Background
Microglia are essential components of the central nervous system (CNS) with a broad range of roles in neurodevelopment, homeostasis, synaptic plasticity, and injury responses [
1‐
3]. In healthy brain, microglia survey the brain parenchyma dynamically [
4]. During pathological conditions, microglia dysfunction contributes to the pathogenesis of most neurodegenerative diseases and psychological disorders [
5]. To define the mechanisms underlying microglia function and to investigate the potential utility of these cells as druggable targets, it is indispensable to establish reliable in vitro models for microglia.
Primary microglia are a useful in vitro model for mechanistic studies and compound testing because they recapitulate a majority of known physiological activities of microglia in vivo, including phagocytosis, migration, and release of pro-inflammatory cytokines and chemokines when stimulated [
6]. For now, however, the major research works of microglia, particularly those on fundamental transcriptome and proteome profiles, were obtained based on primary microglia derived from different isolation methods. For instance, microglia isolated with CD11b magnetic bead sorting (MACS) were used for whole genome analysis among different brain regions [
7]. Microglia proteomic identification [
8] and transcriptome signature upon lipopolysaccharide (LPS) [
9] were based on microglia isolated by the shaking method. Moreover, transcriptional changes of microglia isolated from mild-trypsin digestion were investigated in mouse EAE model [
10]. Despite lots of progress, difficulties exist for parallel comparison between studies with microglia from distinct isolation methods because microglia are sensitive and likely to be activated to a certain extent during each isolation procedure [
11]. Furthermore, microglia activation may mask the differences across interventions. Therefore, establishing a reliable quiescent microglia model is vital, which can be potentially used for comparing gene expression and functions in response to different treatments. Due to the requirement of high yield and further cellular culture for mechanic investigation and drug discovery, we isolated microglia from P0–P3 postnatal mouse brains with a mixed glial culture system. The three most popular microglial isolation methods from the mixed glial culture system are shaking, mild trypsinization, and CD11b MACS. In previous literature, although the yield, purity, and viability of microglia have been compared [
12,
13], no one has ever explored the whole transcriptional level of microglia across different isolation approaches. Hence, one of our aims here is to conduct such a comparison and find out the best approach for in vitro microglial isolation and culture from postnatal mouse brains, which is, we believe, of great value to guide future studies.
In addition to primary microglia, microglia-like cell lines have been created and extensively used for examining mechanistic details of microglial function, which include mouse immortalized BV2 cells. Although immortalized cells replicate readily and are easy to maintain in culture, their validity as a sufficient substitute for primary microglia has been debated [
14]. Functionally, microglia cell lines share similarities with primary microglia but are separated by crucial differences in secretion as well as gene expression upon LPS stimulation [
15,
16]. Although distinct expression patterns of specific microglial genes have been reported between BV2 cells and mouse primary microglia [
17], the whole transcriptome signature and their functional differences still need to be further explored.
To this end, we compared transcriptome of isolated primary microglia across the three different isolation methods (shaking, mild trypsinization, and CD11b MACS), as well as with BV2 cells. According to their distinct transcriptome profiles and pathway enrichment, we then performed in-parallel functional assays between primary microglia and BV2 cells regarding LPS responses, transforming growth factor beta (TGFβ) signaling, and chemotaxis. Taken together, we determined the optimal isolation methods for quiescent microglia; transcriptional and functional analyses also revealed that BV2 cells may not adequately represent primary microglia, which offered valuable insights into the selection of appropriate microglial in vitro models under certain circumstances.
Methods
Animals
Wild-type C57BL/6 pregnant mice were obtained from Charles River Laboratories, Inc. Mice were allowed to acclimate for 7 days after receipt. They were kept on a 12-h light/dark cycle and allowed free access to food and water. All animal care and use were in accordance with the Institutional Animal Care and Use Committee (IACUC)-approved protocols.
Mouse microglia isolation and culture
Cortices from P0–P3 C57BL/6 mouse pups were dissected and stripped of meninges and mechanically dissociated with a hand homogenizer and a 25-gauge needle. The cell suspension was seeded into poly-l-lysine-coated (Sigma-Aldrich) T150 tissue culture flasks and maintained in DMEM/F12 with 10% FBS and 1% penicillin–streptomycin for 10–14 days to grow a confluent mixed astrocyte/microglia population.
Shaking purification
The confluent mixed glia cultures were shaken in an orbital shaker at 37 °C with 230 rpm for 2 h. The floating cells were collected, centrifuged, and plated on a poly-l-lysine-coated plate for 24 h. Cultures were washed twice to remove cell debris before RNA was extracted using RNeasy Plus Mini Kit (QIAGEN).
Mild-trypsin purification
Mild trypsinization was performed as described before [
13]. Briefly, trypsin–EDTA solution (0.25% trypsin, cat. no. 25200-072; Invitrogen) diluted 1:4 in PBS containing Ca
2+ was applied to the mixed glial cells. The upper mixed glial cell layer slowly became detached from the bottom of the flask after incubation with this mild trypsin for at least 30 min depending on the culture confluence at 37 °C. When the top layer of mixed glial cells became completely detached, trypsin was inhibited by adding 10% of serum to the flask and the detached upper cells and the trypsin solution were discarded. All that were left at the bottom of the flask were adherent microglial cells. The cells were either directly scraped for RNA extraction or re-plated into a poly-
l-lysine-coated plate for 24 h before RNA extraction using RNeasy Plus Mini Kit (QIAGEN).
CD11b MACS purification
We gently scraped and applied the cells to an antigen–antibody-mediated magnetic cell-sorting (MACS, Miltenyi Biotec) assay to positively select microglia. Briefly, the mixed glial population was re-suspended in MACS buffer (Miltenyi Biotec) and incubated with CD11b MicroBeads (Miltenyi Biotec). The cell suspension was then applied to LS separation column (Miltenyi Biotec) fitted into a QuadroMACS cell separator (Miltenyi Biotec). Unlabeled cells were allowed to pass through the column while labeled cells remained captured in the magnetic field. After washing the column with MACS buffer, the column was then removed from the magnetic separator and flushed with MACS buffer to collect the purified microglia population. For an increased level of purity, the eluted microglia population was passed through a new LS separation column a second time. The purity of microglia used in our study was more than 95% assessed by immunocytochemistry (data not shown). Microglia either acutely collected from the LS separation column or incubated on a poly-l-lysine-coated plate for 24 h were homogenized, and total RNA was extracted using RNeasy Plus Mini Kit (QIAGEN).
BV2 cell line
The microglia BV2 cell line was obtained from Dr. Dennis Selkoe (Harvard University) and cultured in DMEM with 10% FBS and 1% penicillin–streptomycin.
Immunocytochemistry
Immunocytochemistry was performed as described previously. Briefly, cells were fixed with 4% paraformaldehyde and permeabilized by 0.1% Triton X-100. After blocking with 10% donkey serum, fixed cells were incubated with primary antibodies (Iba1, 1:1000, WAKO Chemicals; GFAP, 1:1000, Abcam) for 2 h followed by fluorochrome-conjugated secondary antibodies (Alexa Fluor 488, Alexa Fluor 555, 1:200, Molecular Probes, respectively). Nuclei were counterstained with DAPI. Fluorescence images were acquired using a confocal-laser microscope (LSM 700; Carl Zeiss MicroImaging) with a multi-track configuration.
Microglial purity and morphological analysis
The purity of isolated microglia with each isolation method was determined by the percentage of Iba1+ cells in total cells, which was indicated by immunocytochemical staining using DAPI and the antibodies against Iba1 and GFAP. For morphological analysis, we defined amoeboid microglia as flat Iba1+ cells without thin processes and calculated the percentage of amoeboid microglia in total microglial cells. At least five randomly selected fields were used for quantification.
RNA sequencing and data processing
RNA quality was assessed by using Agilent RNA 6000 Nano Kit and Agilent 2100 Bioanalyzer according to the manufacturer’s instructions. Qualified total RNA (RIN > 9, 200 ng) from each sample was processed by following TruSeq RNA Library Prep Kit v2 protocol (Illumina, San Diego, CA). In brief, poly-A containing mRNA purified from each total RNA samples was applied to cDNA library construction. The libraries were sequenced at pair end with read length of 100 bp on Illumina HiSeq 4000 platform at a depth of more than 40 million reads. The experiments were carried out by BGI Americas (Cambridge, MA), a fee-for-service provider.
ArrayStudio version 8.0 (Omicsoft, Cary, NC) was applied to quality control (QC) raw RNA sequencing reads, map reads to genome, quantify gene expression, and test expression changes. In brief, low-quality bases and adaptors were trimmed and reads less than 25 bases were discarded. Remaining reads were mapped to mouse GRCm38 genomes (
https://www.ncbi.nlm.nih.gov/grc/mouse) using Omicsoft sequence aligner (OSA) [
18] of the ArrayStudio software. Gene expression read count and TPM (Transcript Per kilobase Million) were calculated based on mouse version m10 of GenCode gene models (
https://www.gencodegenes.org/mouse_releases/10.html/). Samples in each group were QCed-based overall gene expression consistency, and outliers were removed before downstream analysis. Robust center scale was applied to normalize data in all heat maps.
We deposited raw read fastq and sample metadata files in NCBI with BioProject ID PRJNA407656.
Differential gene expression and pathway enrichment analysis
Inference tests based on the Voom algorithm [
19] were applied to adjust read depth differences between samples and estimate changes or differences of gene expression when comparing sample groups. Genes with little or no expression (average TPM < 0.1) were excluded from inference tests. Differentially expressed genes (DEG) from the inference test were selected according to expression changes of more than fourfold and adjusted
P value (calculated by Benjamini–Hochberg procedure) of less than 0.05, or stated otherwise.
MetaCore database version 6.31 (
https://clarivate.com/products/metacore/) was applied to analyze the enrichment of DEGs in biological pathways and processes. Enrichment of significant pathways (adjusted
P value < 0.05, calculated by the database) in each analysis was exported from the database and charted using ArrayStudio version 8.0 or Excel.
Integration of published data
Raw microarray data of published studies on microglia cells with LPS treatment (GSE49329), beta amyloid peptide treatment (GSE55627), and aging (GSE62420) were retrieved from GEO (
https://www.ncbi.nlm.nih.gov/geo/). Custom CDF (ENTREZG version 18,
http://brainarray.mbni.med.umich.edu/www/data-analysis/custom-cdf/) was applied to extract gene expression data from raw CEL files, and standard inference tests were applied in treated versus control comparisons. Genes in treatment groups with expression level significantly (adjusted
P value (calculated by Benjamini–Hochberg procedure) < 0.05) induced more than twofold compared with that in control groups in each study were collected for further analysis.
Quantitative real-time PCR
RNA was reverse-transcribed into cDNA using Superscript III Reverse Transcriptase (Invitrogen) with random hexamer primers. Transcript abundance was determined by quantitative PCR using SYBR Green PCR Mix (Applied Biosystems) with the following primer pairs:
Tspo: GCCTACTTTGTACGTGGCGAG (F), CCTCCCAGCTCTTTCCAGAC (R);
Ptgs2: TTCAACACACTCTATCACTGGC (F), AGAAGCGTTTGCGGTACTCAT (R);
Cd86: TGTTTCCGTGGAGACGCAAG (F), TTGAGCCTTTGTAAATGGGCA (R);
Tnfa: CCCTCACACTCAGATCATCTTCT (F), GCTACGACGTGGGCTACAG (R);
Il6: TAGTCCTTCCTACCCCAATTTCC (F), TTGGTCCTTAGCCACTCCTTC (R);
Il1b: GCAACTGTTCCTGAACTCAACT (F), ATCTTTTGGGGTCCGTCAACT (R);
Tgfb1: CTCCCGTGGCTTCTAGTGC (F), GCCTTAGTTTGGACAGGATCTG (R);
Tgfbr1: TCTGCATTGCACTTATGCTGA (F), AAAGGGCGATCTAGTGATGGA (R);
Tgfbr2: CCGCTGCATATCGTCCTGTG (F), AGTGGATGGATGGTCCTATTACA (R);
Serpine1: TTCAGCCCTTGCTTGCCTC (F), ACACTTTTACTCCGAAGTCGGT (R);
C5a: GAACAAACCTACGTCATTTCAGC (F), GTCAACAGTGCCGCGTTTT (R);
C5ar1: TACCATTAGTGCCGACCGTTT (F), CCGGTACACGAAGGATGGAAT (R);
C5ar2: CTGCTGTCTACCGTAGGCTG (F), AGAGGAATCGAACAGTGGTGA (R);
Gapdh: AGGTCGGTGTGAACGGATTTG (F), TGTAGACCATGTAGTTGAGGTCA (R).
Secretome analysis
Secretome assay was carried out as described before [
20]. Briefly, the relative concentrations of secreted proteins in cell supernatants were measured using antibody-based 38-plex immunoassays (Luminex, R&D systems). The 38 secreted proteins were the following: CCL2/JE/MCP1, CCL3/MIP1α, CCL4/MIP1β, CCL5/RANTES, CCL20/MIP3α, CXCL1/KC, CXCL2/MIP2, CXCL10/IP10/CRG2, CXCL12/SDF1α, FGFb, FGF21, GCSF, GMCSF, IFNγ, IGFI, IL1α, IL1β, IL2, IL4, IL5, IL6, IL10, IL12 p70, IL13, IL17A, IL23 p19, IL33, LIX, MCSF, MMP9, Resistin, TNFα, VEGF, CCL11/Eotaxin, CCL22/MDC, CXCL9/MIG, IL9, and RAGE. We then normalized immunoassay measurements of the listed proteins and clustered them using an unsupervised clustering algorithm (Array Studio) to generate proteomic heat maps. Any undetectable proteins for a sample were removed from the analysis.
Western blot
Cells were homogenized and lysed using RIPA buffer (Amresco) with protease and phosphatase inhibitors (Sigma and Roche, respectively). After centrifugation at 13,000g for 5 min, protein concentrations were measured using the BCA protein assay kit (Pierce) and lysates were separated on a 4–12% Bis–Tris gels (Invitrogen) using MOPS sodium dodecyl sulfate running buffer (Invitrogen). Proteins were transferred with the iBlot system onto nitrocellulose membranes (Novex) and incubated with antibodies p-Smad2 (1:1000, Millipore) and Smad2 (1:1000, Cell Signaling Technology). Signal intensities were detected using ECL Western blotting detection reagents (Amersham Biosciences) and evaluated by ImageJ.
Chemotaxis
Cells were seeded into the upper chamber of an ICAM-precoated separate culture plate inserts (Sartorius) with DMEM/F12 containing 0.5% FBS. The same culture medium and 11 nM C5a were added to the lower chamber. Chemotaxis was monitored every hour for 72 h by IncuCyte Zoom live-cell system (Sartorius).
Statistical analysis
Data were statistically compared using t test between two groups, one-way ANOVA followed by Tukey’s post hoc test among multiple groups, and two-way ANOVA among and within groups using GraphPad Prism 7 (GraphPad Software, Inc.). P < 0.05 was considered statistically significant.
Discussion
In this study, we presented an RNA-Seq transcriptome dataset of various microglial in vitro models, including primary microglia isolated by different methods and immortalized BV2 microglia cell line. Furthermore, based on the analysis of transcriptional differences, we compared cellular functions between primary microglia and BV2 cells by measuring LPS responses, TGFβ signaling, and chemotactic capability in parallel. The major findings are the following: (1) the CD11b MACS method was the most reliable and consistent method, which could keep the isolated microglia in a relatively quiescent state; (2) despite distinct transcriptional signature, BV2 cells shared most immune functions, including the responses to LPS with primary microglia, but showed differences in TGFβ signaling and chemotaxis. Hence, our study characterized the usefulness and limitations for certain microglial isolation methods and BV2 cells and provides valuable insights into the selection of proper microglia as in vitro models for specific investigation.
Our current study focused on microglial isolation from postnatal mouse brains with a mixed glial culture system. Transcriptional differences have been reported between isolated microglia from postnatal and adult brains [
17]. Plus, environmental factors including culture conditions impact microglial transcriptome [
28], which brings into concern the application of microglial culture. Nevertheless, postnatal microglia isolation and culture is still a useful tool for microglial studies due to high yield and relatively easy manipulation and culture. Most importantly, postnatal microglia could recapitulate most phenotypes of microglia in vivo, including cytokine secretion, chemotaxis, and phagocytosis [
6]. Here, we compared mouse microglia from postnatal brains with three most popular isolation methods at the transcriptional level and discovered that cells isolated from CD11b MACS and shaking methods were in a relatively resting state, as compared to those from mild trypsinization isolation. This was evidenced not only by microglial activation genes but also by microglial quiescent genes such as TGFβ signaling-related genes. These findings suggest that CD11b MACS- and shaking-isolated microglia are more suitable for comparison of gene expression profiles and functions when treated by potential therapeutic interventions. However, we could not exclude that subtle differences in purity and subpopulation of isolated microglia from distinct methods may influence the results. Our data are in accordance with previous reports of unwanted transcriptional changes and activation of microglia upon enzymatic digestion such as trypsin [
29]. Moreover, differential adherence-based isolation such as shaking [
30] and mild trypsinization [
13] is difficult to control and reproduce. Shaking speed and duration, as well as trypsinization time, depend on microglial confluence in the mixed glial culture and thus differ batch to batch, which may explain the contradiction between our study and Lin et al.’s study [
31]. In contrast, CD11b MACS method relies on antigen–antibody interaction [
12], which is comparatively consistent and reproducible. Furthermore, CD11b MACS method allows co-harvesting of astrocytes and microglia with high purity from the same mixed culture by depletion or positive selection of microglia from MACS column. Therefore, based on our culture experience and RNA-Seq data, CD11b MACS method is considered an efficient and consistent method to isolate pure and inactive microglia, which can be routinely used for mechanistic studies and compound screening targeting microglial functions.
Due to limited yield of primary microglia produced from mouse brains, a BV2 cell line is frequently used as an alternative owing to a shorter preparation time and its homogeneous population across experiments. However, the validation of BV2 cells as a sufficient substitute for primary microglia has been debated [
17]. At present, we performed RNA-Seq on primary microglia and BV2 cells under non-treated conditions and compared their biological pathways and cellular functions. In addition to common properties of immortalized cell lines (e.g., increased proliferation and adherence), BV2 cells retain most crucial functions of microglia in immune response and inflammation. For example, BV2 cells responded to LPS as primary microglia with enhanced transcript expression and secretion but had distinct transcription and secretion profiles within our selected panel. This finding further supported previous observation that although both BV2 cells and primary microglia express Iba1, a microglia marker, BV2 cells exhibit far less induction of some pro-inflammatory genes and much lower cytokine secretion levels in response to LPS, when compared with primary microglia [
15,
16]. Furthermore, some specific signal pathways and functions, such as TGFβ signaling and chemotaxis upon C5a, substantially differ between primary microglia and BV2 cells from transcriptional and functional aspects. These results characterized the usefulness and limitations of BV2 as an alternative in vitro model. Hence, our analysis raised concerns about appropriate cell models when performing a microglial study to address specific immunological and inflammatory responses.
There are a few directions for future investigation. First, the primary microglia in our study are of postnatal origin. Additional genome-wide transcriptional and functional studies should be performed for direct comparison with microglia from adult tissues. Second, this study only looks into intervention-free situations. Further studies are required to investigate transcriptional changes upon different stimuli. Third, although rodent models are effective systems to investigate the emerging functions of microglia, further research work should be carried out to compare rodent and human microglia at the transcription and function levels to facilitate the translation from preclinical to human studies.