Unique N-glycosylation signatures in human iPSC derived microglia activated by Aβ oligomer and lipopolysaccharide

Human iPSC culture and microglia differentiation

Human iPSCs were obtained from ALSTEM.INC. (Richmond, CA). The line used in this study was Human iPS Cell Line 26 (Episomal, CD34+, ApoE3). Cells were plated onto Matrigel (Fisher) coated plates and cultured with mTeSR plus (Stemcell Technology). For microglia differentiation, we followed a previously described protocol^22,23. Briefly, 2 × 10⁶ iPSCs were plated onto Aggrewell 800 plates (Stemcell Technology) to form embryoid bodies (EBs) in mTeSR1 supplemented with Bone Morphogenetic Protein 4 (BMP4, 50 ng/ml)/Vascular Endothelial Cell Growth Factor (VEGF, 50 ng/ml)/Stem Cell Factor (SCF, 20 ng/ml) and culture for four days with daily medium change. On the fifth day, EBs were plated onto gelatin coated 6-well plates with 20 EBs per well in X-VIVO15 (Lonza) supplemented with M-CSF (100 ng/ml), IL-3 (25 ng/ml), Glutamax (2 mM), Penicillin/streptomycin (100 U/ml and 100 ug/ml) and β-mercaptoethanol (55 μm), and the medium was changed weekly. After 3–4 weeks, floating primitive macrophage precursor (PMP) were collected and plated onto 12-well plates (5 × 10⁴ cells/well), 6-well plates (3 × 10⁵ cells/well), or 100 mm dishes (1.5 × 10⁶ cells) and differentiated in microglia differentiation medium (Advanced DMEM/F12 supplemented with IL-34 (100 ng/ml), GM-CSF (10 ng/ml), N2 supplement (1x), Glutamax (2 mM), Penicillin/streptomycin (100 U/ml and 100 ug/ml) and β-mercaptoethanol (55 μm)) for two weeks. The cultures were routinely tested for mycoplasma. To evaluate the maturity of differentiated microglia, we routinely quantified the marker genes of mature microglia such as TREM2, TMEM119, and P2RY12 by quantitative PCR (Table 1), morphological changes induced by pro-inflammatory stimuli AβO and LPS (Supplementary Fig. 1), as well immunoreactivity of IBA1²². Both PMP and mature microglia were also recognized by high expression of C1QA (Supplementary Fig. 1 A). To ensure consistency, all experiments were conducted using mature microglia obtained following two-week differentiation. In this study, experiments using a batch of differentiated microglia are considered as independent replicates that generate independent data points.

For AβO and LPS stimulation, cells were treated with AβO (3 µM) and LPS (100 ng/ml) for 24 h. Each group, including the Control group, has six replicates.

AβO preparation

AβO composed of Aβ1–42 peptide was prepared following a standard procedure²⁴ with a modification that the HFIP treated Aβ1–42 peptide (Bachem) was dissolved in DMSO and then diluted with Advanced DMEM/F12 culture medium instead of the F12 medium originally described, followed by incubation at 4 °C for 24 h and 10 min centrifugation at 10,000 ×rpm at 4^oC. This preparation of AβO has been extensively characterized in our laboratory¹⁶. Briefly, to ensure consistency of quality, a random sample from each batch was chosen and imaged using electron microscopy and atomic force microscopy to characterize the size and shape of the aggregates. The biological activities of each batch were confirmed by determining for AβO the neurotoxic activity, synaptic binding activity, and ability to rapidly induce exocytosis of MTT formazan, as described previously¹⁶.

RNA isolation, processing, and sequencing

Total RNA from cultured cells was extracted using RNeasy Plus Mini Kit (Qiagen). RNA quality evaluation (yield, purity and integrity), cDNA library construction and Illumina sequencing were performed by NovoGene (Sacramento, CA, USA) using 2 × 150 bp paired-end run. A total of 24 samples, comprising four groups (AβO vs. Control and LPS vs. Control) with six replicates per group, were sequenced with the output of 53.3 ± 8.2 M reads per sample.

Mapping and differential expression analysis

Adapter sequences in the raw FASTQ files were trimmed using Trimmomatic (v0.39). The remaining reads were mapped to the human reference genome hg38 using HISAT2 (v2.2.1)²⁵. The mapping rate was around 97%. The gene counts were summarized by featureCounts (v2.0.1)²⁶, achieving a successful alignment of 70–80% of the reads per library. Differential gene expression was analyzed using edgeR package²⁷ in R version 4.1.0 (R Foundation for Statistical Computing, Vienna, Austria). Low-expressed genes were dropped using the filterByExpr() function. The filter step keeps genes that have count-per-million (CPM) above 10 in 70% of samples. The library sizes were recalculated after filtering. Sample- specific effects were removed by normalizing library sizes with the trimmed mean of M-values (TMM) method using the calcNormFactors() function. A negative binomial model with the quasi-likelihood (QL) F-test was applied to perform the differential expression analysis among diagnosis groups. First, gene counts with normalized library sizes were fitted to a negative binomial generalized linear model. Then, the dispersion was estimated using the estimateDisp() function. The QL dispersion estimation was calculated using the glmQLFit() function, followed by the glmQLFTest() function that conducted the quasi-likelihood (QL) F-test. Finally, we used the topTags() function to output significantly differentially expressed genes. P-values were adjusted for multiple hypothesis testing using Benjamini & Hochberg (BH) with a threshold of adjusted p-value ≤ 0.05.

Sialylation and fucosylation inhibition

For the application of glycosylation inhibitors, microglia were pre-incubated with 100 µM 2 FF (2-Fluorofucose, SynChem Inc) or 100 µM 3 FS (2,4,7,8,9-Penta-O-acetyl-N-acetyl- 3-fluoro-D-neuraminic acid methyl ester, Carbosynth) for 1 h before the addition of AβO (3µM) or LPS (100 ng/ml). Twenty-four hrs later, cells were lysed for the following experiments. The adoption of the concentration of 100 µM for both 2 FF and 3 FS was based on the optimal inhibition determined by Zhou et al. and our previous study^28,29. No apparent toxicity was observed when mature microglia were treated with these two compounds at 100 µM.

qPCR

Total RNA from cultured cells was extracted using RNeasy Plus Mini Kit (Qiagen). cDNA was synthesized using 100 ng RNA and iScript Reverse Transcription Supermix (BioRad). Quantitative real-time polymerase chain reaction (qPCR) was performed using SsoFast™ EvaGreen Supermix and CFX96 qPCR system (BioRad). The forward/reverse primer sequences used are listed in Table 1. Gene expression was normalized to an endogenous gene, β-actin. Relative cDNA levels for the target genes were analyzed by the 2-ΔΔCt method.

Cell membrane enrichment

Samples were prepared for glycomic analysis based upon previously published methodologies³⁰. Briefly, harvested cells were centrifuged at 2,000 RCF for ten minutes after which the cell media was exchanged with 1.2 mL of 0.25 M sucrose (MilliporeSigma, Cat# S7903), 20 mM HEPES (Thermo Fisher Scientific, 1 M, Cat# 15630080) adjusted to pH 7.4 with KOH (MilliporeSigma, Cat# P5958) and a 1:100 protease inhibitor cocktail (Calbio-chem, Cat# 539137) in water and pipette mixed gently. Cells were then lysed by sonication where a total of 60 joules of energy were applied to each sample over a one-minute period. The homogenate was centrifuged at 2,000 RCF for 10 min at 4 °C and then the supernatant was transferred to 1.5 mL ultracentrifuge tubes and the nucleus precipitate was discarded. The cell membrane was precipitated by ultracentrifugation at 42,000 RCF for 45 min at 4 °C after which the cytoplasmic supernatant was discarded taking care not to disturb the membrane pellet. Samples were resuspended in 0.5 mL of 0.2 M Na2 CO3 (MilliporeSigma, Cat# P5958) in water and ultracentrifuged again after which the supernatant, containing membrane-associated proteins, was discarded. A final ultracentrifugation step using the same conditions with 0.5 mL of water was done to wash and desalt the samples, discarding the supernatant after.

N-glycan sample preparation

The enriched membrane fraction was then resuspended in 200 µL of 100 mM NH4HCO3 (MilliporeSigma, Cat# 09830), 5 mM Dithiothreitol (Promega, Cat# V3151) in water to assist in protein denaturing. After gentle homogenization by pipette mixing, samples were placed in a 100 °C water bath for two minutes with 10 s heating and cooling intervals to further denature the membrane proteins. After samples cooled to room temperature, 2 µL of PNGase F (New England Biolabs, Cat #P0705L) was added to enzymatically hydrolyze the N-glycans from their protein substrates. Enzymatic cleavage and conversion of the resulting reducing end amides to hydroxyls occurred at 37 °C for 18 h. Sample were brought to a total volume of 600 µL with water and then ultracentrifuged using the previously described parameters. The supernatant containing the released N-glycans was transferred to a new vial and the remaining membrane pellet was stored at − 20 °C until ready for glycosphingolipid processing. N-glycan samples were cleaned up using a porous graphitic carbon solid phase extraction plate (PGC-SPE, Glygen, Cat# FNSCAR800). SPE wells were first washed with elution solvent consisting of 80% acetonitrile, 0.1% trifluoracetic acid in water (v/v) and then conditioned in wash solvent consisting of 40% acetonitrile, 0.05% trifluoracetic acid in water (v/v). Samples were loaded onto the solid phase and washed three times with 600 µL of washing solvent and all liquids were discarded. N-glycans were then eluted with 600 µL of elution solvent into a new collection plate. For each SPE step, the PGC plate was centrifuged at 150 RCF to quickly ensure solvent flow-through. Enriched N-glycan samples were concentrated by drying to completion in vacuum centrifugation and reconstituted in 30 µL of water prior to analysis.

N-glycan analysis methodology

Released N-glycan analysis was conducted with an Agilent 1200 series ChipCube system operating in positive ion mode and a 6520 Q-TOF mass spectrometer. Chromatographic separations were carried out utilizing a PGC-Chip II, 40 nL enrichment column, 75 × 43 mm analytical column, 5 μm (Agilent Technologies, G4240 - 62003). Mobile phase A (MPA) composition was 3% acetonitrile and 0.1% formic acid in water (v/v). Mobile phase B (MPB) composition was 90% acetonitrile, and 1% formic acid in water (v/v). Samples were cooled with a 4 °C sample tray and 5µL was injected by the autosampler, loaded onto the enrichment column, and washed for 10 min with 100% MPA at 3 µL/min. Next, the chip valve switched to analysis mode moving the enrichment column in line and the analytical gradient began with a flow rate of 0.33 µL/min. 0% MPB was held from 0 to 2 min, at 20 min MPB reached 16%, at 40 min MPB reached 72%, and at 42 min MPB reached 100%. The 100% MPB wash was held until 52 min and decreased back to 0% MPB at 54 min, equilibration of starting composition was held until 65 min. Source parameters used nitrogen as the drying gas with a temperature of 325 °C flowing at 5 L/min and a capillary voltage of 1800 V. Auto MS/MS acquisition mode allowed 4 precursors per cycle with a threshold of 2000 counts and quadrupole filtering with a narrow isolation width for the 600–2000 m/z range. Collision-induced disassociation with N2 gas used a precursor-dependent formula y = 1.8*(m/z)/100 − 3.6 to determine individual collision energies with a range of 100–2000 m/z. Continual mass correction with a reference calibration standard of 1221.990637 m/z was used (Agilent Technologies, HP- 1221). N-glycan compound verification used MassHunter Qual B08.00 software with the ‘Find by Molecular Feature’ algorithm and an in-house database containing N-glycan monoisotopic masses and descriptive information (Supplementary Table S1).

Sphingolipid sample preparation

The membrane pellet from the initial steps was removed from − 20 °C storage and allowed to acclimate to room temperature. Samples were resuspended in 500 µL of freshly prepared sphingolipid extraction solvent consisting of water/methanol/chloroform (3:8:4, v/v%) and sonicated for 15 min to assist in solvation. Proteins were precipitated by centrifugation at 8,800 RCF for 5 min and the supernatant was transferred to a new 1.5 mL Eppendorf vial. A bilayer phase separation was induced by the addition of 100 µL of 0.1 M KCl (MilliporeSigma, Cat#104936). The top aqueous-rich layer containing the sphingolipids of interest was transferred to a new vial and dried by vacuum centrifugation. Sphingolipid samples were cleaned up using a C- 8 solid phase extraction plate (Glygen, Cat# FNSC08.800). SPE wells were first washed with elution solvent consisting of methanol/isopropyl alcohol (1:1, v/v%) and then conditioned in wash solvent consisting of methanol/water (1:1, v/v%). Samples were reconstituted with wash solvent, loaded onto the solid phase by gravity and the flow-through was reloaded. Sample wells were washed three times with 600 µL of washing solvent and all liquids were discarded. Sphingolipids were then eluted with 400 µL of elution solvent into a new collection plate. For each SPE step, the C- 8 plate was allowed to gravity load to ensure analyte binding. Enriched sphingolipid samples were concentrated by drying to completion in vacuum centrifugation and reconstituted in 20 µL of water/methanol (1:1, v/v%) prior to analysis.

Sphingolipid analysis methodology

Sphingolipid analysis was conducted with an Agilent 1200 series nanoflow HPLC equipped with a 10pt/2ps µ-valve(G1316 - 68709), nanoflow ESI operating in positive ion mode (G1992 A), and a 6530 Q-TOF mass spectrometer. Chromatographic separations were carried out utilizing a Zorbax SB- 300 C- 8, 0.075 × 50 mm, 3.5 μm column (Agilent Technologies, Cat# 5065–9923). Four mobile phase compositions were used in this analysis with two low-flow pumps. The first pump (pump 1) operated at 0.7 µL/min and was used to load samples directly and wash the analytical column. The loading mobile phase (MPL) composition was 20 mM ammonium acetate in 56% methanol, 14% isopropyl alcohol, 0.1% glacial acetic acid and water (v/v). The washing mobile phase (MPW) composition was 50% methanol and 50% isopropyl alcohol (v/v). The second pump (pump 2) operated at 0.4 µL/min and was used to carry out the analytical gradient. Mobile phase A (MPA) had a composition of 20 mM ammonium acetate in 0.1% glacial acetic acid and water (v/v). Mobile phase B (MPB) had a composition of 20 mM ammonium acetate in 0.1% glacial acetic acid, 20% isopropyl alcohol, and 80% methanol (v/v). Samples were cooled with a 4 °C sample tray and 1µL was injected and loaded directly onto the analytical column for 4 min with 100% MPL. At 4 min the µ-valve switched positions moving the analytical column into pump 2’s flow path and the gradient began. 70% MPB was held from 4 to 6 min, at 35 min MPB reached 93%, and at 36 min MPB reached 100%. The 100% MPB was held until 39 min and decreased back to 70% at 40 min, equilibration of the starting composition was held until 47 min. Source parameters used nitrogen as the drying gas with a temperature of 300 °C flowing at 5 L/min and a capillary voltage of 1000 V. Auto MS/MS acquisition mode allowed 5 precursors per cycle with a threshold of 600 counts and quadrupole filtering with a narrow isolation width for the 600–2000 m/z range. Collision-induced disassociation with N2 gas used a precursor-dependent formula y = 1.2*(m/z)/100 + 12 to determine individual collision energies with a range of 100–2000 m/z. Continual mass correction with a reference calibration standard of 1221.990637 m/z was used (Agilent Technologies, HP- 1221). Sphingolipid compound verification used MassHunter Qual B08.00 software with the ‘Find by Molecular Feature’ algorithm and an in-house database containing sphingolipid monoisotopic masses and descriptive information (Supplementary Table S2).

Statistical analysis

Single glycans with missing values for more than 80% of samples were excluded from further analysis (Supplementary Fig. 2). The fractions of glycan subtypes (as shown in Fig. 1A), and sialylated and fucosylated N-glycans were calculated as the sum of relative abundances of single glycans. Differential abundance analysis was performed as linear regression models with the glycan abundance as the dependent variable and the treatment as the independent variable in R. ANOVA tests and contrasts between treatments and control were evaluated. The AβO and LPS groups were compared to the control group using Dunnett’s test. The differences of the three treatment groups on the PCA plot were evaluated using the MANOVA test followed by post-hoc linear discriminant analysis (LDA).

Both AβO and LPS induce significant changes in the N-glycan profiles of hiMG

We analyzed the N-glycan profiles of control (mock treatment), AβO-activated, and LPS-activated hiMG using LC-MS/MS technology, with six replicates per group. We treated hiMG for 24 h with 3 µM AβO and 100 ng/ml LPS, optimal concentrations to induce pro-inflammatory cytokines IL-1β, TNF-α, and IL-6. Figure 1A illustrates the N-linked glycosylation types. In all groups, the complex-type N-glycans were the most abundant N-glycans in hiMG, making up ~ 35–40% of total N-glycans (Fig. 1B). This is in keeping with the general composition of brain glycans³¹. Hybrid glycans were the least abundant in hiMG, making up less than 10% of total N-glycans (Fig. 1B). Complex-type N-glycans can be further classified as bi-, tri, and tetra-antennary, as well as bisecting N-glycans, which contains an additional GlcNAc bound to the core (Fig. 1A). In hiMG, the tetra-antennary structure constituted the predominant subtype of complex-type N-glycans, followed by bi-antennary, bisecting, and tri-antennary structures (Fig. 1C). N-Glycans can be decorated by adding fucose, sialic acid, or both to the terminal end of glycans. Fucose can also be added to the core structure. These modifications are important for protein functions such as protein stability, cell adhesion, signal transduction, immunological responses, and cell-to-cell interactions^32,33. We quantified such modifications and found that in hiMG, sialofucosylated N-glycans were the most abundant, making up ~ 50% of glycans. About 30% of N-glycans were not decorated by fucose or sialic acid residues. The rest of N-glycans were only decorated by either fucose or sialic acid residues (Fig. 1D). Considering both subtypes and decorations, sialofucosylated complex N-glycans and undecorated oligomannose N-glycans are the two most abundant structures in hiMG (C-FS and Oligo in Fig. 1E). AβO and LPS treatments had a significant impact on the N-glycan profile of hiMG as shown by the increased dispersion of the AβO-activated hiMG compared to the control cluster and the separation of the LPS-activated hiMG clusters from the control cluster on the PCA plot (MANOVA p-value = 0.0014). The distance between the LPS cluster and the control cluster was greater than the distance between the AβO cluster and the control cluster, indicating that LPS might induce a higher degree of glycosylation changes than AβO (Fig. 1F, Supplementary Fig. 3 A).

Decreased bisecting N-glycans in AβO-activated hiMG

According to the N-glycome, AβO, but not LPS, treatment reduced the percent abundance of bisecting N-glycans in hiMG (Fig. 1C). Consistent with this glycomic change, differential expression (DE) analysis of RNA-seq data revealed that following AβO treatment, beta- 1,4-mannosyl-glycoprotein 4-beta-N-acetylglucosaminyltransferase 3 (MGAT3) was the most downregulated gene across all genes involved in the core N-glycan formation process (Fig. 2A), which was further confirmed by q-PCR (Fig. 2B). MGAT3 encodes N-acetylglucosaminyltransferase-III (GnT-III), which is uniquely responsible for transferring a GlcNAc residue to the mannose of the trimannosyl core of N-glycans to produce a bisecting N-glycan³⁴ (Fig. 2C). Thus, the downregulation of the MGAT3 gene could be the transcriptional basis of the reduction of bisecting N-glycans observed in AβO treated hiMG cells.

Increased sialylation contributed to AβO-induced pro-inflammatory response

AβO treatment also increased the percent abundance of sialylated N-glycans (Fig. 1D), in particular, sialylated complex N-glycans (C-S in Fig. 1E). We therefore further explored the changes in the number of sialic acid residues on N-glycans and found that the increased sialylation was largely attributed to an increased percent abundance of di-sialylated N-glycans (Fig. 2D). DE analysis of RNA-seq data showed that among the genes encoding sialyltransferases and sialidases, ST3GAL2, ST3GAL4, and ST3GAL6 were significantly upregulated by AβO treatment (Fig. 2E). Given that ST3GALs encode for enzymes responsible for transferring sialic acid to the galactose residues with α− 2,3 linkage (Fig. 2C), the results suggest that α− 2,3-linked sialic acids possibly contributed to the increased di-sialylation observed in the AβO-activated hiMG. Of note, ST3GAL4 and ST3GAL6 preferably catalyze the sialylation on N-glycans³⁵, while ST3GAL2 exclusively catalyzes the sialylation of gangliosides and mucin-type O-glycans³⁶.

DE analysis and q-PCR also revealed increased levels of FUT8 transcript encoding α1–6 fucosyltransferase (fucosyltransferase 8 or FUT8), which is uniquely responsible for core fucosylation, in AβO-treated hiMG (Fig. 2A, F). This result corroborates our published data reporting that AβO enhances FUT8-catalyzed core fucosylation, which is a signaling pathway required for AβO-induced microglia activation²². We further co-treated AβO-activated cells with 2-fluoro-fucose (2 FF). 2 FF is a peracetylated derivative of L-fucose that can be converted to its corresponding nucleotide sugars, guanosine diphosphate (GDP)− 2-deoxy- 2-fluoro-L-Fucose, thus competing with and suppressing GDP-L-fucose, the natural substrate of fucosyltransferases. 2 FF particularly suppresses core fucosylation³⁷. We found that 2 FF alleviated AβO-induced overexpression of pro-inflammatory cytokine IL-1β and IL-6. Interestingly, 2 FF co-treatment also completely mitigated AβO-induced suppression of TGF-β in hiMG (Fig. 2G).

Additionally, to investigate the role of sialylation in AβO-induced hiMG activation, we inhibited sialylation by co-treating hiMG with 2,4,7,8,9-Penta-O-acetyl-N-acetyl-3-fluoro-β-D-neuraminic acid methyl ester (3 FS). Similar to the effects observed upon fucosylation inhibition, sialylation inhibition significantly reduced AβO-induced pro-inflammatory response, resulting in IL-1β and IL-6 expression levels returning to those of mock-treated cells, with no significant differences detected (Fig. 2G). 3 FS co-treatment also recovered the expression of TGF-β to the mock treatment level (Fig. 2G). As TGF-β signaling is required for microglia homeostasis and microglial amyloid-β clearance^38,39, our current results, together with our prior study on fucosyltransferase 8-catalyzed core fucosylation²², suggest that fucosylation and sialylation changes play a role in regulating microglial activation. A limitation of our current glycomic approach is that it cannot accurately distinguish core fucose from terminal fucose, and therefore cannot determine their respective changes following activation. However, our prior study supports a role of FUT8-catalyzed core fucosylation in AβO-induced microglial activation, based on various lines of evidence including FUT8 knockdown. Of note, both 2 FF and 3 FS also rectified the AβO-induced downregulation of MGAT3 and upregulation of ST3GAL4 and ST3GAL6 (Fig. 2H).

Decreased complex N-glycans in LPS-activated hiMG

In contrast to AβO-activated hiMG, LPS-activated hiMG exhibited a reduced percent abundance of complex-type N-glycans, specifically the tetra-antennary complex N-glycans (Fig. 1B-C). Complex N-glycans are derived from oligomannose N-glycans. In the endoplasmic reticulum (ER) and cis-Golgi, α-mannosidase I (MAN1B1), α1,2 mannosidases IA, IB, and IC (encoded by MAN1A1, MAN1A2, and MAN1C1, respectively) trim mannose residues from oligomannose N-glycans to yield Man5GlcNAc2, a key intermediate in the pathway to generate hybrid and complex N-glycans. Conversely, misfolded glycoproteins are degraded in the ER with the binding of ER degradation-enhancing α-mannosidase-like proteins (EDEM1 - 3). As a result, these proteins are not processed into complex N-glycans nor transported to the membrane^40,41. In the DE analysis of hiMG transcriptomic data, MAN1A1, MAN1C1, and MAN2A2 were significantly downregulated and EDEM1 - 3 were significantly up-regulated in LPS-treated hiMG (Fig. 3A). The transcript levels of other key mannosidases, including MAN1B1, MAN1A2, and MAN2A1, were also decreased though the differences did not reach statistical significance (Fig. 3A). The downregulation of these mannosidases and upregulation of EDEMs may contribute to the reduced abundance of complex N-glycans.

Increased fucosylation and decreased sialofucosylation associated with LPS-induced pro-inflammatory response

The changes in LPS-activated hiMG also differed from AβO-activated hiMG in terms of fucosylation and sialylation. The extent of fucosylation (i.e., fucosylated-only N-glycans), especially complex- and hybrid-type N-glycans, was increased in LPS-activated hiMG but remained unchanged in AβO-activated hiMG (Fig. 1D-E). This overall change was the sum of changes of individual glycan structures. Fucosylated-only N-glycans with one and two fucose residues increased in LPS-activated hiMG (Fig. 3B), which contributed to the overall increased percent abundance of fucosylated N-glycans.

Fucosyltransferases (FUTs) 3–7 and FUT9 are responsible for transferring fucose to antenna GlcNAc residues (Fig. 3C). According to the DE analysis and q-PCR, FUT4 was significantly upregulated in LPS-treated hiMG (Fig. 3D-E), which possibly contributed to increased fucosylation, especially increased di-fucosylated N-glycans. Fucose residues on N-glycans can be hydrolyzed by fucosidases (FUCAs). Alpha-L-Fucosidase 1 (FUCA1) specifically hydrolyzes the fucose residue at the core structure (Fig. 3C). The transcript level of FUCA1 was significantly downregulated as shown by RNA sequencing and q-PCR (Fig. 3D, F), suggesting less hydrolysis of core fucosylated N-glycans. This, together with enhanced FUT4 action, could lead to increased percent abundance of fucosylated N-glycans.

In contrast, sialofucosylated N-glycans, the most abundant types, were significantly reduced overall in LPS-activated hiMG, especially sialofucosylated complex N-glycans (Fig. 1D-E). The observed decrease in sialofucosylated N-glycans may be attributed to reduced activity of fucosyltransferases, sialyltransterases, fucosidases, neuraminidases, or all. The DE analysis highlighted that neuraminidase 4 (NEU4) was overexpressed in LPS-activated hiMG, potentially contributing to the decreased abundance of sialofucosylated N-glycans (Fig. 3D). However, the effects of sialytransferases were uncertain, as ST3GAL1 and ST3GAL2 were upregulated, while ST3GAL3 were downregulated in LPS-activated hiMG (Fig. 3D). Interestingly, among the sialofucosylated N-glycans those containing one fucose residue decreased while those containing two fucose residues increased (Fig. 3G). Meanwhile, sialofucosylated N-glycans containing two sialic acid residues decreased, consistent with the overexpression of neuraminidase (Fig. 3H). Thus, the decrease in sialofucosylated complex-type N-glycans was the primary effect of LPS treatment, specifically characterized by a decreased in mono-fucosylated and di-sialylated N-glycans and concomitant increase in di-fucosylated N-glycans.

To investigate the role of fucosylation and sialylation in LPS-activated hiMG, fucosylation and sialylation were inhibited using 2 FF and 3 FS, respectively. Both compounds significantly attenuated the LPS-induced upregulation of IL-1β, although the effect is mild (Fig. 3I). Additionally, 3 FS, but not 2 FF, mildly attenuated the LPS-induced upregulation of IL-6 and FUT4. Neither compound had any effect on LPS-induced downregulation of TGF-β or FUCA1 (Fig. 3I).

The glycosphingolipid (GSL) profile of hiMG

GSLs are a subclass of lipids composed of a ceramide part and a mono- or oligosaccharide part. In the adult mammalian brain, the major GSLs are the gangliosides of GM1, GD1a, GD1b, and GT1b that have one to three sialic acid residues respectively^42,43. Because sialic acid residues on a glycoprotein or glycolipid may modulate cell signaling mediated by lectins such as selectins and sialic acid-binding immunoglobulin-type lectins (Siglecs) and because we found significant changes in sialic acid levels following AβO and LPS treatment, we profiled the GSL composition as well. The most abundant GSL was ganglioside GM3, the precursor of other complex gangliosides (Fig. 4A). The second most abundant species was sphingomyelin (SM), a ceramide with a phosphocholine headgroup, followed by neutral GSL G2. The remaining GSLs were less abundant with each accounting for less than 5%. Regarding sialylation, GSLs detected in hiMG were predominantly sialylated GSLs (Fig. 4B). Interestingly, the GSL profile did not change significantly in response to either AβO or LPS activation. The PCA plot showed that the clusters of control, AβO, and LPS groups did not separate from each other (MANOVA p-value = 0.56, Fig. 4C, Supplementary Fig. 3B). The genes involved in GSL biosynthesis also did not show significant changes in response to AβO activation (Supplementary Fig. 4 A). In LPS-activated hiMG, even though the genes involved in GSL biosynthesis, i.e., UGCG, B4GALT5, and B4GALNT1, were significantly upregulated, no significant changes were observed when individual GSLs in either treatment were examined (Fig. 4A, Supplementary Fig. 4B).

Competing interests

The authors declare no competing interests.