Background
Neuroinflammation is a fundamental process contributing to the death of neurons in neurodegerenative diseases, for example, in Parkinson’s Disease (PD) [
1], Alzheimer’s Disease, and Multiple Sclerosis [
2]. The underlying neuroinflammation in these neurodegenerative diseases are characterized by microglia activation [
2,
3]. During this inflammatory process, microglia acquire an amoeboid cell shape [
4], and amongst other mechanisms, they decrease the release of neurotrophic factors, such as the brain-derived neurotrophic factor (BDNF), and secrete cytotoxic substances, which lead to neuronal death [
5]. In contrast, in a healthy brain, microglia are found in a ramified morphology and support a variety of central nervous system (CNS) functions by secreting neurotrophic factors. Interestingly, microglia are not permanent in one or another activity state, but are rather able to switch between different phenotypes and activity states [
6]. Therefore, understanding the molecular machinery that reverses the inflammatory activation of microglia is essential to protect from neurodegeneration. Hence, to identify the intracellular players of this reversion in a systematic manner, we combined two experimental approaches in this study.
Firstly, our previous work indicates that adipose-derived mesenchymal stem cells (ASCs) exert important anti-inflammatory actions on microglia. We observed that primary microglia exposed to ASCs or their secreted factors (conditioned medium, CM) underwent a dramatic cell shape change into a highly elongated morphology in vitro [
7], similar to the phenotype of microglia observed in a healthy brain [
8]. The elongation induced by CM was associated with an upregulation of neurotrophic factors, such as BDNF, which is indicative of the acquisition of a neuroprotective phenotype [
7]. Thus, CM-stimulated microglia represent an ideal tool to study the intracellular events necessary for the transition from inflammatory activated to non-inflammatory neuroprotective microglia. Indeed, we previously identified that the small RhoGTPases Rac1 and Cdc42, which are important regulators of the actin cytoskeleton [
9,
10], play major roles in this phenotypic transition [
7].
Secondly, we present here our novel results of a small interference RNA (siRNA) screen targeting a panel of cytoskeletal and inflammatory genes in primary microglia. For our siRNA screen, we chose a microscopy-based screen as readout, since the phenotype transition of microglia is easily detectable by light microscopy and its outcome is directly visible. Although RNAi screens have been applied to create big datasets of image-based analysis of cell lines in high-throughput format [
11,
12], in the case of microglia available, cell lines do not always resemble the physiology and morphology of primary microglia [
13] and our own experience. Therefore, despite the technical challenge associated with the use of primary cells, we chose primary murine microglia as our model system. We were able to develop a reliable protocol to transfect primary microglia with a panel of siRNA, and optimized the experimental assay for its reproducible readout. Then, we exploited the anti-inflammatory properties of CM on microglia [
7], as described above, in combination with the siRNA screen, and in this way, we identified a list of molecules that were implicated in the reversion from activated to neuroprotective microglia. From this list, we validated seven candidates, of which three of them also downregulated BDNF expression and other three molecules enhanced microglia migration. We found that the atypical RhoGTPase RhoE/Rnd3 was a common hit in both secondary screens enabling us to assign a novel function as a potential regulator of a neuroprotective microglia phenotype to this molecule.
Methods
Animals and ethical statement
Female C57Bl/6 mice (7–8 weeks old) were purchased from Charles River. They were used for obtaining P0 to P2 newborns to isolate primary microglia and for obtaining fat tissue to isolate ASCs. Mice were housed in a controlled-temperature/humidity environment (22 ± 1 °C, 60–70% relative humidity) in individual cages (10 mice per cage, with wood shaving bedding and nesting material), with a 12-h light/dark cycle (lights on at 7:00 a.m.) and fed with rodent chow (GlobalDiet 2018, Harlan) and tap water ad libitum. The experimental protocols of this study conform to EU Directive 2010/63 and followed the ethical guidelines for investigations with experimental animals approved by the Ethics Review Committee for Animal Experimentation of Spanish Council of Scientific Research. Animal studies are reported in compliance with the ARRIVE guidelines [
14].
Cell isolation and cultures
Primary microglia were prepared from P0 to P2 newborn C57Bl/6 mice, as described previously [
7]. Briefly, dissected brains were discarded of olfactory bulb, cerebellum, hindbrain, and meninges. Then, they were homogenized in microglia growth medium consisting in DMEM (Invitrogen), supplemented with 10% fetal bovine serum (FBS, Gibco), 10% horse serum, and 1% penicillin/streptomycin (Gibco), using a Pasteur pipette and a 23-G syringe. Brain homogenates were centrifuged. The resulting cells were plated in poly-
d-lysine-coated flasks and incubated with microglia growth medium for 10–12 days at 37 °C and 5% CO
2. Microglia were harvested on a shaker for 2 h at 200 rpm and plated on poly-
d-lysine-coated cover slips at a density of 12,000 cells/cm
2 or tissue culture 6-well plates in microglia growth medium at a density of 37,000 cells/cm
2. After 24 h, microglia growth medium was replaced by fresh microglia growth medium.
ASCs were isolated from adipose tissue of adult C57Bl/6 mice as previously described [
15]. These cells showed a fibroblast-like morphology and differentiation capacity to the adipocytic and osteocytic lineages and expressed the phenotype MHC
−II
−CD14
−CD18
−CD31
−CD34
−CD45
−CD80
−CD117
−CD144
−CD13
+CD44
+CD29
+CD54
+CD73
+CD90
+CD105
+CD106
+CD166
+. CM was collected from passage 2 until passage 6 of ASC cultures, which were plated at a cell density of 15,000 cells/cm
2, grown for 2 days before collecting their supernatant and then stored at − 20 °C. Before use, CM was quickly thawed and passed through a 0.2-μm filter.
siRNA transfection
A custom-made siRNA library against 157 target genes (with three individual siRNAs against each gene) was ordered from Life Technologies. Primary microglia were plated for 48 h as described above. Then, the growth medium was replaced and cells grown on cover slips in 48-well plates were transfected with 12.5 pmol siRNA and 0.375 μl of Lipofectamine 3000 (Invitrogen) according to the manufacturer’s instructions. For 24-, 12-, or 6-well plate cultures, reagents were proportionally up-scaled. Four days after the siRNA transfection, cells were treated with CM for 4 h and fixed to determine their cell shape. Alternatively, after 4 days, cells were harvested for Western blot or RT-qPCR analysis. Not transfected cells or cells that were transfected with a control Scrambled siRNA were used as reference in all studies.
Determination of cell shape and circularity
Microglia plated on cover slips were fixed for 4 min in ice-cold methanol and stained with Isolectin B4, labeled with AlexaFluor 568 (Invitrogen) for 30 min at room temperature and washed two times with PBS. Images were acquired with a × 10 objective on an Olympus IX 81 fluorescence microscope. To determine the circularity or “form factor” [
7,
16], photos of three fields of view were taken per siRNA and analyzed by a Fiji/ImageJ macro to determine the circularity of each individual cell. Briefly, each image was processed by the median filter at a radius of 8 pixels, then a black and white threshold image was generated, cell surroundings were drawn, and the circularity within shape descriptions was determined as 4
π*area/(perimeter)
2. Cells touching the borders of the image were excluded from the quantification procedure. In total, the siRNA screen was performed three times so that the effect of each target gene on circularity was calculated from three independent experiments. Subsequently, the circularity values were normalized within each experiment and compared to microglia transfected with Scrambled siRNA. The fold change difference between the circularity value for each gene and the Scrambled siRNA was called the differential circularity, since it indicates the difference between the Scrambled siRNA and gene siRNA, and is given as logarithm (log
2).
Scratch assay
Migration of microglia was determined using a scratch assay. Cells were plated at 18,000 cells/cm2 and transfected as described above. Growth medium was replaced 3 days after transfection to CM, and the cover slip was scratched with a yellow pipette tip. Bright field images were taken 1 and 24 h after the scratch. The scratch width was analyzed with the Fiji/ImageJ macro Wound healing tool (MRI_Wound_Healing_Tool.ijm). The scratch width at 24 h was expressed as a percentage of the original scratch width at 1 h. For the normalization, the scratch width of the Scrambled siRNA-transfected cells was set to 1.0 in each experiment, and the scratch width of the siRNA-transfected cells were set to values proportional to 1.0.
Western blot analysis
Microglia were transfected as described above and then harvested with cold lysis buffer consisting in 10 mM Tris-HCl pH 8.0, 150 mM NaCl, 1% Nonidet-P40, 1 mM EDTA, 10 mMNaF, 1 mM Na3VO4, and a cocktail of commercially available protease inhibitors (Sigma, containing 104 mM AEBSF, 80 μM Aprotinin, 4 mM Bestatin, 1.4 mM E-64, 2 mM Leupeptin, 1.5 mM Pepstatin A). After centrifugation for 15 min at 14,000 rpm, the protein concentration of the supernatants was determined by Bradford assay (Bio-Rad), and samples were prepared for SDS-PAGE with Laemmli SDS sample buffer. After semi-dry blotting or Western blotting in a wet chamber for Tiam1, PVDF membranes were blocked with 5% milk in TBS-Tween (0.1%) and incubated 8 h at 4 °C with the primary antibodies, rabbit anti-Ahrgef4, (Antibodies-Online), rabbit anti-IκBα (Cell Signaling), mouse anti-Rac1 (BD Transduction laboratories), rabbit anti-GAPDH (Sigma), mouse anti-Tiam1, and mouse anti-Map3k2 (both from Santa Cruz) diluted in 2% BSA/TBS-Tween (0.1%) or incubated 1 h at RT with the primary antibodies mouse anti-Creb1, rabbit anti-RhoE, rabbit anti-GM-CSFR, and rabbit anti-Mapk11 (all from Antibodies-Online) diluted in the blocking solution. Horseradish peroxidase-conjugated secondary antibodies (DakoCytomation) and ECL (GE Biotech) were used for detection. If necessary, membranes were stripped with stripping buffer (100 mM β-mercaptoethanol, 2% SDS, 62.5 mMTris pH 6.8) for 30 min at 55 °C.
RNA extraction and RT-qPCR
Total RNA was extracted using Tripure (Roche) from microglia plated in 6-well plates. After DNase I treatment (Sigma), RNA (1 μg/sample) was reverse transcribed using RevertAid First Strand cDNA Synthesis kit (Fermentas) and random hexamer primers. The cDNA was analyzed by qPCR in triplicates on a Cfx96-Cycler (Bio-Rad) with the SensiFAST™ SYBR® No-ROX Kit (Bioline) and 2.5 pmol of the following primers: mouse Rac1 forward, CCC AAT ACT CCT ATC ATC CTC G; mouse Rac1 reverse, CAG CAG GCA TTT TCT CTT CC; mouse BDNF forward, CCC TCC CCC TTT TAA CTG AA; mouse BDNF reverse, GCC TTC ATG CAA CCG AAG TA with a primer efficiency of 100.7% and GAPDH forward and reverse primers from the RevertAid First Strand cDNA Synthesis kit (Fermentas), with a primer efficiency of 86.3%. After 42 cycles, the Ct values were determined. To normalize the samples, ΔCt between the gene of interest and GAPDH Ct values as reference gene was calculated. The x-fold difference in expression between the different treatments was then determined by subtraction of the ΔCt values and called ΔΔCt. Finally, the total change was calculated as 2−ΔΔct and the relative amount compared to Scrambled siRNA-transfected cells was deducted.
Elisa
BDNF protein levels in culture supernatants of siRNA-transfected primary microglia were determined using the Quantikine® ELISA Kit (R+D Systems®), a quantitative sandwich enzyme immunoassay, according to the manufacture’s recommendations.
Statistical analysis
All data are expressed as the mean ± SEM. The data and statistical analysis comply with the recommendations on experimental design and analysis in pharmacology (Curtis et al., 2018). Statistical analysis was carried out with two-way ANOVA followed by Student’s
t test. We assumed significance at
p < 0.05. For the statistical analysis of the siRNA screen, cell HTS2 package version 2.38.0 [
17] from R version 3.3.2 was applied to calculate and normalize all circularity values within each experiment, as it is the case in the normalization by experiments. Finally, these normalized values were compared for each gene to a Scrambled siRNA used as negative control sample using limma software version 3.30.11 [
18]. Achieved results are presented as a “fold change” in the circularity, given as logarithm (log
2) value. The threshold value between significant and not significant genes was determined by an adjusted
P value: the Fold Discovery Rate (FDR). We applied the standardized FDR, as previously described [
19].
Discussion
Highly elongated and ramified microglia are predominantly found in a healthy brain [
8]. But it is commonly accepted that microglia can display an amoeboid morphology during a neuroinflammatory response [
6], for example, as seen in PD [
4]. Although microglia ramification is equally important as the development of axons and dendrites for neurons to exert their functions, very little is known about the pathways that induce this ramification. To systematically unravel this process, we performed a microscopy-based siRNA screen in primary microglia and have identified a list of 45 genes that are positively involved in the ramification of microglia. Since we are aware of high numbers of false positive hits in siRNA screens [
28], we have applied a rigorous validation procedure in our siRNA screen in order to make our hits reliable and reproducible results. The execution of this validation step seems to be essential, because we observed that three of ten initially selected positive hits were false positives.
Two of the seven positive hits, the Rho guanine nucleotide exchange factor (RhoGEF) Tiam1 and the small RhoGTPase RhoE (also known as Rnd3) are regulators of the cytoskeleton. The NF-κB inhibitor IκBα, the kinases p38β (also known as Mapk11) and Map3k2, the transcription factor Creb1, and the cytokine receptor for granulocyte macrophage colony stimulating factor (GM-CSFR, also known as Csf2ra) are molecules involved in inflammatory responses. Since their knock down inhibited microglia ramification upon CM treatment, a potent anti-inflammatory neuroprotective stimulus [
7], these seven proteins must play an essential role in this process and their identification sheds the first light on these so far poorly characterized signaling routes.
The fact that the BDNF expression is affected by the downregulation of three out of seven positive hits indicates that this might be a possible mechanism for acquiring a neuroprotective phenotype. Indeed, a decreased release of BDNF by microglia is one of the mechanisms by which microglia can kill neurons [
27]. Since BDNF transcription is mediated by Creb1 [
29], our result that BDNF mRNA expression was reduced by Creb1 siRNA was actually not surprising. Interestingly, RhoE has been identified as BDNF-induced Creb1-regulated gene in hippocampal neurons [
30]. Now, we report here for the first time that RhoE vice-versa also regulates BDNF expression, establishing a potential positive feedback similar to that previously described for Creb1 activation-BDNF expression [
29]. Because this positive feedback-loop is one of the reasons to consider to Creb1 as a key player in neuronal survival [
31], our new findings with RhoE could result of relevance from a physiological point of view.
The MAPK p38 is expressed as two major isoforms in the brain, namely p38α and p38β, which have around 70% homology. Most of the published work so far refers to p38α and only few data is available distinguishing both isoforms, mainly because the main p38 inhibitors target both isoforms [
32]. However, a recent report showed that siRNA-mediated downregulation of p38β, but not of p38α, resulted in impaired Creb1 activation in microglia [
33]. These results are in good concordance with our results showing that p38α (Mapk14, Additional file
3) did not come up as positive hit in our primary screen for deactivating-microglia drug targets. Thus, these findings suggest a differential role of both isoforms of p38 in microglia activity states, with p38β mainly being implicated in a neuroprotective phenotype.
Besides affecting particular changes in cell morphology and gene expression, positive hits also changed microglia migration, being a more global cellular read-out. Migration is one of the important functions of microglia to assess a lesion site once injury or insult in the brain has occurred and is typically associated with an inflammatory phenotype [
27]. Three of our positive hits, namely Tiam1, RhoE, and IκBα, enhanced cell migration when downregulated by siRNA. It was previously described that downregulation of RhoE promoted glioblastoma cell migration and invasion [
34]. On the other hand, although the RhoGEF Tiam1 is traditionally believed to enhance cell migration, its role in cell adhesion makes its involvement in cell migration controversial depending on cell type and extracellular matrix used [
35]. Under our experimental conditions, Tiam1 would be rather involved in cell adhesion, since its downregulation enhances primary microglia migration. Regarding IκBα, it is likely that it exerts its neuroprotective role in microglia by inhibiting the inflammatory signaling of the transcription factor NF-κB, probably by retaining it in the cytoplasm [
36]. The role of IκBα itself in cell migration is less studied, but for example curcumin suppresses inflammatory-induced NF-κB signaling and macrophage migration [
37]. In addition, pharmacological inhibition of nuclear translocation of NF-κB blocked endothelial cell migration in a scratch assay [
38]. These results correlate with ours, namely that the downregulation of IκBα, which probably increases nuclear NF-κB, enhanced microglia migration.
Since in our secondary screens not all hits overlap, we must assume that our seven validated hits probably do not act on the same pathway, probably because the CM we applied is a mix of several growth factors and cytokines and can stimulate several pathways at once. It is however remarking that RhoE came up as hit in both analyses, placing RhoE in a central role to act on microglia activation states. RhoE is an atypical RhoGTPase, because its GTPase domain is not able to hydrolyze GTP, being its activity regulated by its expression level and phosphorylation [
39]. As mentioned above, its gene expression can be regulated by BDNF [
30]. RhoE’s main function is to act as endogenous antagonist to RhoA-mediated actin cytoskeleton remodeling, which is especially important in cell migration and polarization. For example, polarization and axonal and dendritic lengths are reduced in RhoE-null neurons [
40], a comparable phenotype to that we observed in RhoE-downregulated microglia, which were less ramified. Taken all these data together, we suggest RhoE is a promising candidate gene, whose action is required for the reversion of microglia into a neuroprotective phenotype, being especially indicative its role in BDNF production. However, further studies are necessary to establish its precise molecular role in this process. Since RhoE, similar to p38β, has not been previously identified publicly available databases related to PD, these findings are completely novel in the context of neurodegenerative diseases.
Finally, our primary siRNA screen also identified 50 genes (Additional file
3) that potentially block the acquisition of a ramified morphology in microglia. This suggests that selective inhibitors against these molecules would favor this neuroprotective phenotype. Although the validation of these negative hits is out of the scope of the present study, it offers this possibility to other researchers.