|Home | About | Journals | Submit | Contact Us | Français|
Reactive astrogliosis is characterized by a profound change in astrocyte phenotype in response to all CNS injuries and diseases. To better understand the reactive astrocyte state, we used Affymetrix GeneChip arrays to profile gene expression in populations of reactive astrocytes isolated at various time points after induction using two mouse injury models, ischemic stroke and neuroinflammation. We find reactive gliosis consists of a rapid, but quickly attenuated induction of gene expression after insult and identify two induced genes, Lcn2 and Serpina3n, as strong markers of reactive astrocytes. Strikingly, reactive astrocyte phenotype strongly depended on the type of inducing injury. Although there is a core set of genes that is up-regulated in reactive astrocytes from both injury models, at least 50% of the altered gene expression is specific to a given injury type. Reactive astrocytes in ischemia exhibited a molecular phenotype that suggests that they may be beneficial or protective, whereas reactive astrocytes induced by LPS exhibited a phenotype that suggests that they may be detrimental. These findings demonstrate that, despite well established commonalities, astrocyte reactive gliosis is a highly heterogeneous state in which astrocyte activities are altered to respond to the specific injury. This raises the question of how many subtypes of reactive astrocytes exist. Our findings provide transcriptome databases for two subtypes of reactive astrocytes that will be highly useful in generating new and testable hypotheses of their function, as well as for providing new markers to detect different types of reactive astrocytes in human neurological diseases.
Reactive astrogliosis is a universal response of astrocytes to brain injuries and diseases as diverse as trauma, infection, neurodegeneration and ischemia. Astrocytes’ abilities to help support neurons, regulate the blood brain barrier, remodel the extracellular space, control immune cells, and control synapse formation and function may all be of great import in influencing how the brain fares during and following injury (Pekny and Nilsson, 2005; Sofroniew, 2009).
Reactive astrocytes undergo dramatic morphological changes (Wilhelmsson et al., 2006) and various alterations in gene expression have been observed (Sofroniew, 2009). It has been long debated whether reactive astrocytes are harmful or beneficial. In the past few years, both types of effects have been observed. For instance, reactive astrocytes can inhibit axon regeneration after CNS injury (McKeon et al., 1991; Bradbury et al., 2002; Fitch and Silver, 2008; Alilain et al., 2011) and can produce pro-inflammatory cytokines that exacerbate spinal cord injuries (Brambilla et al., 2005; Brambilla et al., 2009). Conversely, elegant work involving ablation of reactive astrocytes has demonstrated that reactive astrocytes are crucial for withstanding insult and improving recovery after CNS trauma, ischemia and in EAE (Bush et al., 1999; Faulkner et al., 2004; Voskuhl et al., 2009). Together these findings demonstrate that reactive astrocytes can play both beneficial and detrimental roles and raise the question of whether there might be different subtypes of reactive astrocytes, elicited depending on the nature of the injury or disease, that differ in their functions.
To more fully characterize complex molecular changes that occur during reactive astrogliosis, we used Affymetrix GeneChip analysis to profile mRNA isolated from populations of quiescent and reactive astrocytes acutely isolated from healthy and injured brains. Gene profiling analysis gives a detailed snapshot of the transcriptional character of a particular cell or tissue state. Though often used to look for differences between healthy and injured tissue, the isolation of individual cell types permits a detailed understanding of which cells exhibit which changes in gene expression (Cahoy et al., 2008; Doyle et al., 2008; Heiman et al., 2008) and increases the resolution of differences between similar cells (Lobo et al., 2006; Sugino et al., 2006). We used expression profiling to investigate the changes that occur during reactive astrogliosis after two complementary injury models: ischemic stroke which causes extensive cell death, and systemic LPS injection, which causes neuroinflammation (Buttini and Boddeke, 1995; Qin et al., 2007) but leaves cell structure intact. We followed the progression of reactive astrogliosis over the course of the first week and found that it began with a burst of transcription that moderated over the course of a week. Stroke and LPS-induced neuroinflammation are likely to induce reactive gliosis through divergent mechanisms. Nonetheless, there were core changes in gene expression that are shared between the two types of reactive astrocytes. There were, however, as many gene expression differences that differed either qualitatively or quantitatively between these reactive astrocyte populations. These transcriptome datasets give insights into the nature and evolution of the reactive astrocyte response and provide a framework for considering the balance of activities provided by reactive astrocytes during injury and repair.
All experiments were carried out on mice from the transgenic mouse line, Tg(Aldh1l1-eGFP)OFC789Gsat (GENSAT project) maintained on a Swiss Webster strain background. Young adult, postnatal day 30-35 (P30-P35), mice were used for the initial FACS purifications of astrocytes. The in situ hybridization (ISH) and immunohistochemistry experiments were done on mice aged 1-3 months.
A single dose of 5 mg/kg of the endotoxin LPS from Escherichia coli O55:B55 (Sigma) dissolved in normal saline and diluted into endotoxin-free PBS or endotoxin free PBS control was administered by intraperitoneal (i.p.) injection to a mixture of male and female mice.
Transient ischemia was induced by occluding the middle cerebral artery (MCAO) in young adult male mice as previously described (Han et al., 2009). Occlusion was carried out for 1 hour followed by reperfusion. Control animals underwent a sham surgery during which no suture was inserted.
Live single cell suspensions for each control and experimental condition were made as follows. For the neuroinflammation model, the cortices and corpus callosum from two LPS-injected or saline-injected control animals were combined for each replicate. For the MCAO model, the ipsilateral cortex, corpus callosum, hippocampus and striatum from three mice that had undergone MCAO and both hemispheres from one mouse that had undergone sham control surgery were used for each replicate. Dissected tissue was treated as described in Cahoy et al., (2008). Briefly, dissected tissue was first diced to 1-3 mm and then digested and with 200 units of papain enzyme for 90 min at 34°C in bicarbonate buffered Earle’s balanced salt solution with 0.46% glucose, 26 mM sodium bicarbonate, 0.5 mM EDTA and 125 units/ml of DNase I (Worthington). Digested tissues were dissociated into single cell suspensions by gentle trituration, filtered through a 15 μm Nitex mesh (Tetko) to remove any remaining clumps of tissue, collected by centrifugation and resuspended in Dulbecco’s PBS containing 0.02% BSA and 125 U/ml DNase I and 1 μg/ml propidium iodide (PI) for fluorescence activated cell sorting (FACS).
Live astrocytes were isolated at room temperature by FACS at the Stanford Shared FACS Facility on the basis of their GFP expression on a BD Vantage running CellQuest software (for 1 day MCAO and sham samples) or a BD Aria II running BDFACSDiva software (for 1 day LPS and saline samples and 3 and 7 day MCAO and sham samples). Cell suspensions were sorted twice sequentially using forward light scatter and SSC to gate single cells followed by gating for GFP fluorescence in the absence of PI to select live astrocytes. The Flowjo software (Treestar) was used to analyze purity of the final astrocyte populations.
Qiagen Qiashredder and microeasy spin columns were used to lyse purified astrocyte populations and to purify their total RNA. The integrity and concentration of the isolated total RNA was confirmed by analysis on an Agilent Bioanalyzer. Approximately 20 ng total RNA from each sample was amplified and biotin-labeled using the Two-cycle target labeling kit (Affymetrix) and hybridized to the GeneChip® Mouse Genome 430 2.0 arrays (Affymetrix). Samples were processed by the Protein and Nucleic Acid Facility at the Stanford University School of Medicine. Samples were hybridized using an Affymetrix GeneChip® Hybridization Oven 640, processed using an Affymetrix GeneChip® Fluidics Station 450 and scanned using an Affymetrix GeneChip® Scanner 3000 7G. The data discussed in this publication have been deposited in NCBI’s Gene Expression Omnibus (Edgar et al., 2002) and are accessible through the GEO Series accession number GSE35338 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE35338).
The .cel files from the scanned GeneChip® arrays were processed with Arraystar 4.0 software (DNAstar) using RMA processing with quantile normalization. The .cel files from Cahoy et al, (2008) used in this study for comparison purposes were reprocessed along with the samples in this study to allow for direct comparison of signal intensities. The Significance Analysis of Microarrays (SAM) (Tusher et al., 2001) excel add-in, two class unpaired response type was used on the normalized expression values in log2 to generate lists of significantly changed genes between quiescent and reactive astrocytes. Due to differences in variability between datasets, an FDR of <1% was used for the MCAO stroke comparison, an FDR of <3.5% was used for the LPS neuroinflammation comparison and a FDR or <5.5% was used for the time course data using signed area option. These cutoffs maximized the ratio of significantly changed probe sets to the calculated false positives. Significantly changed probe sets were re-imported back into Arraystar 4.0 and consolidated to one probe set per gene for further analysis. Probe sets were filtered for a minimum average expression value in log2 of 8 in any astrocyte population to be included in the analysis. Significantly changed probe sets were analyzed for statistically enriched pathways using Ingenuity Pathway analysis (Ingenuity® Systems, www.ingenuity.com) and categorized for biological function using Amigo (The Gene Ontology Consortium, http://www.godatabase.org/cgibin/amigo/go.cgi.) accessed June-July, 2011.
Brain tissue for immunostaining was prepared as follows. Mice were deeply anesthetized with a ketamine/xylazine cocktail and transcardially perfused with DPBS followed by 4% paraformaldehyde (PFA). Brains were removed and immersion fixed for 2-3 hours on ice with 4% PFA. Fixed brains were cryoprotected by immersion in 30% sucrose in PBS overnight at 4°C. Cryoprotected brains were embedded in 2 parts O.C.T. compound (Tissue-Tek) to 1 part 30% sucrose in PBS. Embedded tissue was quick frozen and cut into 10 μm cryosections using a Leica cryostat. For immunohistochemistry, the following antibodies were used: rabbit polyclonal to GFAP at 1:2000 (Dako) and rabbit polyclonal to nestin at 1:1000 (Abcam). For these antibodies, cryosections on microscope slides were fixed for an additional 10 minutes in 4% PFA. No additional fixation was used for chicken polyclonal against vimentin used at 1:2000 (Novus Biologicals) and rabbit polyclonal against Iba1 used at 1:500 (Wako). The GFP fluorescence was enhanced using chicken polyclonal against GFP used at 1:1000 (Chemicon) where protocol allowed. Brain sections were rehydrated in PBS and then blocked with 10% goat serum in PBS with 0.2% triton x-100 for 1 hour at room temperature. All primary antibody incubations were done overnight at 4°C.
Immunostaining was visualized using Alexa secondary antibodies (Invitrogen). Quantification of fluorescence area was done in NIH ImageJ 1.45s (Rasband, W.S., ImageJ, U. S. National Institutes of Health, Bethesda, Maryland, USA, http://imagej.nih.gov/ij/, 1997-2011.) Images from control and injured sections were thresholded within each experimental replicate. Percentage of area with signal above threshold was measured in representative areas of cortex. For MCAO sections, area measured was immediately adjacent to the core as defined by loss of eGFP signal. The percentage area above threshold in each control and injured section was normalized to the average percentage area above threshold from all control sections within the experimental replicate to obtain fold increase in expression relative to control. The one-tailed T-test was used on the resulting ratios to assess significance.
For ISH, deeply anesthetized animals were perfused with DPBS, quick frozen in OCT and cut into 10 μm cryosections using a Leica cryostat. Colorimetric ISH was carried out using digoxygenin (dig) labeled RNA probes to target genes. Color development was done by anti-dig-AP Fab fragments from sheep (Roche) using NBT/BCIP (Roche). Fluorescent ISH was carried out using FITC-labeled and Dig-labeled RNA probes. Fluorescence development was done by anti-FITC-POD and anti-dig-POD Fab fragments (Roche) using FITC-tyramide and Cy3-tyramide amplification (Perkin Elmer).
We confirmed the induction of astrocyte reactive gliosis in two complementary brain injury models: focal ischemic stroke produced by transient MCAO and neuroinflammation induced by systemic LPS injection. One hr MCAO leads to destruction in the ipsilateral hemisphere of parts of cortex, striatum, and hippocampus (Han et al., 2009; Xiong et al., 2011). The core of the lesion is marked by extensive cell, including astrocyte, death (Liu et al., 1999) surrounded by the relatively intact, but stressed, tissue of the lesion penumbra in which astrocytes become reactive and subsequently form the glial scar (Kindy et al., 1992; Yamashita et al., 1996). Neuroinflammation was induced by a single i.p. injection of the bacterial endotoxin LPS. Although LPS itself largely fails to cross the blood brain barrier (Banks and Robinson, 2010), this treatment induces microglia activation in response to induction of inflammatory cytokines in the periphery (Buttini and Boddeke, 1995; Qin et al., 2007) that subsequently leads to astrocyte activation (Herx and Yong, 2001). Astrocytes, marked by eGFP expression driven by the Aldh1l1 promoter in Gensat Bac Aldh1L1-eGFP mice (Anthony and Heintz, 2007; Cahoy et al., 2008), showed little GFAP immunoreactivity in sections from control, healthy cortices (Fig. 1 A, I). By 1 day post-MCAO, increased GFAP immunoreactivity was seen in astrocytes in the MCAO penumbra of the ipsilateral cortex which persisted for at least 7 days post-MCAO (Fig. 1 A-D). eGFP expression was reduced in astrocytes in the lesion core at both 1 day post-MCAO (data not shown) and 7 days post-MCAO (starred areas in Fig. 1 D), consistent with rapid astrocyte death (Liu et al., 1999). By 1 day after injection, cortex from LPS-injected animals showed patches of increased GFAP immunoreactivity (Fig. 1 J arrowheads) which persisted for at least one week (Fig. 1 I-L). Particularly strong activation was seen in astrocytes near the pial surface (Fig. 1 L). The reactive astrocyte response fully resolved by 30 days post LPS injection (data not shown). There was no loss of eGFP expression in astrocytes from brains experiencing neuroinflammation in response to LPS consistent with lack of brain cell death in this inflammation model (Deng et al., 2003). The fold increase in astrogliosis was quantified by percentage of area with GFAP immunoreactivity above threshold after MCAO (Q) and LPS (S). Both MCAO and systemic LPS injection induced astrogliosis in the cortex as indicated by increased GFAP expression.
Astrocyte activation was concomitant with activation of microglia, the endogenous brain macrophages. In healthy cortex, Iba1 lightly labels resting microglia cell bodies and their thin, highly ramified processes (Fig. 1 E and M). At 1 day post-MCAO in the lesion penumbra (Fig. 1 F) and core (data not shown) microglia showed stronger immunoreactivity for Iba1, which increases upon macrophage activation, and thickening of their processes indicating a shift in state towards a more activated amoeboid morphology (Jonas et al., 2012). Since infiltration of peripheral macrophages into the brain is low for the first few days following MCAO (Schilling et al., 2003), Iba1 positive cells are likely to be mainly microglia. At 7 days post-MCAO, strong activation in the penumbra was evident (Fig. 1 G, H), and there were increased numbers of immune cells in the core of the lesion (starred areas in Fig. 1 H), largely reflecting immune cell infiltration from the periphery. Microglia activation was seen 1 day post-LPS-injection in the cortex (Fig. 1 N vs. M). Activation persisted for at least 7 days after LPS injection (Fig. 1 P). The fold increase in microglia activation was quantified by percentage of area with Iba1 immunoreactivity above threshold after MCAO (R) and LPS (T). Increased Iba1 expression and morphology changes indicated that microglia became activated in both models.
FACS was used to acutely isolate pure populations of astrocytes from mice aged P30-35 from control and injured Aldh1l1-eGFP brains on the basis of their astrocyte-restricted GFP expression using a simplified version of the protocol used to isolate astrocytes from S100b-eGFP mice (Cahoy et al., 2008). Astrocytes have reached their mature gene profiles by P30 (Cahoy et al., 2008), and the use of P30-35 mice significantly enhanced the yield of viable astrocytes compared to mice of older ages. Two consecutive rounds of sorting enriched astrocytes from 17.5% ± 4.4 (S.D.) of the starting cell suspension (Representative FACS plots in Fig. 2 A, B) to 98.8% ± 1.3 (S.D.) of live cells in the final isolated population (Representative FACS plots in Fig. 2 C, D). We were routinely able to isolate 50,000-100,000 live GFP positive astrocytes in the final cell population.
Reactive astrocytes were no more or less amenable to isolation than resting astrocytes. Despite the lowered expression of GFP in the astrocytes of the MCAO core lesion (Figure 1 D), there was no significant difference between percentage of GFP positive astrocytes present in starting cell suspensions made from MCAO injured and sham control brains (p = 0.19 unpaired two-tailed t-test). Likewise, there was no significant difference in percentage of GFP positive astrocytes in cell suspensions made from brains from LPS-injected and saline-injected mice (p = 0.29 unpaired two-tailed t-test). Astrocytes were 15.6% ± 5.4 (mean ± S.D.)(n=11) and 18.6% ± 3.3 (n=10) in MCAO and sham suspensions respectively and 17.4% ± 1.9 (n=5) and 20.2% ± 5.1 (n=4) in LPS-injected and saline-injected brain suspensions respectively. There were no significant differences between the final purity of MCAO astrocytes (98.5% ± 1.4% (S.D)) and sham control astrocytes (98.2% ± 1.4% (S.D)) (p = 0.56 unpaired two-tailed t-test), or between LPS astrocytes (99.8% ± 0.3% (S.D)) and saline control astrocytes (99.9% ± 0.1% (S.D)) (p = 0.32 unpaired two-tailed t-test). The healthy and injured brains yielded comparable populations of purified astrocytes.
We confirmed that we had isolated relatively pure populations of astrocytes, both quiescent and reactive, by semi-quantitative RT-PCR for the astrocyte marker, GFAP, and additional cell-type specific markers for oligodendrocyte lineage cells, neurons, endothelial cells and microglia (data not shown) and then through analysis of the subsequent Genechip expression levels for cell-type specific markers. We normalized the quiescent and reactive astrocyte Genechip expression files to the neuron and oligodendrocyte lineage cell expression profiles from previous work (Cahoy et al., 2008) as well as microglial profiles from the Aldh1l1-eGFP mice (J.L. Zamanian, B.A. Barres and R.G. Giffard unpublished observations). As expected for purified populations of astrocytes, Genechip analysis of the isolated quiescent and reactive astrocyte populations showed high expression of the astrocyte markers: Glt1, Aqp4, Connexin 30 (Cx30) as well as Aldh1l1, expression of which did not change between astrocytes isolated from healthy and injured brains (Fig. 2 E). The purified astrocyte populations expressed only low levels of markers specific for neurons: neurofilament, Syt1, Gabra1 and Snap25, (Fig. 2 G), microglia: Cd68, Ptprc, Itgam and Iba1, (Fig. 2 F) and oligodendrocytes: Mog, Sox10, connexin 47 (Cx47) and Mbp (Fig. 2 H). By Genechip expression comparison, the astrocyte populations were contaminated by neurons to 4.0% ± 0.7% (S.E.M.), by microglia to 1.6% ± 0.1% (S.E.M.) and by oligodendrocytes by 1.8% ± 0.8% (S.E.M.). Contamination levels were used to filter out confounding signals from the astrocyte dataset. We thus isolated highly pure populations of astrocytes by FACS from both healthy and injured mouse brains.
In order to confirm that we had isolated reactive astrocytes from injured brains, we next assessed changes in established markers of reactive astrocytes in the Genechip expression profiles. Since we isolated astrocytes on the basis of their astrocyte-specific GFP expression and not on the basis of a reactive astrocyte marker, the astrocyte populations isolated from the injured brains will be a mix of quiescent and reactive astrocytes (the majority of astrocytes isolated, however, were reactive, see below). Classic reactive astrocyte markers GFAP and vimentin (Vim) were strongly up-regulated in both the MCAO and LPS reactive astrocyte populations (Fig. 3 A). At 1 day, mRNAs for vimentin and GFAP were 7-fold increased in the MCAO reactive astrocyte population and 5-fold increased in the LPS reactive astrocyte population, indicating a similar level of activation between the two stresses. GFAP and vimentin expression continued to rise for 3 days after MCAO injury and persisted for at least one week. Nestin (Nes), another intermediate filament protein that is up-regulated in reactive astrocytes after stroke (Clarke et al., 1994; Duggal et al., 1997), was induced 7-fold at 1 day in the MCAO reactive astrocytes but was not induced in LPS reactive astrocytes. In contrast to GFAP and vimentin, induction of nestin expression does not persist and has returned to near baseline by 7 days after MCAO injury. A fourth marker, tenascin c, an extracellular matrix protein secreted by reactive astrocytes (Laywell et al., 1992), was induced only in reactive astrocytes from the MCAO model (data not shown). The identification of expression changes in well-established reactivity markers in both MCAO and LPS astrocyte populations confirms that using these methods we could successfully identify expression changes indicative of reactive astrocytes, and also provides a clear indication that astrogliosis differs depending on the nature of the inducing stimulus.
We used immunohistochemistry to confirm these similar and divergent gene expression changes in reactive astrocytes identified by Genechip expression profiling. Vimentin immunoreactivity is normally very low and restricted to the pial layer in control brain sections (Fig. 3 J). One day after MCAO, vimentin immunoreactivity was modestly increased in penumbral astrocytes (Fig. 3 C). By 7 days after MCAO, vimentin was strongly expressed in the astrocytes in the penumbra as seen by co-localization of vimentin with GFP positive astrocytes (Fig. 3 D, E). Induction of vimentin protein expression after LPS-injection in astrocytes was seen most clearly in astrocytes near the pial layer at 7 days (Fig. 3 K-M). No nestin protein was seen in healthy cortex (Fig. 3 F, N). Despite induction of nestin transcription by 1 day after stroke, little protein expression was seen by immunostaining at that time point (Fig. 3 G). Strong nestin immunoreactivity in astrocytes in the penumbra was observed by 7 days after MCAO (Fig. 3 H, I). Nestin expressing reactive astrocytes were less widespread than GFAP expressing astrocytes, restricted to those astrocytes closest to the lesion core (asterisks throughout the figure) and absent in more distal regions. In contrast, GFAP expression at 7 days post MCAO was found in astrocytes more distal to the lesion (Fig. 1 D). As predicted from the Genechip expression profiling, no nestin immunoreactivity was seen in the cortex of LPS-injected animals (Fig. 3 O-Q). The expression of established markers of reactive astrocytes is therefore heterogeneous in marker composition, localization and the complement of markers expressed.
Having established that we had isolated purified populations of reactive astrocytes and that we could use Genechip expression profiling to identify gene expression differences between quiescent and reactive astrocyte populations, we conducted a comparison analysis between our healthy and stressed astrocyte populations to more thoroughly characterize the gene expression changes observed in reactive astrocytes. Expression in log2 of all 45037 probe sets on the Affymetrix GeneChip® Mouse Genome 430 2.0 arrays are represented on scatter plots comparing astrocytes from LPS-injected mice (LPS reactive astrocytes) to astrocytes from saline-injected animals (saline astrocytes; Fig. 4 A), and astrocytes from mice that had undergone MCAO (MCAO reactive astrocytes) to astrocytes from mice that had undergone the sham surgery (sham astrocytes)(Fig. 4 B). Gene expression in LPS and MCAO reactive astrocytes differed to a similar degree when compared to their control populations. The R2 value for best fit to a straight line was 0.9686 for LPS vs. saline astrocytes and 0.9566 for stroke vs. sham astrocytes. The scatter plots demonstrate that the vast majority of expression changes (4-fold cutoff), 206 of 220 genes for MCAO and 113 of 116 for LPS, involved induction of gene expression.
We identified 263 individual genes whose expression levels are significantly induced in astrocytes at least 4-fold at 1 day following injury in reactive astrocytes: 206 by MCAO and 113 by LPS. The heat map generated by cluster analysis of the quiescent (n=8) and reactive (n=10) astrocyte population replicates using the identified reactive astrocyte genes revealed that injured astrocyte populations fall into distinct groups depending on how the astrocytes were made reactive (Fig. 4 B). Clustering using the top 1057 genes significantly changed > 2-fold in MCAO and/or LPS reactive astrocytes gave the same clustering result (data not shown). The corresponding quiescent astrocyte population replicates from the saline-injected and sham operated animals were clustered away from both sets of reactive astrocyte replicates and interspersed with each other. Using the identified astrogliosis genes, cluster analysis demonstrates that different injuries produce different patters of reactive astrocyte gene expression.
To validate the Genechip expression profiling data, we chose two potential reactive astrocyte markers that were among the most highly expressed genes induced in both reactive astrocyte populations. Lcn2, a secreted lipophilic protein that is induced after infection and that limits bacterial growth by sequestering bacterial iron sidephores (Goetz et al., 2002; Flo et al., 2004) and which was recently implicated in astrocyte reactive gliosis (Lee et al., 2009; Chia et al., 2011), was induced 228-fold and 355-fold in MCAO and LPS reactive astrocytes respectively relative to their respective control astrocyte populations 1 day after treatment. Serpina3n, a secreted peptidase inhibitor whose expression is induced by inflammation and nerve injury (Takamiya et al., 2002; Gesase and Kiyama, 2007) was induced 9.1-fold in MCAO reactive astrocytes and 30-fold in LPS reactive astrocytes 1 day after injury. ISH confirmed that Lcn2 and Serpina3n were induced by injury in astrocytes based on the stellate morphology of many of the stained cells (arrowheads) in Figure 4 and co-localization with the astrocyte glutamate transporter, Glast, in Figure 12. Lcn2 and Serpina3n could not be detected in healthy brain sections (Fig. 4 D-G, zoom in H-L). Lcn2 (Fig. 4 M, zoom in Q) and Serpina3n (Fig. 4 O, zoom in S) were up-regulated in astrocytes (arrowheads in Fig. 4 Q, S) 1 day after LPS treatment in the cortex and throughout the brain (Fig. 4 M, O). From the gene expression profiles, Serpina3n up-regulation was specific to astrocytes after both LPS-induced neuroinflammation and MCAO (J.L.Z., B.A.B. and R.G.G. unpublished observations). Lcn2 was strongly induced after LPS injury not only in astrocytes, but also in endothelial cells (J.L.Z., B.A.B. and R.G.G. unpublished observations) and as seen by Ip et., al (2011), strongly in choroid plexus (Fig. 4 H and Marques et al., (2008)) and to a lesser extent in microglia (J.L.Z., B.A.B. and R.G.G. unpublished observations) and Ip et al., (2011). 1 day after MCAO, Lcn2 was induced specifically in astrocytes (Fig. 4 R arrowheads) in the penumbra (Fig. 4 N, zoom in R) and also in endothelial cells (Fig. 12 H, white arrowhead). Induction of Serpina3n expression was more widespread, extending further than Lcn2 from the lesion (Fig. 4 P, zoom in T). Thus, Lcn2 and Serpina3n gene induction are both markers of the early phase of astrocyte reactive gliosis in both models.
In order to study how reactive astrocyte gene expression changes over time, we isolated astrocytes 3 days (n=3) and 7 days (n=3) after MCAO. Using S.A.M. (Tusher et al., 2001), we identified the top 317 genes that changed significantly over time. We used cluster analysis to separate the 317 time courses of gene expression into 6 groups (Fig. 5 A). Relative expression in log2 is shown for time points sham, 1 day, 3 days and 7 days after MCAO. Group 1 contains 37 genes that were not induced at 1 day after MCAO, but were induced at 3 days, and which were moderating their expression by 7 days. Many genes suggestive of proliferation are present in this group including late phase cyclins b1 and b2 (Ccnb1 and Ccnb2), Cdk1, Top2a and the proliferation marker Ki67 (Figs. 5 A and and6).6). Group 2 contains 44 genes whose expression was induced at 1 day, increased further at 3 days, but was decreasing by 7 days after MCAO. This group contains the classic reactive gliosis marker, vimentin (see also Fig. 3 A), galectins Lgals3 and Lgals1, and osteopontin (Spp1). The largest cluster, group 3, contains 135 genes including many of the most highly induced genes by fold induction. These genes were highly up-regulated at 1 day, were decreasing by 3 days, and continued down, but remained elevated by 7 days after MCAO. Lcn2, Serpina3n, tweak receptor (Tnfrsf12a), S1pr3, and all Ptx3 fall into this group (see also Fig. 5 B). The 14 group 4 genes were up more modestly 1 day after MCAO, stayed elevated at 3 days and decreased by 7 days. Group 5 contains 22 genes that were increased at 1 day, and remained elevated at 3 and 7 days after MCAO. The chemokines, CXCL1, CXCL2 and CXCL10 (see also Fig. 5 C), as well as universal reactive gliosis marker, GFAP, fell into this group. Group 6 contains 65 genes that were induced at 1 day after MCAO, but were rapidly decreased over 7 days to baseline or below. Bdnf, the oncostatin M receptor (Osmr), and transcription factor tumor suppressor klf6 fell into this category. The expression of genes with diverse functions was rapidly induced and moderated during reactive gliosis. Genes involved in adhesion, ECM modification, immune response and the neurotrophic cytokines all followed this trend (Fig. 5 B). Chemokines were a major class of genes that were stably induced (Fig. 5 C), retaining high expression even out to 30 days after MCAO (data not shown). Even within this class of genes, there was variation of expression course with the chemokine for monocytes, CCL2, following the rapidly moderating group 3 expression pattern while other cytokines, CXCL1, CXCL2, and CXCL10 (Cartier et al., 2005) remained elevated as part of a group 5 expression pattern. Overall, gene expression profiling of reactive astrocytes reveals a dramatic burst of induced expression which is rapidly moderated.
We confirmed the rapid induction and reduction of expression by ISH on tissue sections from brain 1, 3 and 7 days after MCAO and LPS (Fig. 5 D). Consistent with the Genechip expression values, Lcn2 had the fastest time course for reduction in expression. Its expression was clearly reduced by 3 days after MCAO and was below detectable limits at 7 days (Fig. 5 D, top row and Fig. 5 B). Expression persisted for longer in the LPS tissue, present at 3 days, but was nearly absent by 7 days (Fig. 5 D, second row). The rapid induction and decrease in gene expression of lcn2 in astrocytes was similar to the time course of induction and repression in choroid plexus after LPS (Marques et al., 2008). Consistent with the expression profiling result (Fig. 5 B), induction of Serpina3n expression persisted for longer, for at least 3 days after LPS (Fig. 5 D, fourth row) and for at least 7 days after MCAO (Fig. 5 D, third row and Fig. 5 B) as seen by ISH in sections adjacent to those used for Lcn2. ISH on injured brain sections confirmed the time course of induction and moderation of expression of reactive astrogliosis genes seen in the gene expression profiles.
Whether reactive astrocytes proliferate after injury or simply undergo hypertrophy as long been controversial (Sofroniew, 2009). We recently analyzed this question with BrdU labeling in these Aldh1l1-GFP mice and found significant numbers of astrocytes colabeling with BrdU on day 2 after MCAO, with only modest additional numbers of cells if labeling was extended through day 6 (Barreto et al., 2011). We analyzed the reactive astrocyte expression profiles for cell cycle genes and markers of proliferation (Fig. 6). Early phase cyclin D (Ccnd1) was induced 4 to 5-fold and growth arrest gene, Gas1, was repressed 50% by 1 day after injury in MCAO reactive astrocytes. Many cell cycle genes including the late phase cyclin B (Ccnb1) and cyclin dependent kinase, Cdk1, were not induced at 1 day after MCAO but were elevated 3 to 4-fold in MCAO reactive astrocytes 3 days later. By 7 days post-MCAO, the cell cycle genes were decreasing towards their baseline expression, consistent with our prior BrdU labeling study (Barreto et al., 2011). The expression of cell proliferation marker, Ki67, was induced about 4-fold at 3 days and was returning towards baseline by 7 days after MCAO. Similar changes in LPS reactive astrocyte gene expression occurred at 1 day after injection. Cyclin D1 expression was induced 3.4-fold and gas1 expression was decreased by 40%. These data support previous findings that reactive astrocytes divide with a brief delay after injury, but that this proliferation is limited.
Hierarchical clustering of the quiescent and reactive astrocyte populations by Genechip expression revealed that reactive astrocytes are separated into groups depending on whether their activation was induced by MCAO or LPS (Fig. 4 B). We further analyzed the similarities and differences between the two types of reactive astrocytes. 56 of the >4-fold induced reactive gliosis genes representing 50% of the genes induced by LPS and 25% of the genes induced by MCAO were shared between the two types of reactive astrocytes (Fig. 7 A). We identified 57 genes whose expression were induced significantly at >4-fold in LPS reactive astrocytes but not MCAO reactive astrocytes and 150 genes whose expression were induced significantly at >4-fold in MCAO reactive astrocytes but not LPS reactive astrocytes. In LPS reactive astrocyte genes (At a 2-fold cutoff for both, the Venn diagram was similar to that for 4-fold, with 166 genes induced, representing 22% of genes (766) induced by MCAO and 57% of the genes (291) induced by LPS (data not shown). Some of the genes excluded from one reactive astrocyte gene set at a cutoff level were induced to a lesser degree. 90 of the 113 genes (80%) that were induced >4-fold induced by LPS are induced by >2-fold by stroke and 82 of the 220 (37%) of the >4-fold induced by stroke were induced by >2-fold by LPS indicating that reactive astrocyte gene induction by individual injuries varies in both gene representation and fold induction. The Top 50 gene changes with fold induction are listed in Table 1, for MCAO reactive astrocytes, and Table 2, for LPS reactive astrocytes.
We analyzed the identified reactive astrogliosis genes using gene ontology (GO) classification (The Gene Ontology Consortium, http://www.godatabase.org/cgibin/amigo/go.cgi.). The categorization by class and/or biological process for genes >4-fold induced is shown in pie charts for MCAO (Fig. 7 B) and LPS (Fig. 7 C). The gene constituents of each category induced >4-fold in MCAO and LPS reactive astrocytes are listed in Table 3. Proteins involved in extracellular matrix modification and adhesion were the largest class for both types of reactive astrocytes. Constituents of this class (Fig. 8) included not only ECM proteins such as collagen (Col12a1, Col6a1) and versican (Vcan), but proteins that interact with the ECM, such as thrombospondin (Thbs1) and fibulin 5 (Fbln5), proteins involved in cell adhesion, such as Cd44 and neurofascin (Nfasc), and enzymes that modify the carbohydrate side chains of extracellular molecules, such as Ggta1 and Galntl2. The dendrogram resulting from hierarchical clustering of extracellular matrix and adhesion proteins demonstrates that MCAO and stroke reactive astrocytes, while both showing gene induction strongly suggestive of modification of the extracellular space, differed greatly in the specifics of the changes (Fig. 8 A). Whereas both types of reactive astrocytes exhibited a large array of induced genes, the degree to which any gene is induced depended on the stimulus (Fig. 8 B). Prominently, collagen (Col6a1, Col12a1) and versican (Vcan) were more strongly induced by MCAO as might be expected to seal off the dying tissue and form the glial scar. Conversely, other genes in the class, Fbln5 and Amigo2, were more strongly up-regulated in LPS reactive astrocytes.
Proteins involved in transport, especially of metal ions and immune response also figure prominently. In fact, a full 50% of all LPS and 25% of MCAO reactive astrocyte genes had a GO categorization that involved them in immune response. Cytokine signaling in particular was induced in both MCAO and LPS reactive astrocytes. Even within this gene class, differences in induction were clear (Fig. 9 A). The C-X-C class of chemokines were induced to a similar degree by stroke and neuroinflammation. For instance, CXCL1, on average, was induced ~5-fold in both types of astrocytes, CXCL2 ~8-fold, CXCL10 11-15-fold. Alternatively, CCL2, a macrophage chemokine, was more prominently up-regulated by MCAO reactive astrocytes, 8-fold vs. 2 fold by LPS. The neurotrophic cytokines, LIF and CLCF1 (Bauer et al., 2007) were greatly induced in stroke reactive astrocytes, but only marginally induced in LPS reactive astrocytes. IL6, another cytokine known to be important in stroke, with both beneficial and deleterious effects depending on timing and context (Gadient and Otten, 1997; Monje et al., 2003; Suzuki et al., 2009; Voloboueva et al., 2010), also follows this pattern. The dendrogram made from hierarchical clustering (Fig. 9 B) shows that part of this difference was due to variation in astrocyte response during LPS-induced neuroinflammation.
Certain categories of genes were more prominently represented in one type of reactive astrocyte. Increased metabolic activity, cell cycle genes and transcription factors were prominent categories for MCAO reactive astrocytes (Fig. 7 B) but not LPS reactive astrocytes (Fig. 7 C). In contrast, the antigen presentation pathway, complement pathway and response to interferon figured more prominently in the LPS reactive astrocytes (Fig. 7 C and Fig. 10) than in MCAO reactive astrocytes. After injury, genes within the antigen presentation pathway (York and Rock, 1996) including class I MHC molecules (H2-D1, H2-K1, H2-T10) and the Tapbp and B2m genes utilized in peptide processing and MHC association were up-regulated by 2- to 30-fold in LPS reactive astrocytes, but only by 10% to 3-fold in MCAO reactive astrocytes (Fig. 10 A). Hierarchical clustering using probe sets for genes in the antigen presentation pathway showed that all but one replicate of MCAO reactive astrocytes cluster with quiescent astrocyte populations. One replicate showed strong induction of part of this pathway, again demonstrating variability in reactive gliosis even within an injury model (Fig. 10 B). Interestingly, the genes of the initiating part of the complement cascade, C1r, C1s, C3, and C4B as well as complement inhibitor, Serping1 (Gasque, 2004), were all induced 4.5- to 34-fold (15-fold on average) in LPS reactive astrocytes, but only 2.5- to 7-fold (4-fold on average) in stroke reactive astrocytes (Fig. 10 C). Hierarchical clustering of complement pathway genes showed that 4 of our 5 LPS reactive astrocyte replicates cluster apart from all other astrocyte populations (Fig. 10 D). Although MCAO and neuroinflammation both induced reactive gliosis based on classical markers, the characteristics of the activation greatly differed by inducing signal.
We also analyzed the reactive astrocyte genes using the canonical pathways analysis by IPA (Ingenuity® Systems, www.ingenuity.com). In order to increase capture of induced pathways, we used the 2-fold cutoff gene set list from each reactive astrocyte subtype for analysis. The top 20 significant pathways, with significance value, are shown in Table 4. Constituent members of each pathway induced in the reactive astrocytes are also listed. IPA supports the finding that, although there are some pathways that are induced in both types of reactive astrocytes, astrogliosis is qualitatively different between the two inducing injuries. The acute phase signaling and hepatic stellate cell activation, two pathways indicative of cellular activation are induced in both reactive astrocyte populations. Prominently, the IL6 and IL10 signaling pathways, as well as aminosugar metabolism, which suggests increased metabolism, are enhanced in the MCAO reactive astrocytes. Conversely, the antigen presentation, complement and response to interferon pathways are significantly induced in LPS reactive astrogliosis. IPA analysis supports the idea that astrogliosis differs depending on the inducing stimulus.
Having discovered by Genechip expression profiling that the character of reactive astrogliosis is different in response to different stimuli, we confirmed this using ISH on sections from injured brain tissue. We chose H2-D1, a class I MHC molecule, and Serping1, a C1q inhibitor that is a critical regulator of complement activity (Cicardi et al., 2005), as representatives of the antigen presentation and complement pathways that expression profiling revealed were more strongly induced in astrocytes by LPS than by MCAO. H2-D1 is expressed at low levels in the healthy brain by all cell types in our expression profiling datasets. After injury, the most dynamically induced H2-D1 probe set was induced 30-fold by LPS, but only 3-fold by MCAO (Fig. 10 A). ISH shows that H2-D1 was expressed in sparse cells in brain sections from healthy brain from saline injected animals (Fig. 11 A). After MCAO, H2-D1 expressing cells were present at higher density (Fig. 11 B), but LPS increased the density to a still greater degree (Fig. 11 C). Expression was observed not only in astrocytes, but based on morphology, other cell types. This is consistent with the recent findings that MCAO significantly induces H2-D1 and H2-K1 in neurons (Adelson et al., 2012). Genechip expression profiling also identified the complement pathway as being induced to a greater extent in LPS reactive astrocytes. Serping1 is expressed at very low levels in the Genechip expression profiles for all cell types. After injury, it was induced 6.5-fold in reactive astrocytes after MCAO and 34-fold in reactive astrocytes after LPS (Fig. 10 C). ISH detected no expression in the cortex of healthy brain from a saline-injected animal (Fig. 11 D). Very sparse cells expressed Serping1 after MCAO (Fig. 11 E). After LPS, however, astrocytes throughout the cortex expressed Serping1 (Fig. 11 F).
Of the large number of other genes more highly expressed by astrocytes after MCAO than after LPS, many are involved in immune response including the opsinin Ptx3 and signaling receptors for tweak and S1P. In our gene expression profiles Ptx3 was induced 44-fold after MCAO, but only 5.5-fold after LPS; tweak receptor (Tnfrsf12a) was induced by 14-fold after MCAO but only 3.3-fold by LPS; and S1P receptor 3 (S1PR3) was induced 46-fold after MCAO but only 6.4-fold by LPS. ISH confirmed no detectable expression of Ptx3, tweak receptor, and S1PR3 in the cortex of animals that had undergone a sham surgery (Fig. 11 G, J, M) and demonstrated expression in the penumbra (Fig. 11 H, K, N) 1 day after MCAO. Little to no expression of these markers is seen in the cortex of LPS treated mice (Fig. I, L, O). The ISH studies, thus, confirmed that MCAO and LPS induced different subtypes of reactive astrocytes.
Given the identified heterogeneity of reactive gliosis between stimuli, we also wondered whether there would be heterogeneity in astrocyte phenotype even within the response to a single inducing stimuli. To investigate this question, we used our newly identified markers of reactive astrogliosis to investigate the uniformity of the reactive astrocyte response. We used double fluorescent ISH to look at the extent and distribution of reactive astrocytes in the cortex after LPS injection and penumbra after MCAO. An ISH probe to the GLAST astrocytic glutamate transporter was used to mark the astrocytes in green in sections from healthy brain (Fig. 12 A, C, E, G, I, K) and 1 day after injury (Fig. 12 B, D, F, H, J, L). As expected based on the colorimetric ISH results, astrocytes throughout the cortex expressed Lcn2 (Fig. 12 B) and Serpina3n (Fig. 12 F), both shown in red, after LPS treatment but not in healthy cortex from saline-injected (Fig. 12 A, E) and sham-operated mice (Fig. 12 C, G). Astrocytes in the MCAO lesion penumbra also express Lcn2 (Fig. 12 D) and Serpina3n (Fig. 12 H). Endothelial cells in this region also express Lcn2 (Fig. 12 H white arrowhead). Reactive astrocytes, as defined by Lcn2 or Serpina3n expression (red arrows in Fig. 12 B, D, F, H), were interspersed with adjacent quiescent and lightly reactive astrocytes (red arrowheads in Fig. 12 B, D, F, H) demonstrating that neighboring astrocytes can differ in reactivity.
A major finding from previous expression profiling of astrocytes is that cultured neonatal astrocytes produced by the McCarthy De Vellis method (McCarthy and de Vellis, 1980)(MD-astrocytes) are highly dissimilar to mature astrocytes acutely purified from the healthy brain (astrocytes in vivo) (Cahoy et al., 2008; Foo et al., 2011). We were surprised to notice that many of our newly identified reactive astrocyte markers, including the markers, Lcn2 and Serpina3n, common to both inflammation and MCAO injury stimulated astrocytes, were also expressed at a much higher level in MD-astrocytes than in acutely-purified postnatal and adult astrocytes in vivo isolated by FACS (Cahoy et al., 2008; Foo et al., 2011) and as determined by the Bac-Trap method (Doyle et al., 2008). In the expression profiling data from Cahoy et al., 2008, Lcn2 was expressed 593-fold higher and Serpina3n 11-fold higher in MD-astrocytes than in astrocytes in vivo. A full 60% (126 of 206) of stroke reactive astrocyte genes were >4-fold more highly expressed by MD astrocytes than by astrocytes in vivo, while 47% (53 of 113) of LPS reactive astrocyte genes were >4 fold more highly expressed by MD-astrocytes than by astrocytes in vivo (Fig. 13 A). Hierarchical clustering using our most highly induced reactive astrocyte genes shows that MD-astrocyte replicates cluster with MCAO reactive astrocytes and away from both quiescent astrocytes and LPS reactive astrocytes (Fig. 13 B). Using subsets of reactive astrocyte genes, MD-astrocytes cluster with MCAO reactive astrocytes and/or LPS reactive astrocytes depending on the specific pathways analyzed (Fig. 13 C). When using ECM binding and adhesion genes, cytokine signaling molecules, and IL6 signaling pathway genes, MD astrocytes clustered with MCAO reactive astrocytes. When using complement pathway genes, MD-astrocytes clustered with LPS reactive astrocytes. When using antigen presentation pathway genes, MD-astrocytes clustered with MCAO reactive astrocytes and quiescent astrocytes. When using peptidase inhibitors (closer to MCAO reactive astrocytes), transporters/channels (closer to LPS reactive astrocytes), acute phase signaling (closer to MCAO reactive astrocytes) and interferon response (closer to LPS reactive astrocytes), MD-astrocytes clustered with both LPS and MCAO reactive astrocytes and away from quiescent astrocytes. These expression profiling analyses demonstrate the MD-astrocytes share many of the characteristics of reactive astrocytes.
The role of reactive astrocytes has long been mysterious. Expression profiling offers a powerful approach to understanding the molecular changes that characterize reactive astrocytes, which in turn provide a starting point for generating new hypotheses about their function. Our reactive astrocyte transcriptomes show that reactive astrocytes undergo extensive changes in gene expression relative to quiescent astrocytes. We identified over 1000 genes from across a broad spectrum of biological processes whose levels of expression were induced at least 2-fold. The extensive shift in gene expression supports the idea that reactive astrogliosis is a highly complex change in astrocyte cell state. The fold changes are dramatic, with 260 genes having >4-fold induction of expression. Several of the most strongly expressed genes were induced over a hundred fold, with several dozen genes over 10-fold induced. In brain tissue that had been injured by MCAO or activated by LPS exposure, most or all cell types in the brain undergo dramatic gene expression changes (J.L.Z., B.A.B., and R.G.G unpublished observations). Separation of reactive astrocytes enables analysis of their relative contribution to the expression changes in CNS tissue. Although our protocol was not designed to separate reactive astrocytes from quiescent astrocytes, our in situ hybridization studies demonstrated that the majority of the astrocytes in the LPS treated cortex and MCAO penumbra were in an activated state and contamination with quiescent astrocytes would only lead to underestimation of the observed changes in gene expression. The transcriptomes that we generated are therefore highly representative of the reactive astrocyte state and offer new insight into their phenotypes and possible functions.
One well studied role for reactive astrocytes is the modification of the extracellular milieu. Reactive astrocyte modification of the extracellular matrix can inhibit axon regeneration, but also limit spread of damage after injury by formation of the glial scar. In accordance with this known activity, extracellular matrix genes were amongst the most highly represented in our induced gene expression profiles. In addition, many of the induced peptidases and peptidase inhibitors are secreted and will also influence the extracellular space. Up-regulation of intermediate filament proteins such as GFAP, vimentin, and nestin, and changes in actin cytoskeleton binding proteins, are also a well represented class in the reactive astrocyte transcriptome as would be expected based on the well described hypertrophy and morphological alterations of these cells. Many cytokines that are known to powerfully control the immune system were also represented in our gene profiles suggesting that reactive astrocytes may play particularly important roles in controlling and interacting with immune cells. Interestingly, many of the reactive astrocyte genes are well described genes that are induced after injury and in wound healing in peripheral tissues suggesting important commonalities between the way the brain and other tissue systems respond to injury. For instance, AHNAK, Anxa2, and s100a10 were highly up-regulated and have recently been described to form a complex that promotes the sealing of damaged membranes (Lennon et al., 2003; Rezvanpour et al., 2011).
The wide array of genes induced demonstrates the complexity of reactive astrocyte influence on these biological processes and suggest new targets for manipulating reactive astrocytes after injury. Transporters, especially those that regulate metal ion homeostasis, and channels whose activities regulate intercellular ionic homeostasis are also well represented. One day after injury, there were surprisingly few changes in transcription factors. No transcription factor was >4-fold induced in LPS reactive astrocytes. One of the few transcription factors identified as strongly induced in the MCAO reactive astrocytes was Klf6, which has recently been identified as being induced in astrocytes in an epilepsy injury model (Jeong et al., 2011). Signal transduction changes, which are largely post-translational modification, are largely not picked up by gene expression profiling, but one central player in reactive gliosis, Stat3 (Sriram et al., 2004; Okada et al., 2006; Herrmann et al., 2008), is increased 2-fold in both types of reactive astrocytes and exhibits a sustained 50% elevation over the course of a week. Of great interest, many signaling receptors were very strongly upregulated in reactive astrocytes including the oncostatin receptor Osmr, S1PR3, Tnfrsf12a, and c-Met. These signaling pathways are thus new candidates to investigate for their roles in inducing or maintaining reactive astrogliosis.
Our findings also demonstrate that neonatal astrocyte cultures (MD-astrocytes) prepared by the method of McCarthy and de Vellis (1980) share many aspects of the reactive astrocyte phenotype, as shown by comparison of transcriptomes. They express over 100 genes not normally expressed by mature astrocytes in vivo (Cahoy et al., 2008; Foo et al., 2011), but do express 56% of the genes that are strongly induced in reactive astrocytes at elevated levels. These MD-astrocytes are also very highly proliferative compared with prospectively isolated mature brain astrocytes and may thus represent glial progenitor cells in a reactive state.
Although prolonged GFAP expression is widely reported after injury, and particularly in scar forming astrocytes (Sofroniew, 2009), it has not been clear to what extent the reactive astrocyte state is stable, transient or represents multiple phenotypes. To get at this question, we examined reactive astrocyte gene expression over 1 week. We found that during one week after MCAO or LPS exposure, broad classes of genes with roles in many different processes all demonstrated a strong burst of increased gene expression that was subsequently rapidly moderated. Neurotrophic cytokines and growth factors (Bauer et al., 2007) are among those genes rapidly down-regulated. Conversely, pro-inflammatory cytokines are among the most persistently elevated genes. The powerful macrophage chemoattractant chemokine, CCL2, is rapidly induced and then down-regulated. Recruitment of neutrophils and macrophages may be beneficial to clear bacteria and debris from dying cells. An interesting question raised by this observation is whether drugs that alter the time course of these reactive changes could be beneficial.
Reactive astrocytes are frequently identified by their strong expression of the intermediate filament protein, GFAP, considered a universal marker for reactive astrocytes. However GFAP is also normally expressed by many astrocytes and it would be useful for analysis of human brain injuries and diseases to have reactive astrocyte specific markers that do not identify quiescent astrocytes. Our gene expression profiling shows that there is a core group of genes, including GFAP, that is induced in both of our populations of reactive astrocytes. We identified the acute phase protein, Lcn2, and proteinase inhibitor, Serpina3n, as markers common to reactive astrocytes induced by MCAO and LPS exposure. Their expression has previously been shown to be induced in several other injuries including inflammation and excitotoxicity for Lcn2 (Chia et al., 2011), and after nerve injury (Gesase and Kiyama, 2007) and amyotrophic lateral sclerosis (Fukada et al., 2007) for Serpina3n. Lcn2 may be a marker of the earliest part of the reactive astrogliosis response as its expression is rapidly down-regulated after initial induction. Serpina3n may be a useful marker of the more persistent reactive gliosis response as its expression remains relatively elevated at least over the course of a week. Given their expression after diverse injuries, they are possible universal markers for reactive astrocytes. Ptx3, Tgm1, and Cd109 are additional potential markers of reactive astrocytes. Like Serpina3n, the expression of these genes in our expression profile dataset is specific to astrocytes. Although they have low induced expression in LPS reactive astrocytes, they are induced to a high degree in MCAO reactive astrocytes and may be markers of this reactive astrocyte subtype.
It has been unclear whether there is one primary type of reactive astrocyte or whether there are multiple different reactive phenotypes depending on response to different acute and chronic diseases and injuries. Despite a core set of gene expression changes, we found that reactive astrocyte gene expression differed dramatically depending on whether the inducing stimulus was MCAO or LPS. Out of a total of 263 reactive glial genes that we identified, 150 of these were preferentially expressed by MCAO reactive astrocytes, 57 were preferentially expressed by LPS reactive astrocytes, and only 56 were shared. For instance, although MCAO reactive astrocytes and LPS reactive astrocytes each had extensive induction of genes involved in modifying the extracellular space and interaction with the immune system, the particular genes within each of these classes differed. This is likely due to differences in the inducing signals for each of these types of reactive astrocytes. Whereas LPS activates TLR4 receptors on microglia and macrophages (Buttini and Boddeke, 1995; Poltorak et al., 1998; Olson and Miller, 2004) which in turn secrete signals that induce reactive astrogliosis, MCAO exposes all brain cells within its territory to ischemia and causes direct cell death and breakdown of the blood brain barrier allowing serum influx. In MCAO, reactive astrogliosis is likely triggered by a combination of signals such as signals from dying cells, hypoxia, acidosis, serum factors, direct trauma, and immune system signals. MD-astrocytes, which are grown in serum, express many of the MCAO reactive astrocyte genes (Cahoy et al., 2008; Foo et al., 2011). Different inducing signals thus induce different phenotypes of reactive astrocytes.
Our findings clearly demonstrate that there are distinct subtypes of reactive astrocytes and raise important new questions. How many types of reactive astrocytes are there and how are their functions distinguished? With immune cells, the concept of functionally different subclasses of cell types such as helper and suppressor lymphocytes or M1 and multiple M2 macrophages is well established. Analogously, it will be important in future work to begin to understand the roles of the different types of reactive astrocytes. We identified many reactive astrocyte upregulated genes that were unique to the MCAO subtype or the LPS subtype of reactive astrocytes. Ptx3, S1pr3 and tweak receptor are markers for the MCAO subtype of reactive astrocytes and H2-D1 and Serping1 are markers for the LPS subtype of reactive astrocytes. These genes will be useful as new markers of distinct reactive astrocyte subtypes so that these subtypes can now be identified and explored in different brain injuries and diseases.
A longstanding question concerning reactive astrocytes is to determine whether they are beneficial or detrimental. The pioneering studies of Sofroniew and colleagues has established that scar forming astrocytes induced by ischemia play a critical role in repairing the blood-brain barrier, limiting immune cell influx, and minimizing neuron death (Sofroniew, 2009). In analogy with the immune system, the gene profiles we have elucidated, raise the hypothesis that reactive astrocytes can be polarized in a relatively protective orientation or instead in a relatively destructive orientation. Consistent with the work of Sofroniew and colleagues, our gene profiles strongly suggest that MCAO reactive astrocytes are protective. They express high levels of neurotrophic factors and cytokines including CLCF1, LIF, and IL6, and thrombospondins that may help repair and rebuild lost synapses (Christopherson et al., 2005; Eroglu et al., 2009), In contrast, LPS reactive astrocytes more strongly up-regulate the initial part of the classical complement cascade—C1r, C1s, C3, and C4--that is involved in synapse pruning during development (Stevens et al., 2007) and is hypothesized to drive the synapse loss which drives neuron loss in neurodegenerative disease (Stevens et al., 2007; Stephan et al., in press). LPS injection itself is sufficient to causes degeneration of the dopaminergic neurons that mimics Parkinson’s disease (Herrera et al., 2000; Qin et al., 2007). Similarly, inflammation caused by bacterial infection, which can be experimentally mimicked by LPS injection, is known to be a priming factor for neurodegenerative disease (Perry, 2004). Because LPS up-regulates the classical complement cascade components, which together with microglial derived C1q, provide a fully active complement system, it is possible that LPS is sufficient to activate complement-mediated destruction of synapses. Could activation of astrocytes in this detrimental orientation thereby sensitize the CNS to neurodegeneration? Using the markers of these potentially protective and detrimental reactive astrocyte subtypes identified in this work, this hypothesis can begin to be tested. If so, a future therapeutic avenue may be to develop new drugs that control the polarization of astrocytes, inhibiting the detrimental orientation and enhancing the protective orientation.
This work was supported by NIH R01 NS059893 (B.A.B.), the Myelin Repair Foundation (B. A. B.), the Adelson Medical Foundation (B. A. B), the Christopher & Dana Reeve Foundation (B.A.B), an A-Star Fellowship (L.C.F.), and NIH RO1 GM49831 (R.G.G).
Conflicts of Interest: The authors have no conflicts of interest.