Quantitative proteomics of acutely-isolated mouse microglia identifies novel immune Alzheimer’s disease-related proteins

Female C57BL/6 J and female 5xFAD mice (on a C57BL/6 J background) used for the studies were housed in the Department of Animal Resources at Emory University under standard conditions. Institutional Animal Care and Use Committee approval was obtained prior to in-vivo work and all work was performed in strict accordance with the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. Adult mice were given intraperitoneal LPS injections (10 μg/dose × 4 daily doses) to induce acute neuroinflammation [12, 13]. If ≥25% of weight loss was observed, animals were euthanized. Three groups of adult 6–7 mo old C57BL/6 J mice (n = 3 pools each, 2 mouse brains per pool) were used for our studies. Wild-type (WT) mice (untreated), WT mice treated with LPS for 4 consecutive days (IP, 20 μg/dose) and 5xFAD mice were euthanized followed by cardiac perfusion as previously described [13].

The brain was dissected and brain mononuclear cells were isolated as previously described by Percoll gradient centrifugation [13]. Mechanical homogenization of the brain was performed rather than enzymatic digestion due to two reasons: i) in our experience, live microglial cell yield using enzymatic or mechanical dissociation methods are very similar (≈200,000 mononuclear cells per brain), and ii) enzymatic digestion may cleave cell surface proteins, thereby potentially decreasing the likelihood of identifying cell surface proteins by mass spectrometry. Isolated cells were then further purified by CD11b positive selection using a MACS column (Miltenyi Biotec Cat#130–093-636) to selectively enrich microglia and brain mononuclear phagocytes. Positive selection enriched the CD11b⁺ population from 60 to 95% as confirmed by flow cytometry. On average, 200,000 live CD11b⁺ cells were isolated from each brain. Whole cell microglial lysates were prepared and processed for analysis.

Cell pellets were flash frozen and stored at − 80 °C until ready for protein extraction. Cells were lysed in 8 M urea lysis buffer (8 M urea, 100 mM NaHPO4, pH 8.5) with HALT protease and phosphatase inhibitor cocktail (ThermoFisher). Each sample was then sonicated for 3 cycles consisting of 5 s of active sonication at 30% amplitude followed by 15 s on ice. Samples were then centrifuged for 5 min at 15,000 g and the supernatant was transferred to a new tube. Protein concentration was determined by bicinchoninic acid (BCA) assay (Pierce) and 20 μg of each sample was aliquoted.

Protein digestion and sample cleanup was performed as previously published [14]. Briefly, protein samples were reduced with 1 mM dithiothreitol (DTT) for 30 mins, alkylated with 5 mM iodoacetamide (IAA) in the dark for another 30 mins and then 8-fold diluted with 50 mM triethylammonium bicarbonate (TEAB). Overnight digestion was performed with 1:100 (w/w) Lysyl endopeptidase (Wako) followed by an additional 12-h digestion with Trypsin at 1:50 (w/w). The peptide solutions were acidified, and desalted with a C₁₈ Sep-Pak column (Waters). A 2 μg equivalent of each sample elution was pooled and used to create a global internal standard (GIS) and all samples were dried under vacuum.

Tandem mass tag (TMT) peptide labeling was performed according to manufacturer’s instructions and as previously described [14]. One batch of 10-plex TMT kits (Thermo Fisher) was used to label the 9 samples and one GIS mixtures. In this batch, TMT channel 131 was used to label the GIS standard, while the 9 remaining TMT channels were used to label each individual sample. Electrostatic repulsion-hydrophilic interaction chromatography (ERLIC) offline fractionation was performed as previously described [14, 15]. Briefly, dried samples were re-suspended in 100 μL of ERLIC buffer A (90% acetonitrile with 0.1% acetic acid) and separated on a PolyWAX LP column (20 cm by 3.2 mm packed with 300 Å 5 μm beads (PolyLC Inc.) and elution fractions were recovered over a 45-min gradient from 0 to 50% ERLIC buffer B (30% ACN with 0.1% FA). The original 44 fractions were combined as previously described into 21 and dried under vacuum [14, 15].

Assuming equal distribution of peptide concentration across all ERLIC fractions, 10 μL of loading buffer (0.1% TFA) was added to each of the fractions and 2 μL was separated on a 25 cm long by 75 μm internal diameter fused silica column (New Objective, Woburn, MA) packed in-house with 1.9 μm Reprosil-Pur C₁₈-AQ resin. The LC-MS/MS platform consisted of a Dionex RSLCnano UPLC coupled to an Orbitrap Fusion mass spectrometer with a Flex nano-electrospray ion source (Thermo Fisher). Sample elution was performed over a gradient of 3 to 30% Buffer B (0.1% formic acid in ACN) over 105 mins (flow rate started at 300 nl/min and ended at 350 nl/min), from 30 to 60% B over 20 mins at 350 nl/min, and from 60 to 99% B over 5 mins at 350 nl/min. The column was equilibrated with 1% B for 10 min at a flow rate that increased from 350 nl/min to 400 nl/min. The MS was operated in positive ion mode and utilized the synchronous precursor selection (SPS)-MS3 method for reporter ion quantitation as described [14]. The full scan range was 380–1500 m/z at a nominal resolution of 120,000 at 200 m/z and automatic gain control (AGC) set to 2 × 10⁵. Collision-induced dissociation (CID)-Tandem MS/MS at 35% normalized collision energy (CE) and higher energy collision dissociation (HCD) SPS-MS3 at 65% normalized collision energy (CE) were collected at top speed with 3 s cycles. For SPS, the top 10 product ions were notched and fragmented.

Raw data files from Orbitrap Fusion were processed using Proteome Discover (version 2.1). Collected MS/MS spectra were searched against the UniProt mouse proteome database (54,489 total sequences). SEQUEST parameters were specified as: trypsin enzyme, two missed cleavages allowed, minimum peptide length of 6, TMT tags on lysine residues and peptide N-termini (+ 229.162932 Da) and carbamidomethylation of cysteine residues (+ 57.02146 Da) as fixed modifications, oxidation of methionine residues (+ 15.99492 Da) and deamidation of asparagine and glutamine (+ 0.984 Da) as a variable modification, precursor mass tolerance of 20 ppm, and a fragment mass tolerance of 0.6 Da. Peptide spectral match (PSM) error rates were determined using the target-decoy strategy coupled to Percolator [16] modeling of true and false matches. Reporter ions were quantified from MS3 scans using an integration tolerance of 20 ppm with the most confident centroid setting. An MS2 spectral assignment false discovery rate (FDR) of less than 1% was achieved by applying the target-decoy strategy. Following spectral assignment, peptides were assembled into proteins and were further filtered based on the combined probabilities of their constituent peptides to a final FDR of 1%. In cases of redundancy, shared peptides were assigned to the protein sequence with the most matching peptides, thus adhering to principles of parsimony. The search results and TMT quantification as well as raw LC-MS/MS files are included in the ProteomeXchange online repository with identifier PXD009137. Even though TMT labeling limits missing values, normalized abundances were zero for some channels in the dataset, up to 3 out of the 9 channels representing individual samples. Missing values were imputed according to the informative missing-ness assumption, such that these values fit into the left tail of the population-wide Gaussian distribution, with a mean 1.8 population-wide standard deviations less than the mean for non-missing data, and randomness of the imputed population was allowed to vary within ±0.3-fold of the same standard deviation.

Functional enrichment of the differentially expressed proteins was determined using the GO-Elite (v1.2.5) python package [17]. The set of total proteins identified and quantified (n = 4133) was used as the background. Input lists included proteins significantly differentially expressed (p < 0.05 by Student’s t test) comparing either 5xFAD vs WT or LPS-treated WT vs untreated WT mouse microglial proteomic data. Z-score determines over-representation of ontologies and Fisher’s exact test p-value was used to assess the significance of the Z-score. GO Elite v1.2.5 command line application with Z score cutoff of 1.96 and minimum 5 genes per ontology identified ontology enrichments for LPS or AD model-affected proteins, similar to our previously published method [18]. Enrichment Map v2.1.0 Cytoscape plugin [19] was called from Cytoscape v3.5.1 on DAVID-formatted GO-Elite pruned Z-Score output tables. Edge connectivity (Jaccard similarity metric) ranges from 0 (white) to 1 (dark red or blue). Ontology network node colors represent the degree of average log₂ fold change for all differentially expressed members of the ontology in the AD-WT and LPS-WT/CT networks. Since no overlap of gene symbols occurs between differential expression list-derived up and down subnetworks, respectively red and blue subnetworks were overlaid to produce the visualizations shown. Differentially expressed protein lists were also used for bioinformatics analysis (MetaCore software, Thomson Reuters) including pathway analyses to identify over-represented molecular and metabolic pathways as well to identify potential upstream transcriptional regulators as previously described [20].

For mouse tissue, brains were harvested after cardiac perfusion with ice cold saline, followed by 4% paraformaldehyde. After overnight fixation, brains were transferred to 30% sucrose for 24 h after which 30 μm sections were obtained using a cryostat. For all immunohistochemistry studies, three sagittal fixed brain sections from wild-type and 5xFAD mice from equivalent regions were placed on each slide to control for any heterogeneity in staining. For Aβ retrieval, sections were treated with 70% formic acid for 10 min, washed in buffer, blocked with 3% hydrogen peroxide and 10 μg/ml of Avidin for 30 min and then blocked in 10% normal horse serum prepared in Tris-buffered saline (TBS) for 30 min followed by overnight incubation with primary antibody (anti-Aβ42 4G8 antibody [Signet, Cat # 9220–02] 1:1000, anti-Apoe antibody [Meridian Life Science, Cat # K74180B] 1:100 [21], anti-Iba1 antibody [Abcam ab178846], anti-GFAP [Millipore Sigma, Cat # MAB360] 1:100, anti-CD68 [Abcam ab955] 1:100). Sections were rinsed in TBS and then incubated in the appropriate fluorophore-conjugated secondary antibody (1:500) for 30 min after which tissues were mounted on slides, dried and mounted with mounting medium containing DAPI for nuclear staining (Fluoreshield, Sigma-Aldrich F6057). Confocal microscopy was performed on Leica SP8 multi photon microscope and all image processing was performed using Imaris software. For co-localization analyses, appropriate thresholds for each channel were determined using appropriate negative controls. Immunofluorescence microscopy was performed on an immunofluorescence microscope (Microscope: Olympus BX51 and camera: Olympus DP70) and image processing was performed using ImageJ.

Cryopreserved pathology-confirmed AD (n = 3) and age- and sex-matched non-disease control (n = 3) frontal cortex tissues were obtained from the Emory Neuropathological core and sectioned using a cryotostat (30 μm slice thickness). Endogenous peroxidase was quenched with 0.3% hydrogen peroxide for 30 min and then the sample was blocked with 10% normal goat serum/0.4% Triton X-100/Tris-buffered saline (TBS) for 1 h, followed by avidin/biotin binding (Vector labs SP2001) per manufacturer’s instructions, followed by overnight incubation with primary antibodies (anti-Cotl1 [Sigma-Aldrich HPA008918, also validated in the Human Protein Atlas] rabbit polyclonal Ab 1:100) [22]. Sections were rinsed with TBS and incubated with appropriate biotin-conjugated secondary antibody followed by Vectastain Elite ABC and diaminobenzidine (DAB) as per manufacturer’s recommendations. To exclude nonspecific staining unrelated to polyclonal and monoclonal antibodies, immunostaining was performed with omission of the antibodies but with all other procedures unchanged. Slides were lastly counterstained with hematoxylin. Light microscopy was performed with an Olympus light microscope (Olympus, Center Valley, PA). Images acquired from 4 AD cases and 4 non-AD/non-disease post-mortem cases (frontal cortex) were used for quantitative analysis to compare Cotl1 immunoreactivity (ImageJ software) as described [23].

To describe the proteome of adult mouse microglia, we analyzed whole cell lysates of acutely isolated CD11b⁺ MACS-purified microglia isolated from brains of 6–7 mo old wild-type (WT) mice treated with saline or intra-peritoneal LPS for 4 days to induce neuroinflammation [13], as well as from 5xFAD mice (Fig. 1a, 2 brains pooled per sample, n = 3 samples per group) [26]. MACS purification resulted in > 95% enrichment of CD11b⁺ CNS immune cells (Fig. 1b). Among CD11b-enriched cells, CD45^high mononuclear phagocytes accounted for only 1.3, 2.1 and 4.1% cells in untreated WT, LPS-treated WT and 5xFAD mice respectively (Fig. 1c) confirming minimal contamination by CNS-infiltrating macrophages. Microglial whole-cell lysates in 8 M urea buffer were digested by LysC and trypsin, followed by labeling of peptides with TMT reagents, admixture of the TMT 10-plex peptides, followed identification and synchronous precursor selection based MS3 (SPS-MS3) mass spectrometry reporter quantitation (Fig. 1d), and then differential expression and comparative analyses (Fig. 1e).

We obtained a total of 31,865 unique peptides that mapped to 4259 unique mouse gene symbols. Of these, 4133 proteins were identified across the three groups (see Additional File 1: Table S1).. Not allowing any missing values, 3598 proteins were quantified. The distribution of protein abundance was highly skewed such that 5% of proteins (n = 211) accounted for 50% of the cumulative protein weight in the samples (Fig. 2a). Most of these highly abundant proteins were housekeeping and cytoskeletal proteins such as Atp1a3 (α3 subunit of Na⁺/K⁺ ATPase), Sptan1 (α spectrin), Actb (β actin), Sptbn1 (β spectrin) and Tubb4b (beta4 tubulin). Using a previously published purified mouse brain and cell type-specific proteome as a reference (7,198 identified and quantified proteins) [8], 3379 proteins identified in our dataset were also identified in the reference microglial proteome. Of these, 156 proteins were highly abundant (≥90th percentile of abundance) in both datasets. Overall, there was modest correlation of expression levels between the reference microglial proteome and our microglial proteome (Fig. 2b, Spearman’s rank correlation R = 0.28 for all proteins in common, Spearman’s R = 0.23 for the top 25th percentile in common). This modest correlation suggests possibility of methodological and age-related differences in microglial protein expression and myelination, since our proteome was derived from significantly older mice (26–30 weeks of age) as compared to the reference proteome (8–9 weeks of age). In the reference proteome, 633 microglia-specific proteins were also identified as having highest expression in microglia as compared to neurons or other glial cells. Of these, 189 proteins were identified in our microglial dataset. Among the most highly abundant proteins in both microglial proteomes (≥80th percentile of abundance), 17 proteins also met criteria for microglia-specific expression (Fig. 2c). These proteins consistent across two independent datasets (including Myh9, Msn, Coro1a, Cotl1 and Bin1), represent the most highly abundant and microglial-specific proteins in mice.

We also performed gene ontology (GO) enrichment analyses on the entire list of 4133 proteins and identified 180 significant GO terms (adjusted p-value < 0.05). The top 10 GO terms for each category are shown in Fig. 2d (also see Additional file 2: Table S2). This GO analysis shows that our microglial proteome was enriched for cytosolic/cytoplasmic proteins involved in transport and several catabolic, nucleotide/nucleoside metabolic and mitochondrial processes. Interestingly, “neuron projection” and “synapse” were also identified as significantly enriched GO terms in this dataset, which may be explained by the known role of microglia in neuronal synaptic pruning [27, 28]. WikiPathways enriched in our microglial proteome included electron transport chain, oxidative phosphorylation/TCA cycle, proteasome degradation, glycolysis/gluconeogenesis, fatty acid beta oxidation and G protein signaling pathways (Fig. 2e, Additional file 2: Table S2). Results from GO analyses were not affected by the choice of background proteins used (all known mouse gene symbols or 11,937 proteins identified in a reference mouse brain proteome) [8]. In summary, the global microglial proteome from adult mice shows that the large majority of microglial proteins are ubiquitously expressed housekeeping proteins although highly abundant and microglia-specific proteins also exist. The observed enrichment of glycolytic, proteasome degradation, mitochondrial oxidative phosphorylation and fatty acid beta oxidation pathways suggest that adult microglia represent a highly metabolically active and dynamic group of glial cells, rather than bystander resting cells as they were once perceived to be.

Of the 4133 proteins identified across all three groups, 187 proteins were differentially expressed in 5xFAD microglia and 1230 proteins were differentially expressed following LPS treatment in WT mice (Fig. 3, Additional file 3: Figure S1). Of 187 proteins differentially expressed in 5xFAD mice (Fig. 3a), 56 proteins demonstrated ≥1.25-fold increased expression (top 5: App, Drg2, Apoe, Nop56 and Clu) while 105 proteins demonstrated ≥1.25-fold decreased expression (top 5: Plekhg1, Lpgat1, Zc3h14, Ampd3 and Exd2). By applying Benjamini-Hochberg false-discovery-rate (FDR) correction to group-wise ANOVA p-values, 71 proteins demonstrated significant differential expression (FDR < 0.05) across the three groups. Of these, 23 proteins were differentially expressed comparing 5xFAD to WT microglia (6 up-regulated and 17 down-regulated) and the 6 proteins with increased expression in 5xFAD microglia included App, Apoe, Clu, Htra1, Rab2b and Necab1. GO analysis of differentially expressed proteins showed that extracellular proteins, carbohydrate binding, post-transcriptional regulation, GTP binding and regulation of phosphorylation were specifically enriched in proteins increased in 5xFAD microglia while extracellular proteins, anatomic structure development, ion channel activity, transmembrane transport inhibition of cell proliferation, locomotion and learning and memory were enriched GO terms in proteins decreased in 5xFAD microglia (Fig. 3b, Additional file 4: Table S3). Pathway analyses of differentially expressed proteins identified aberrant lipid trafficking, altered lipid metabolism and ATP metabolism as being enriched among proteins with increased expression in 5xFAD microglia while anti-inflammatory IL4 signaling was enriched in proteins with decreased expression in 5xFAD microglia (Additional file 5: Table S4). Transcriptional factors predicted to be upstream regulators of proteins overexpressed in 5xFAD microglia included PUR2α, AP-2 and p53 while Dec1 (Stra13) and c-Myc were predicted as upstream regulators of proteins decreased in 5xFAD microglia (Additional file 4: Table S3).

In addition to identifying neurodegeneration-associated microglial proteomic changes, we used the systemic LPS-induced neuroinflammation model to compare acute neuroinflammatory changes with chronic neuroinflammation seen in more indolent neurodegeneration models [12, 13]. In microglia from LPS-treated WT mice, 363 proteins were ≥ 1.25-fold upregulated and 823 were ≥ 1.25-fold down-regulated (Fig. 3c). Proteins increased following LPS treatment were enriched for GO terms including cellular responses to stress/cytokines, mitochondrial localization, increased cell proliferation and ribosomal processes (Fig. 3c, Additional file 4: Table S3). Conversely, proteins decreased by LPS treatment were enriched for GO terms including oxidoreductase activity, cell adhesion and cell contact, carboxylic acid metabolism and ATP metabolism, suggesting a general suppression of oxidative phosphorylation and motility (Fig. 3d, Additional file 2: Table S2). We identified several pro-inflammatory LPS-upregulated microglial proteins that were increased in 5xFAD microglia (top 5: Clu, Nudt2, Glipr2, Diablo and Cstf2), and 30 LPS-downregulated proteins that were decreased in 5xFAD microglia (top 5: Bcorl1, Plekhg1, Rasgef1a, Ipo4 and Bicd1). Proteins differentially expressed in 5xFAD mouse microglia showed mostly concordant directions of change in microglia derived from LPS-treated WT mice (Fig. 4a, R = 0.47 p < 0.001) suggesting that AD pathology, in general, induces a pro-inflammatory state in microglia. However, differences between LPS-activated microglia and 5xFAD microglia were also observed. Several proteins were differentially expressed in only in 5xFAD mouse microglia without any LPS-induced changes, including increased levels of App (Aβ peptide), Nop56, Ptn, Tmem151b and Mgrn1 and decreased levels of Hist2h2be, Sar1a, Ttc9a and Aimp3 (Fig. 4c), suggesting that these proteins may be uniquely relevant to AD pathology. A comparison of GO enrichment terms from each analysis (5xFAD vs WT and LPS-WT vs WT, Fig. 3) reveals that while both LPS and AD pathology resulted in a general inhibition of microglial motility/locomotion, LPS-treatment induced proliferative responses and mitochondrial metabolic processes while AD pathology inhibited cell differentiation and proliferation while promoting apoptosis, highlighting key metabolic and functional differences between these two states of microglial activation despite anatomical/structural similarities.

To compare the proteomic profile of microglia from acute neuroinflammation and neurodegeneration models with that of advanced aging, we contrasted our findings with an existing microglia proteomic study that compared young (3–5 mo) and very old mice (20–24 mo) [10]. Of 267 aging-associated microglial proteins identified in this reference dataset, 198 were identified in our microglial proteome of which 62 were differentially expressed either in 5xFAD or LPS-treated WT mouse microglia. These 62 proteins could be broadly classified into 4 clusters based on direction of change in protein expression in aged and LPS-treated mice (Fig. 4d). Only 8 proteins with altered expression in 5xFAD microglia also showed aging-related changes. Of the proteins increased in 5xFAD microglia, Hexb, Apoe and Cotl1 increased with aging while Cstf2 and Set decreased with aging. Interestingly, all three proteins that increased in aged WT and 5xFAD mouse microglia are highly expressed by microglia in the CNS [29], highlighting the microglial specificity of these AD- and aging-related protein alterations. In summary, this comparative microglial proteomic analysis from models of acute neuroinflammation, aging and AD pathology demonstrate model-specific as well as overlapping inflammatory changes in microglia.

To assess the degree of concordance between transcriptomic and proteomic changes in microglia, we contrasted our proteomic findings in WT, LPS-induced neuroinflammation and 5xFAD models with existing RNAseq or microarray data from CD11b⁺ microglia acutely isolated from similar model systems [5, 25]. Of the common gene symbols in existing RNAseq and our proteomic dataset from LPS-treated WT mouse microglia, we identified 72 genes/proteins were differentially expressed in both datasets of which only 35 (48.6%) showed concordant changes (11 up-regulated, 24 down-regulated, Fig. 5a). Similarly, we contrasted our proteomics findings to existing transcriptomic data obtained from CD11b⁺ microglia from WT and 5xFAD mice [24], in which 3200 genes were found to be differentially expressed. Our proteomic dataset comparing 5xFAD to WT microglia included 902 of the 3200 genes which were altered at the transcript level. Among the 187 differentially expressed proteins (Fig. 5b), 43 were also altered at the transcriptomic level. Of these 43, only 22 (51.2%) showed concordant changes (8 up-regulated, 14 down-regulated). Among gene symbols identified in both proteomics and transcriptomics datasets, we also identified expression changes in 5xFAD mice that occurred exclusively at the protein or transcript levels (Fig. 5c). Proteins upregulated in 5xFAD microglia that did not demonstrate changes at the transcript level included Nop56, Pnoc, Ikbkg, Nudt2, Glipr2, Diablo and Htra1. Conversely, genes upregulated in 5xFAD microglia without proteome-level changes included Fabp3, Fabp5, Myo1e, Myo5a, Atp1a3, Lmnb1 and Clic4. Overall, these results highlight the modest levels of concordance between expression changes at the transcriptome and proteome levels and emphasize the need to perform molecular profiling at both transcript and protein levels since each can provide unique and novel biological insights into neurodegenerative disease processes.

Since amyloid β accumulation seen in 5xFAD mice only represents one of the pathological hallmarks of human AD, we do not expect all protein changes in mouse 5xFAD microglia to be relevant to human disease [26]. To identify proteomic changes observed in mouse 5xFAD microglia that are consistent with human AD, we compared our mouse microglial proteome with an existing human post-mortem brain proteome in which frontal cortex from non-disease controls and pathology-confirmed AD cases were compared [14]. Three thousand six hundred forty-four proteins identified in our microglial proteome were also identified in this reference human brain proteome. While the overall correlation between relative protein expression data between mouse microglia and human brain samples for all identified proteins was low (R² = 0.002), moderate correlation (R² = 0.34) was observed for 55 proteins that were differentially expressed in both datasets. Of these 55 proteins, 11 (including App, Apoe, Htra1, Cotl1 and Clu) were concordantly increased while 23 (including Vgf, Rtn4, Alpl, Scn3a and Camk4) were concordantly decreased in both 5xFAD microglia and human AD brain (Fig. 6a).

Since App was identified as the most highly differentially expressed and human AD-relevant protein in 5xFAD microglia, we further investigated individual App peptide-level expression to determine whether this finding was driven by de-novo App synthesis or Aβ phagocytosis by microglia. Five unique App peptides were identified in the dataset (Fig. 6b) of which 2 mapped to the Aβ sequence, 2 mapped to the C-terminal YNPTY endocytosis motif and 1 mapped to N-term residues 439–450. Since all five App peptides were identical in mice and humans, we were unable to directly confirm the source (mouse or human) of the App peptides identified in microglia. Selective increased expression of both Aβ42 peptides was noted in 5xFAD microglia while the C-term peptide and the N-term sequences showed no group-wise differences. This pattern of App peptide differential expression strongly suggests that intra-microglial Aβ is phagocytosed and not produced within microglia.

Among the top 5 proteins (Fig. 6a) with increased expression in 5xFAD microglia and post-mortem human AD brain, Apoe and Cotl1 were also highly expressed at the transcriptional level by microglia [8, 29]. Therefore, we performed additional neuropathological studies to characterize Cotl1 and Apoe protein expression in AD. Cotl1 or coactosin like F-actin binding protein 1 is an actin-binding protein that promotes the activity of 5-lipoxygenase, as enzyme that converts arachidonic acid into leukotriene A4 [30]. Cotl1 is also expressed by immune cells including macrophages and has been recently reported to regulate actin dynamics at the immune synapse in addition to promoting the function of lipoxygenase 5 (5LOX), an enzyme that converts arachidonic acid to leukotriene A4 (LTA4) [31]. Cotl1 is highly and specifically expressed by microglia at the transcript and protein levels (Fig. 7a) [8, 29], consistent with our finding that Cotl1 is a highly abundant microglial protein. In an analysis of an existing proteomic dataset comparing frontal cortical regions from AD, Parkinson’s disease and non-disease post-mortem brain tissues [18], we found that Cotl1 protein expression was increased in AD cases (Fig. 7b). Cotl1 protein expression was also strongly correlated with neurofibrillary tangle pathology (Braak stage) (Fig. 7c) and belonged to a glial co-expression protein module that was positively associated with AD pathology (Fig. 7d) [32, 33]. In addition to Cotl1, this glial module also contained other proteins that are pertinent to eicosanoid biosynthesis such as LTA4 hydrolase (Lta4H) which catalyzes the conversion of LTA4 to LTB4 [34] and, prostaglandin D2 synthase (Ptgds) which catalyzes the conversion of prostaglandin H2 (PG-H2) to PG-D2 [35]. In immuno-histochemical studies of human post-mortem brains, we found that Cotl1 was specifically expressed by CD68⁺ microglia (Fig. 7e) but not by GFAP⁺ astrocytes (Fig. 7f). In a comparative immunohistochemistry analysis of AD and non-AD post-mortem frontal cortex tissues, we found that cortical and sub-cortical white matter microglia showed significantly higher Cotl1 protein expression as compared to non-AD controls (Fig. 7g). We also observed that microglia highly expressing Cotl1 in the AD brain had an activated morphology with larger cell bodies and numerous processes (Fig. 7g).

The second protein we validated was Apoe (apolipoprotein E) which regulates cholesterol and Aβ transport into the brain and is strongly implicated in AD pathogenesis [36]. Carriers of the Apoε4 allele have significantly higher risk of developing late-onset AD and in mouse models of AD pathology, Apoε4 promotes neuroinflammatory responses leading to neurodegeneration, independent of its effects of Aβ processing [6, 36, 37]. Apoe is highly expressed in astrocytes and microglia at both transcriptomic and proteomic levels (Fig. 8a) and is a specific marker of disease-associated-microglia that are highly prevalent in neurodegenerative diseases [5, 8, 29]. We observed that Apoe protein expression in mouse 5xFAD brains closely resembled immunostaining of Aβ plaques (Fig. 8b) which was not affected by preincubation with Aβ fibrils (Additional file 3: Figure S2). Apoe also partly co-localized with Iba1 immuno-reactive microglia (Fig. 8b) confirming the presence of intra-microglial Apoe protein. Although Apoe and Aβ (4G8) showed overlapping plaque-like immunostaining patterns with high degree of co-localization, Apoe within microglia also co-localized with Aβ, suggesting shared intracellular compartmentalization, presumably within phagosomes (Fig. 8c and d). These data suggest that Apoe within microglia could represent either phagocytic co-uptake of Apoe along with Aβ, or may represent functional interaction between microglial Apoe in the phagocytic uptake of extra-cellular Aβ and shared subsequent subcellular compartmentalization in microglia. These results show that Cotl1 is specific microglial protein that is selectively increased in AD while Apoe, at least at the protein level, may be involved in phagocytic uptake of Aβ in plaque-associated microglia.