1, and App-Ps1 animals and not in young or aged-matched controls (Additional file 9: Figure S3). Mouse model-specific expression of several hub genes was confirmed using quantitative RT-PCR analysis of Ercc1 and App-Ps1 microglia. CD14, Gm1673, and Ldlr expression was restricted to the App-Ps1 blue module and significantly induced in 15 months old App-Ps1 microglia while their expression was not increased in Ercc1 microglia (Figure 3d). The Ccl3 gene was restricted to the general aging module and the Oas2 gene was most significantly associated with the Ercc1 blue module. Their expression level was significantly increased in Ercc1, but not in App1-Ps1 microglia, confirming the specificity of the identified sub-modules.
Signatures of acute and primed microglia are preserved in brain tissue samples
To determine, if the transcriptional profiles associated with primed microglia are preserved in mouse brain tissue, expression sets of App-Ps1, aged, rTg4510 and ME7 prion infected mice (-/+ LPS) were analyzed. rTg4510 mice overexpress the a mutant form of human tau that causes fronto-temporal dementia and parkinsonism linked to chromosome 17 (FTDP-17). rTg4510 mice provide a model for tauopathies, including Alzheimer’s disease. ME7 prion brain infection is a frequently used model system to induce microglia priming. The overlap between the up-regulated blue modules in pure microglia and significantly up-regulated genes in brain tissue expression data was determined as a measure of preservation. Significant overlap was observed between pure microglia up-regulated blue modules and the up-regulated genes in App-Ps1 (p-values ranging from 7.35E-10 to 2.68E-32) and aging (p-values ranging from 9.24E-7 to 7.56E-20; Additional file 10: Table S7) brain tissue. No significant overlap of the upregulated App-Ps1 and aging genes was observed with the up-regulated acute red module of microglia from LPS injected mice. Interestingly, the ME7 response genes (main effect of ME7) significantly overlapped with the primed blue modules (p-values ranging from 7.31E-17 to 2.58E-50), but much less with the acute red module (p = 3.5E-9). In contrast, a highly significant overlap between the LPS response genes (main effect of LPS) with the up-regulated acute red module was observed (Additional file 10: Table S7, p = 8.7E-36) and the overlap with the blue primed modules was less pronounced (p-values ranging from 1.67E-4 to 1.74E-14). Similar results were obtained with the rTg4510 dataset; a highly significant overlap with the primed blue microglia modules was found (p-values ranging from 5.95E-8 to 2.68E-30). These data suggest that signatures of primed, but not acute activated, microglia are preserved in brain tissue expression data from models of Alzheimer’s disease, prion infection and, aging.
Comparative WGCNA analysis of brain tissue and pure microglia expression data
The overlap between the consensus brain tissue green module with the identified individual pure microglia blue modules, the consensus blue module, and the acute LPS red module was determined (Additional file 11: Table S8b). A significant overlap was observed with all primed microglia blue modules (p = 5.16E-48 to 4.95E-18) but a less significant overlap was present with the acute activation red module (1.61E-7). Next, this consensus green microglia-enriched profile was compared to the consensus blue primed microglia profile (Figure 1a), hub genes were allocated, and five categories were defined (see Methods): “brain tissue derived microglia signature”, “mainly brain tissue derived microglia signature”, “general microglia”, “mainly pure microglia”, and “pure microglia”. The “general microglia”-hub consists of highly connective genes that were found both in the pure microglia and in the microglia-enriched brain tissue modules and contains genes like: Spp1, Csf1, Axl, B2m, Lgals3bp, and Tlr2. The “pure microglia” hub contains, amongst others, the Clic4, Rap2B, and Gapdh genes that are highly connective in microglia but not in the microglia-enriched brain tissue module. The “brain tissue-derived microglia signature” hub contains genes like C1QB, C1QC, and Irf8. The intermediate “mainly pure microglia” hub contains genes like Cybb and Igf1, and the “mainly brain tissue microglia” hub contains previously reported microglial hub genes Tyrobp and Trem2, as well as the astrocyte marker Gfap (Figure 4c,d). The genes of these pure microglia and brain tissue modules are listed in Additional file 6: Table S5. Since these data indicate that C1QB, C1QC, Tyrobp and Trem2 expression is not increased in primed or acute activated microglia, the expression level of these genes was checked a recently published database by Zhang and colleagues  and we found that all these genes are very highly expressed in microglia and therefore likely identified as hub genes for microglia in brain tissue expression data.
These data show that the consensus profile of microglia (-enriched) modules found in different pure microglia and brain tissue expression datasets share similarities at the hub gene level, but also are critically different. These differences are likely caused by a combination of changes in microglia cell numbers in the brain under neuropathological conditions and microglia-intrinsic changes in gene expression. Several papers have shown that neuropathology is associated with increased microglia cell proliferation [10,24,42]. As a consequence, typical hub genes of microglia modules identified using brain tissue expression data are not necessarily hub genes in pure microglia expression data.
Discussion and conclusions
One of the main objectives of this study was to investigate the hallmarks of gene expression profiles of primed microglia isolated from mouse models for aging and neurodegeneration. We show that in all mouse models investigated, independent of the origin and platform used, these primed microglia expressed a core gene expression profile, which substantially differed from the inflammatory gene network observed in acutely activated, pro-inflammatory, microglia. The degree of preservation of this core gene expression profile in physiological aging, Ercc1, App-Ps1, Sod1, and Me7 mice made it very likely that these microglia acquired a similarly primed phenotype. In the current paper, we show that primed microglia are clearly different from M1 and M2 macrophages or M1 ex-vivo isolated microglia. The observation that activated microglia in chronic brain disease/in a neurodegenerative disease do not resemble an M1 or M2 phenotype was already suggested by Chiu et al. , whose Sod1-dataset was used in the current study. They proposed that the pattern of regulation of a particular set of genes, including Axl, can distinguish LPS stimulated microglia from Sod1-associated microglia. Based on the primed microglia gene expression network we predict that that primed microglia are characterized by expression of cell surface markers like Itgax, Lgals3, Axl, Clec7a, MHC class 2, and Cxcr4.
The major functions of the primed microglia gene expression network show that these cells are involved in immune-, phagosome-, lysosome-, oxidative phosphorylation, and antigen presentation signaling pathways. These functions fit the needs of chronically degenerating brain tissue. In response to tissue damage, microglia migrate to the site of injury and phagocytose tissue debris or damaged cells , thereby potentially degrading healthy synapses and contributing to the ongoing degenerative process . In the used mouse models for neurodegeneration and aging, the phagosome and closely related lysosome KEGG pathways were indeed significantly enriched in the consensus blue module, suggesting aging- or neuropathology-induced phagocytic activity of primed microglia.
Using WGCNA, hub genes were identified that are likely candidate genes that drive the observed phenotype [1,19-21]. Interestingly, four hub genes unique to the primed microglia gene expression network, Galactin-3, Igf1, Csf1, and Axl were previously shown to be instrumental for microglia functions including proliferation, activation, and phagocytosis. Igf1 signals through Galactin-3 and inactivation of the Galactin-3 gene resulted in Igf1 insensitivity, decreased microglia activity and a significant increase in the ischemic lesion size [38,45]. Igf1 signaling is related to neuroplasticity and neuroprotection  and is shown to mediate motor neuron protection and prolonged survival in ALS animal models. Csf-1-mediated microglia proliferation has been shown to be important in chronic neurodegeneration  and inhibition of Csf-r1 signaling in mice resulted in complete ablation of microglia . The tyrosine kinase receptor Axl is up-regulated in microglia in various neurodegenerative and demyelinating conditions, such as cuprizone-induced demyelination, EAE and in MS lesions, and is shown to play an important role in phagocytosis of apoptotic cells and myelin [48-50]. Axl KO mice experience enhanced inflammation and delayed myelin removal in EAE-mice , and fewer mature oligodendrocytes and more axonal damage in cuprizone induced demyelination . Contrarily, a recent paper showed that Axl is an inflammatory phagocytic receptor whose expression was induced by pro-inflammatory mediators . The up-regulation of these hub-genes suggests that microglial priming also has adaptive aspects, necessary to cope with increased neurodegeneration and environmental stress.
In a recent study, microglia were compared to other myeloid immune cells and a microglial signature, which is dependent on TGFβ signaling, was reported . This list of genes was enriched in the down-regulated modules of both primed and acutely activated microglia. Two genes that are particularly interesting in this respect are SPARC and Cx3cr1, which are hub genes of the down-regulated consensus module. SPARC regulates the activity of growth factors and cytokines. Enhanced microglia proliferation, microgliosis around stroke lesions, and enhanced recovery is observed in SPARC null mice . Cx3cr1 is ubiquitously expressed by microglia and plays an important role in microglia-neuron communication . It was shown that Cx3cr1 deficiency resulted in microglia activation, and increased neurodegeneration following LPS injections in PD and ALS-models . Moreover, Cx3cr1 deficiency worsens the AD-related neuronal deficits, associated with microglial activation and elevated chemokines . In contrast, others reported that in two mouse models for AD, Cx3cr1-deficiency resulted in increased beta amyloid clearance and prevented neuron loss [54,55]. Furthermore, lack of Cx3cr1 was shown to reduce infarct size, ischemic damage and inflammation . The notion is that constitutively expressed Cx3cl1 (the ligand for Cx3cr1) provides a tonic inhibitory signal to microglia to remain quiescent, and that deficiency results in hyperactivated microglia [57,58]. This indicates that upon activation, microglia partially lose their resting signature and acquire a priming or acute signature.
Besides the aforementioned common primed microglia gene expression network, additional, specific elements of the microglia gene expression networks were found, exclusively for the aging, AD, or ALS mouse models used in our study. Although it is impossible to eliminate potential confounding factors like isolation protocols, mouse strain, age, CNS regions used, and different expression profiling methodologies, we could confirm the model-specific differences in gene expression in App-Ps1 and Ercc1 mice with quantitative RT-PCR. These model-specific changes in gene expression were related to an increased interferon-type 1 signature in both aging models, an altered cell-cycle GO in Sod1 and decreased neurotrophin signaling in App-Ps1. An aging-associated type-1 interferon signature is recently described in the choroid plexus of aging mice and humans . This interferon signature had a negative effect on brain function, and was induced by brain-derived signals derived from the CSF. In addition, it was suggested that interferon type 1 plays a role in microglial priming .
Microglia isolated from App-Ps1 mice were hallmarked by a decreased expression of neurotrophin related genes. In App-Ps1 mice, amyloid plaque load increases with age and is associated with a strong immune signaling profile to which microglia contribute . Interestingly, two App-Ps1 model system-specific hub-genes, LDL-receptor and CD14, are associated with Amyloid Beta (Aβ) clearance. Microglia surround amyloid fibril deposits and have been suggested to be involved in their phagocytosis . Increased LDL receptor expression prevented amyloid deposition and led to an increased Aβ clearance . CD14 is required for Aβ stimulation of microglia and inhibition of CD14 prevents initiation of Aβ phagocytosis . This indicates that using WGCNA, we could identify disease-specific module components with known biological relevance in AD.
WGCNA is often used to generate microglia specific profiles in brain tissue expression data [1,19,21]. Similarly, we identified a microglia-enriched consensus module in brain datasets, indicating that primed microglia in Me7, aging, rTg4510, and App-Ps1 mice had a similar gene expression profile. However, comparison of this brain tissue microglia-enriched module to the pure microglia expression profiles, showed substantial differences in hub genes. In the brain tissue data, Trem2 and Tyrobp are identified as hub genes, where in pure microglia they are only weakly associated with the consensus priming module, suggesting that these genes might not play a critical role in microglia priming. Tyrobp and Trem2 are highly expressed in microglia, and therefore often identified as hub genes of microglial modules in brain tissue datasets. Changes caused by altered relative cell numbers in brain tissue expression data are readily detected using WGCNA, resulting in cell type-specific modules. However, it is very difficult to discriminate between cell intrinsic alterations in gene expression levels and changes in cell numbers in cell type-specific modules. In addition, other cell types, such as astrocytes, possibly contaminate these modules. Summarizing, our data indicate that a complementary WGCNA analysis of both pure cell populations and brain tissue expression data is required in order to fully unveil regulatory gene networks.
In this study, we analyzed primed microglia from different neurodegenerative conditions. Microglia priming is often regarded as a confounding factor, resulting in exacerbation of neurodegeneration in a wide range of conditions . The core microglia priming module described in this study supports the notion of a generic microglia response in different neuropathologies, but this module mostly contains genes related to phagocytosis and cell proliferation, with tissue protective elements. This indicates that primed microglia adopt an altered inflammatory profile predominantly adapted for phagocytic clearance and in a state of immune readiness, possibly necessary to cope with inflammatory and neurodegenerative conditions.
The brain tissue App-Ps1 datasets was kindly supplied by Koen Bossers and Joost Verhaagen. The microglia illustrations were generated by Hilmar van Weering. This work was supported by funds from the Internationale Stichting Alzheimer Onderzoek [ISAO 08504 and 12509], NANONET COST [BM1002], the Netherlands Organization for Scientific Research [NWO; VICI], the Graduate School of Medical Sciences of the University Medical Center Groningen and the Dutch Hersenstichting. The authors acknowledge technical support by the FACS facility of the University Medical Center Groningen.
- Zhang B, Gaiteri C, Bodea L-G, Wang Z, McElwee J, Podtelezhnikov AA, Zhang C, Xie T, Tran L, Dobrin R, Fluder E, Clurman B, Melquist S, Narayanan M, Suver C, Shah H, Mahajan M, Gillis T, Mysore J, MacDonald ME, John R, Lamb JR, Bennett DA, Molony C, David J, Stone DJ, Gudnason V, Myers AJ, Schadt EE, Neumann H (2013) Integrated systems approach identifies genetic nodes and networks in late-onset Alzheimer’s disease. Cell 153:707–20View ArticlePubMed CentralPubMedGoogle Scholar
- Cribbs DH, Berchtold NC, Perreau V, Coleman PD, Rogers J, Tenner AJ, Cotman CW (2012) Extensive innate immune gene activation accompanies brain aging, increasing vulnerability to cognitive decline and neurodegeneration: a microarray study. J Neuroinflammation 9:179View ArticlePubMed CentralPubMedGoogle Scholar
- Crotti A, Benner C, Kerman BE, Gosselin D, Lagier-Tourenne C, Zuccato C, Cattaneo E, Gage FH, Cleveland DW, Glass CK (2014) Mutant Huntingtin promotes autonomous microglia activation via myeloid lineage-determining factors. Nat Neurosci 17:513–521View ArticlePubMed CentralPubMedGoogle Scholar
- Chiu IM, Morimoto ETA, Goodarzi H, Liao JT, O’Keeffe S, Phatnani HP, Muratet M, Carroll MC, Levy S, Tavazoie S, Myers RM, Maniatis T (2013) A neurodegeneration-specific gene-expression signature of acutely isolated microglia from an amyotrophic lateral sclerosis mouse model. Cell Rep 4:385–401View ArticlePubMed CentralPubMedGoogle Scholar
- Li MD, Burns TC, Morgan AA, Khatri P (2014) Integrated multi-cohort transcriptional meta-analysis of neurodegenerative diseases. Acta Neuropathol Commun 2:93View ArticlePubMed CentralPubMedGoogle Scholar
- Kettenmann H, Hanisch U-K, Noda M, Verkhratsky A (2011) Physiology of microglia. Physiol Rev 91:461–553View ArticlePubMedGoogle Scholar
- Gordon S (2003) Alternative activation of macrophages. Nat Rev Immunol 3:23–35View ArticlePubMedGoogle Scholar
- Boche D, Perry VH, Nicoll JAR (2013) Review: Activation patterns of microglia and their identification in the human brain. Neuropathol Appl Neurobiol 39:3–18View ArticlePubMedGoogle Scholar
- Perry VH, Holmes C (2014) Microglial priming in neurodegenerative disease. Nat Rev Neurol 10:217–24View ArticlePubMedGoogle Scholar
- Raj DDA, Jaarsma D, Holtman IR, Olah M, Ferreira FM, Schaafsma W, Brouwer N, Meijer MM, de Waard MC, van der Pluijm I, Brandt R, Kreft KL, Laman JD, de Haan G, Biber KPH, Hoeijmakers JHJ, Eggen BJL, Boddeke HWGM (2014) Priming of microglia in a DNA-repair deficient model of accelerated aging. Neurobiol Aging 35:2147–60View ArticlePubMedGoogle Scholar
- Norden DM, Godbout JP (2012) Microglia of the Aged Brain: Primed to be Activated and Resistant to Regulation. Neuropathol Appl Neurobiol. doi: 10.1111/j.1365-2990.2012.01306.xGoogle Scholar
- Rajaram S, Oono Y (2010) NeatMap–non-clustering heat map alternatives in R. BMC Bioinformatics 11:45View ArticlePubMed CentralPubMedGoogle Scholar
- Santacruz K, Lewis J, Spires T, Paulson J, Kotilinek L, Ingelsson M, Guimaraes A, DeTure M, Ramsden M, McGowan E, Forster C, Yue M, Orne J, Janus C, Mariash A, Kuskowski M, Hyman B, Ashe KH (2005) Tau suppression in a neurodegenerative mouse model improves memory function. Science 309:476–481View ArticlePubMed CentralPubMedGoogle Scholar
- Dalby NO, Volbracht C, Helboe L, Larsen PH, Jensen HS, Egebjerg J, Elvang AB (2014) Altered function of hippocampal CA1 pyramidal neurons in the rTg4510 mouse model of tauopathy. J Alzheimers Dis 40:429–42PubMedGoogle Scholar
- Lunnon K, Teeling JL, Tutt AL, Cragg MS, Glennie MJ, Perry VH (2011) Systemic inflammation modulates Fc receptor expression on microglia during chronic neurodegeneration. J Immunol 186:7215–24View ArticlePubMedGoogle Scholar
- Xiang W, Windl O, Wünsch G, Dugas M, Kohlmann A, Dierkes N, Westner IM, Kretzschmar HA (2004) Identification of differentially expressed genes in scrapie-infected mouse brains by using global gene expression technology. J Virol 78:11051–60View ArticlePubMed CentralPubMedGoogle Scholar
- Langfelder P, Horvath S (2008) WGCNA: an R package for weighted correlation network analysis. BMC Bioinformatics 9:559View ArticlePubMed CentralPubMedGoogle Scholar
- Kidd BA, Peters LA, Schadt EE, Dudley JT (2014) Unifying immunology with informatics and multiscale biology. Nat Immunol 15:118–127View ArticlePubMed CentralPubMedGoogle Scholar
- Miller JA, Horvath S, Geschwind DH (2010) Divergence of human and mouse brain transcriptome highlights Alzheimer disease pathways. Proc Natl Acad Sci U S A 107:12698–703View ArticlePubMed CentralPubMedGoogle Scholar
- Langfelder P, Mischel PS, Horvath S (2013) When is hub gene selection better than standard meta-analysis? PLoS One 8:e61505View ArticlePubMed CentralPubMedGoogle Scholar
- Miller JA, Woltjer RL, Goodenbour JM, Horvath S, Geschwind DH (2013) Genes and pathways underlying regional and cell type changes in Alzheimer’s disease. Genome Med 5:48View ArticlePubMed CentralPubMedGoogle Scholar
- Wang J, Duncan D, Shi Z, Zhang B (2013) WEB-based GEne SeT AnaLysis Toolkit (WebGestalt): update 2013. Nucleic Acids Res 41:W77–83View ArticlePubMed CentralPubMedGoogle Scholar
- Orre M, Kamphuis W, Osborn LM, Jansen AHP, Kooijman L, Bossers K, Hol EM. (2014) Isolation of glia from Alzheimer’s mice reveals inflammation and dysfunction. Neurobiol Aging. doi: 10.1016/j.neurobiolaging.2014.06.004Google Scholar
- Wirz KTS, Bossers K, Stargardt A, Kamphuis W, Swaab DF, Hol EM, Verhaagen J (2013) Cortical beta amyloid protein triggers an immune response, but no synaptic changes in the APPswe/PS1dE9 Alzheimer’s disease mouse model. Neurobiol Aging 34:1328–42View ArticlePubMedGoogle Scholar
- Kedmi M, Orr-Urtreger A (2011) The effects of aging vs. α7 nAChR subunit deficiency on the mouse brain transcriptome: aging beats the deficiency. Age (Dordr) 33:1–13View ArticleGoogle Scholar
- Xiang W, Hummel M, Mitteregger G, Pace C, Windl O, Mansmann U, Kretzschmar HA (2007) Transcriptome analysis reveals altered cholesterol metabolism during the neurodegeneration in mouse scrapie model. J Neurochem 102:834–47View ArticlePubMedGoogle Scholar
- El Chartouni C, Rehli M Comprehensive analysis of TLR4-induced transcriptional responses in interleukin 4-primed mouse macrophages. Immunobiology 215:780–7.Google Scholar
- Hickman SE, Allison EK, El Khoury J (2008) Microglial dysfunction and defective beta-amyloid clearance pathways in aging Alzheimer’s disease mice. J Neurosci 28:8354–60View ArticlePubMed CentralPubMedGoogle Scholar
- Smyth GK (2005) Limma : Linear Models for Microarray Data. Bioinformatics pages:397–420.Google Scholar
- Johnson WE, Li C, Rabinovic A (2007) Adjusting batch effects in microarray expression data using empirical Bayes methods. Biostatistics 8:118–27View ArticlePubMedGoogle Scholar
- Gautier L, Cope L, Bolstad BM, Irizarry RA (2004) affy–analysis of Affymetrix GeneChip data at the probe level. Bioinformatics 20:307–15View ArticlePubMedGoogle Scholar
- Zhang B, Kirov S, Snoddy J (2005) WebGestalt: an integrated system for exploring gene sets in various biological contexts. Nucleic Acids Res 33:W741–8View ArticlePubMed CentralPubMedGoogle Scholar
- Miller JA, Cai C, Langfelder P, Geschwind DH, Kurian SM, Salomon DR, Horvath S (2011) Strategies for aggregating gene expression data: the collapseRows R function. BMC Bioinformatics 12:322View ArticlePubMed CentralPubMedGoogle Scholar
- Martinez FO, Helming L, Milde R, Varin A, Melgert BN, Draijer C, Thomas B, Fabbri M, Crawshaw A, Ho LP, Ten Hacken NH, Cobos Jimenez V, Kootstra NA, Hamann J, Greaves DR, Locati M, Mantovani A, Gordon S (2013) Genetic programs expressed in resting and IL-4 alternatively activated mouse and human macrophages: similarities and differences. Blood 121:e57–69View ArticlePubMedGoogle Scholar
- Robinson MD, McCarthy DJ, Smyth GK (2010) edgeR: a Bioconductor package for differential expression analysis of digital gene expression data. Bioinformatics 26:139–40View ArticlePubMed CentralPubMedGoogle Scholar
- Hickman SE, Kingery ND, Ohsumi TK, Borowsky ML, Wang L, Means TK, El Khoury J (2013) The microglial sensome revealed by direct RNA sequencing. Nat Neurosci 16:1896–905View ArticlePubMedGoogle Scholar
- Butovsky O, Jedrychowski MP, Moore CS, Cialic R, Lanser AJ, Gabriely G, Koeglsperger T, Dake B, Wu PM, Doykan CE, Fanek Z, Liu L, Chen Z, Rothstein JD, Ransohoff RM, Gygi SP, Antel JP, Weiner HL (2014) Identification of a unique TGF-β-dependent molecular and functional signature in microglia. Nat Neurosci 17:131–43View ArticlePubMed CentralPubMedGoogle Scholar
- Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, Paulovich A, Pomeroy SL, Golub TR, Lander ES, Mesirov JP (2005) Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A 102:15545–50View ArticlePubMed CentralPubMedGoogle Scholar
- Venkatesan C, Chrzaszcz M, Choi N, Wainwright MS (2010) Chronic upregulation of activated microglia immunoreactive for galectin-3/Mac-2 and nerve growth factor following diffuse axonal injury. J Neuroinflammation 7:32View ArticlePubMed CentralPubMedGoogle Scholar
- Hawrylycz MJ, Lein ES, Guillozet-Bongaarts AL, Shen EH, Ng L, Miller JA, van de Lagemaat LN, Smith KA, Ebbert A, Riley ZL, Abajian C, Beckmann CF, Bernard A, Bertagnolli D, Boe AF, Cartagena PM, Chakravarty MM, Chapin M, Chong J, Dalley RA, Daly BD, Dang C, Datta S, Dee N, Dolbeare TA, Faber V, Feng D, Fowler DR (2012) An anatomically comprehensive atlas of the adult human brain transcriptome. Nature 489:391–9View ArticlePubMed CentralPubMedGoogle Scholar
- Zhang Y, Chen K, Sloan SA, Bennett ML, Scholze AR, O’Keeffe S, Phatnani HP, Guarnieri P, Caneda C, Ruderisch N, Deng S, Liddelow SA, Zhang C, Daneman R, Maniatis T, Barres BA, Wu JQ (2014) An RNA-Sequencing Transcriptome and Splicing Database of Glia, Neurons, and Vascular Cells of the Cerebral Cortex. J Neurosci 34:11929–47View ArticlePubMed CentralPubMedGoogle Scholar
- Gómez-Nicola D, Fransen NL, Suzzi S, Perry VH (2013) Regulation of microglial proliferation during chronic neurodegeneration. J Neurosci 33:2481–93View ArticlePubMedGoogle Scholar
- Cunningham C (2013) Microglia and neurodegeneration: the role of systemic inflammation. Glia 61:71–90View ArticlePubMedGoogle Scholar
- Schafer DP, Lehrman EK, Kautzman AG, Koyama R, Mardinly AR, Yamasaki R, Ransohoff RM, Greenberg ME, Barres BA, Stevens B (2012) Microglia sculpt postnatal neural circuits in an activity and complement-dependent manner. Neuron 74:691–705View ArticlePubMed CentralPubMedGoogle Scholar
- Lalancette-Hébert M, Swarup V, Beaulieu JM, Bohacek I, Abdelhamid E, Weng YC, Sato S, Kriz J (2012) Galectin-3 is required for resident microglia activation and proliferation in response to ischemic injury. J Neurosci 32:10383–95View ArticlePubMedGoogle Scholar
- Llorens-Martín M, Torres-Alemán I, Trejo JL (2009) Mechanisms mediating brain plasticity: IGF1 and adult hippocampal neurogenesis. Neuroscientist 15:134–48View ArticlePubMedGoogle Scholar
- Elmore MRP, Najafi AR, Koike MA, Dagher NN, Spangenberg EE, Rice RA, Kitazawa M, Matusow B, Nguyen H, West BL, Green KN (2014) Colony-stimulating factor 1 receptor signaling is necessary for microglia viability, unmasking a microglia progenitor cell in the adult brain. Neuron 82:380–97View ArticlePubMed CentralPubMedGoogle Scholar
- Hoehn HJ, Kress Y, Sohn A, Brosnan CF, Bourdon S, Shafit-Zagardo B (2008) Axl-/- mice have delayed recovery and prolonged axonal damage following cuprizone toxicity. Brain Res 1240:1–11View ArticlePubMedGoogle Scholar
- Weinger JG, Brosnan CF, Loudig O, Goldberg MF, Macian F, Arnett HA, Prieto AL, Tsiperson V, Shafit-Zagardo B (2011) Loss of the receptor tyrosine kinase Axl leads to enhanced inflammation in the CNS and delayed removal of myelin debris during Experimental Autoimmune Encephalomyelitis. J Neuroinflammation 8:49View ArticlePubMed CentralPubMedGoogle Scholar
- Weinger JG, Omari KM, Marsden K, Raine CS, Shafit-Zagardo B (2009) Up-regulation of soluble Axl and Mer receptor tyrosine kinases negatively correlates with Gas6 in established multiple sclerosis lesions. Am J Pathol 175:283–93View ArticlePubMed CentralPubMedGoogle Scholar
- Zagórska A, Través PG, Lew ED, Dransfield I, Lemke G (2014) Diversification of TAM receptor tyrosine kinase function. Nat Immunol 15:920–928View ArticlePubMed CentralPubMedGoogle Scholar
- Lloyd-Burton SM, York EM, Anwar MA, Vincent AJ, Roskams AJ (2013) SPARC regulates microgliosis and functional recovery following cortical ischemia. J Neurosci 33:4468–81View ArticlePubMedGoogle Scholar
- Harrison JK, Jiang Y, Chen S, Xia Y, Maciejewski D, McNamara RK, Streit WJ, Salafranca MN, Adhikari S, Thompson DA, Botti P, Bacon KB, Feng L (1998) Role for neuronally derived fractalkine in mediating interactions between neurons and CX3CR1-expressing microglia. Proc Natl Acad Sci U S A 95:10896–901View ArticlePubMed CentralPubMedGoogle Scholar
- Cho S-H, Sun B, Zhou Y, Kauppinen TM, Halabisky B, Wes P, Ransohoff RM, Gan L (2011) CX3CR1 protein signaling modulates microglial activation and protects against plaque-independent cognitive deficits in a mouse model of Alzheimer disease. J Biol Chem 286:32713–22View ArticlePubMed CentralPubMedGoogle Scholar
- Fuhrmann M, Bittner T, Jung CKE, Burgold S, Page RM, Mitteregger G, Haass C, LaFerla FM, Kretzschmar H, Herms J (2010) Microglial Cx3cr1 knockout prevents neuron loss in a mouse model of Alzheimer’s disease. Nat Neurosci 13:411–3View ArticlePubMed CentralPubMedGoogle Scholar
- Cardona AE, Pioro EP, Sasse ME, Kostenko V, Cardona SM, Dijkstra IM, Huang D, Kidd G, Dombrowski S, Dutta R, Lee J-C, Cook DN, Jung S, Lira SA, Littman DR, Ransohoff RM (2006) Control of microglial neurotoxicity by the fractalkine receptor. Nat Neurosci 9:917–24View ArticlePubMedGoogle Scholar
- Dénes A, Ferenczi S, Halász J, Kornyei Z, Kovacs KJ (2008) Role of CX3CR1 (fractalkine receptor) in brain damage and inflammation induced by focal cerebral ischemia in mouse. J Cereb Blood Flow Metab 28:1707–21View ArticlePubMedGoogle Scholar
- Biber K, Neumann H, Inoue K, Boddeke HWGM (2007) Neuronal “On” and “Off” signals control microglia. Trends Neurosci 30:596–602View ArticlePubMedGoogle Scholar
- Baruch K, Deczkowska A, David E, Castellano JM, Miller O, Kertser A, Berkutzki T, Barnett-Itzhaki Z, Bezalel D, Wyss-Coray T, Amit I, Schwartz M (2014) Aging-induced type I interferon response at the choroid plexus negatively affects brain function. Science 346:89–93View ArticlePubMedGoogle Scholar
- Orre M, Kamphuis W, Osborn LM, Melief J, Kooijman L, Huitinga I, Klooster J, Bossers K, Hol EM (2014) Acute isolation and transcriptome characterization of cortical astrocytes and microglia from young and aged mice. Neurobiol Aging 35:1–14View ArticlePubMedGoogle Scholar
- Itagaki S, McGeer PL, Akiyama H, Zhu S, Selkoe D (1989) Relationship of microglia and astrocytes to amyloid deposits of Alzheimer disease. J Neuroimmunol 24:173–82View ArticlePubMedGoogle Scholar
- Sokolowski JD, Mandell JW (2011) Phagocytic clearance in neurodegeneration. Am J Pathol 178:1416–28View ArticlePubMed CentralPubMedGoogle Scholar
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.