Blog

Mice

All animal experiments were approved by the Ethical Review Board of the Government of Upper Bavaria. Mice were group housed with littermates in standard sized, individually ventilated cages on a 12-hour light/dark cycle, with enriched environment and ad libitum access to food and water. Both sexes were used for all experiments. APP-SAA^ki/ki x hTfR^ki/ki (APP-SAA x hTfR KI) mice [47,48,49] were acquired from Denali Therapeutics or bred in our mouse facility and maintained on a C57BL/6J genetic background. hTfR KI was bred into these mice in preparation of future antibody dosing studies that exploit antibody transport vehicle (ATV) technology, but was not investigated in the current study and was previously not found to impact microglia phenotypes in response to Aβ [49] (and Fig. S1). Shipped mice were acclimated for a minimum of two weeks before entering experiments. For anti-Aβ treatment, the chimeric anti-Aβ antibody Aducanumab was used, which contains a mouse IgG2 Fc domain with full effector function [31]. For isotype control, the antibody 4D5 was used, which has a mouse IgG2 Fc domain and is raised against human HER2, a non-existent target in mice [50]. Mice were randomly assigned to a treatment arm and two mouse cohorts underwent treatment. For cohort 1, isotype antibody was dosed at 1 mg/kg and for cohort 2, 10 mg/kg isotype antibody was dosed at 10 mg/kg. Anti-Aβ antibody was dosed at 1 mg/kg, 3 mg/kg or 10 mg/kg. Mice were treated from the average age of 4.48 ± 0.12 months (cohort 1) or 4.68 ± 0.14 months (cohort 2) via weekly intraperitoneal (i.p.) injection of antibody, which was thawed at 4 °C and diluted with phosphate-buffered saline (PBS). Mice from cohort 1 underwent FBB-PET and mice from cohort 2 were subjected to FDG-PET at 8 months of age. Mice were sacrificed by cardiac perfusion 7 days after the last antibody injection, at an average age of 8.26 ± 0.13 months (cohort 1) or 8.62 ± 0.15 months (cohort 2). From cohort 1, one hemibrain was fixed for immunofluorescent staining and another snap frozen for protein extraction (Fig. 1A). From cohort 2, terminal cerebrospinal fluid (CSF), blood plasma and microglia were collected for microglial RNA-seq and lipidomics, and CSF proteomic analysis.

Small animal PET/MRI

All rodent PET procedures followed an established standardised protocol for radiochemistry, acquisition times, and post-processing [51] which was transferred to a novel PET/MRI system [52]. In brief, [¹⁸F]-FBB-PET (florbetaben) and [¹⁸F]-FDG-PET (fluorodeoxyglucose) were used to measure fibrillar amyloidosis and glucose metabolism respectively after antibody treatment. We studied PET images of 8.12 ± 0.13-month-old APP-SAA mice (n = 32) for FBB-PET and 8.48 ± 0.20-month-old APP-SAA mice (n = 35) for FDG-PET using at least n = 8 per treatment group and tracer. All mice were scanned with a 3T Mediso nanoScan PET/MR scanner (Mediso Ltd., Budapest, Hungary) with a triple-mouse imaging chamber. Isoflurane anaesthesia was applied for all PET experiments (1.5% at time of tracer injection and during imaging; delivery 3.0 L/min). Two 2-minute-long anatomical T1 MR scans (sagittal and axial) were performed after tracer injection (head receive coil, matrix size 96 × 96 × 22 mm³, voxel size 0.24 × 0.24 × 0.80 mm³, repetition time 677 ms, echo time 28.56 ms, flip angle 90°). Injected dose was 13.1 ± 2.1 MBq for [¹⁸F]-FBB and 19.1 ± 1.5 MBq for [¹⁸F]-FDG delivered in 200 µl saline via venous injection. PET emission was recorded in a dynamic 0–60 min window for FBB-PET and in a static 30–60 min window for FDG-PET. List-mode data within 400–600 keV energy window were reconstructed using a 3D iterative algorithm (Tera-Tomo 3D, Mediso Ltd., Budapest, Hungary) with the following parameters: matrix size 55 × 62 × 187 mm³, voxel size 0.3 × 0.3 × 0.3 mm³, 8 iterations, 6 subsets. Decay, random, and attenuation correction were applied. The T1 image was used to create a body-air material map for the attenuation correction. Framing for FBB-PET was 6 × 10 s, 6 × 30 s, 6 × 60 s, 10 × 300 s.

All analyses were performed by using PMOD software (version 3.5, PMOD Technologies, Basel, Switzerland). To normalise FBB-PET data we generated V_T images with an image-derived input function [53, 54] using the methodology described by Logan et al. implemented in PMOD [55]. The plasma curve was obtained from a standardised voxel of interest (VOI) placed in the myocardial ventricle. A maximum error of 10% and a V_T threshold of 0 were selected for modelling of the full dynamic imaging data. Normalization of the injected activity for FDG-PET was performed by generating standardised uptake values (SUV), reflecting the common read-out in clinical setting. A cortical volume-of-interest (comprising 40.9 mm³) was selected and served for extraction of FBB-PET values. FDG-PET values were extracted from a bilateral entorhinal VOI (comprising 13.0 mm³) which was delineated by regions of the Mirrione atlas [56].

Mouse brain, CSF, and plasma sampling

CSF collection was performed as previously described from treatment cohort 2 [57]. Briefly, mice were anesthetised using a mix of medetomidine (0.5 mg/kg), midazolam (5 mg/kg), and fentanyl (0.05 mg/kg) (MMF) injected i.p. After complete anaesthesia, mice were head-fixed in a stereotaxic frame and the cisterna magna was surgically exposed. The dura was punctured using a borosilicate glass capillary (Sutter, B100-75-10) attached to medical-grade tubing and CSF was gently extracted by applying a negative pressure on the tubing using a syringe (equipped with a 28G needle). CSF samples were deposited from the capillary into protein Lo-Bind tubes (Eppendorf, 0030108094) and kept on ice until centrifugation at 2000 g for 10 min at 4 °C to pellet any red blood cells and visually check for contamination. After CSF collection, blood was extracted via cardiac puncture using a syringe, inserted into Microvette^® 500 EDTA K3E tubes (Sarstedt, 20.1341.100), slowly inverted 10 times and kept on ice. Within 1 h, blood was centrifuged at 12700 rpm at 4 °C for 10 min in a tabletop centrifuge. Plasma was then transferred to a protein Lo-Bind tube and snap-frozen. Mice were perfused via cardiac puncture with ice-cold PBS. For cohort 1, brains were split into two hemispheres and one hemisphere was fixed in 4% paraformaldehyde (PFA) with 0.05% NaN₃ for 48 h. The other hemisphere was snap-frozen in liquid nitrogen and stored at -80 °C. For cohort 2, brains were kept in Hanks’ buffered salt solution with Ca²⁺ and Mg²⁺ (HBSS) (Gibco, 14025092) + 7 mM HEPES (Gibco, 15630080) + 2x GlutaMAX (Gibco, 35050061) on ice until proceeding with microglia isolation.

Immunofluorescence staining of mouse brain

50-µm brain sections were cut using a vibratome and stored in 15% glycerol + 15% ethylene glycol in PBS for 2 days at 4 °C, before transferring them to a -20 °C freezer for long-term storage. For immunostaining, free-floating sections were washed 5x in PBS on a shaker to remove storage medium. Antigen retrieval was performed in citrate buffer (pH 6) or Tris-EDTA buffer (pH 8 or pH 9) at 80–95 °C for 30 min, depending on the antibody. After antigen retrieval, sections were cooled down to room temperature (RT), briefly washed in PBS and incubated in 10% normal donkey serum (NDS) in PBS + 0.3% Triton X-100 (blocking solution) on a shaker for 1–1.5 h. Section were incubated overnight in blocking solution containing primary antibodies. The next day, sections were washed 3x in PBS + 0.3% Triton X-100 and incubated in secondary antibodies in blocking solution for 1–2 h. In case of co-staining with Thiazine Red (Morphisto, 12990.001), the dye was added to the secondary solution, sections were washed 3x in PBS + 0.3% Triton X-100. For Methoxy-X04 (MX-04, Tocris, 4920) co-staining, sections were incubated in 50% EtOH in PBS with MX-04 for 30 min at RT, washed 5 min in 50% EtOH in PBS, and washed 3x in PBS. In case of HS169 (courtesy of Peter Nilsson, Linsköping University, Sweden) staining, dye was incubated 1:2500 in PBS for 15 min and washed 3x in PBS. If applicable, 40,6-diamidino-2-phenylindole (DAPI) was added to the secondary antibody solution (1:1000). Sections were mounted onto Superfrost Plus slides with ProLong Gold antifade reagent (Thermo Fisher, P36980) or Fluoromount-G (Thermo Fisher, 00-4958-02). After 24 h of drying, slides were stored at 4 °C.

Prussian blue staining of haemosiderin deposits in mouse brain

For quantification of haemosiderin deposits, slides were mounted onto Superfrost Plus slides and dried for 2 h at room temperature (RT). Slides were rehydrated in PBS and incubated in Prussian blue solution (2 g potassium hexacyanoferrate (II) trihydrate (Sigma, P9387) in 100 mL dH₂O) for 20 min and in 0.1% Nuclear Fast Red solution (Morphisto, 10264.00500) for 5 min and washed in dH₂O. Slides were dehydrated from 70 to 100% EtOH and mounted using VectaMount Express Mounting Medium (Vector Labs, VEC-H-5700). The number of Prussian Blue deposits was quantified from 5 brain sections of each mouse by stereology using the Leica DMi8 fluorescence microscope. Images of deposits were acquired using a 40x air lens (0.65 NA, Leica). Area and number of observed deposits was quantified from images using Fiji [60].

Microscopy and image acquisition

Epifluorescence images were acquired with a Leica DMi8 equipped with a mercury lamp (EL6000, Leica) using a 20x air lens (0.4 NA, Leica) or an Olympus VS200 Slideview slide scanner using a 20x air lens (0.8 NA, 0.274 μm/pixel). Leica scanned tiles were acquired using the Leica Application Suite X software using an overlap of 10% per image and a resolution of 1024 × 1024 (0.651 × 0.651 μm per pixel). Confocal images were acquired with a 63x oil immersion lens (1.4 NA, Zeiss), using a Zeiss LSM800 confocal microscope and the ZEN 2.5 Zeiss software package, at a resolution of 2048 × 2048 (0.0495 × 0.0495 μm per pixel).

Quantification of plaque number and microglia/plaque association

Image analysis was conducted blinded using a semi-automated ImageJ pipeline, where the user draws the outline of the region of the brain in each image to be analysed and inputs Gaussian filter values and thresholds for each channel. For each image, the pipeline then automatically applies a difference-of-Gaussian filter using Clij2 [61], followed by automated thresholding and subsequently measures total area and intensity of the selected channels. For individual plaque analysis, the total plaque region of interest (ROI) is split into individual ROIs, then using the ROI Manager, each ROI is given a unique name and subsequently area and intensity are measured for each plaque. For concentric ring analysis, the plaque ROI is enlarged and using logical operations (XOR) the original ROI is subtracted from the enlarged ROI to generate concentric rings with a user defined increase in size around the original selection (here 3 × 10 μm). Each concentric ring is given the same name as the original ROI they were generated from + a suffix to denote its increase in size. For each of these rings and the plaque ROIs, the total ROI size as well as selected channel area and intensity within these ROIs is measured. To quantify the area that a selected channel occupies in the vicinity of each plaque specifically within microglia, a threshold for Iba1 is set to obtain an ROI for the entirety of microglia. Then, using logical operators with the ROI manager (AND), ROIs corresponding to microglia colocalizing with plaque ROIs and concentric rings are obtained. Lastly, the total ROI size as well as selected channel area and intensity within these ROIs is measured. For each processed image a.csv file is created, which was subsequently processed, analysed and plotted using R (4.1.1) and R Studio (2024.09.0 + 375) [62]. For percent area calculations, thresholded signal area was divided by the total ROI area and multiplied by 100.

3D evaluation of plaque morphology and microglial clustering around plaques

For the evaluation of plaque size, sphericity and proximity of microglia and Aβ to plaques, 5 plaques per mouse were picked randomly and 63x confocal z-stacks were acquired along the cortex (z-distance 1.7 μm). First, images were deconvolved using point spread functions generated with the PSF generator Fiji plugin (Hagai Kirshner and Daniel Sage, Biomedical Imaging Group at EPFL) with the Born & Wolf optical model and 10 iterations of Richardson-Lucy deconvolution with the CLIJx plugin [61]. Then, using an automated Fiji script, a 3D difference-of-Gaussian filter was applied and images were made isotropic using Clij2. Then, using the 3D ROI manager [63] individual ROIs were imported from each microglia nucleus (based on PU.1⁺ nuclei) or Aβ (3552), from each thiazine⁺ plaque (excluding objects touching the image edges, as well as top and bottom z-slices) and from the total image volume. 3D measurements of each plaque (volume, sphericity, etc.), distance of each PU.1⁺ nucleus or Aβ ROI to each plaque and colocalization between each plaque and the total volume of microglia were obtained using 3D manager built-in functions. These measurements were exported as.csv files and further processed, analysed and plotted using R (4.1.1) and R Studio (2024.09.0 + 375). Measurements of 5 plaques per mouse were averaged and images where PU.1⁺ ROI separation was not achieved were excluded. Representative 3D isotropic images were made using napari [64].

Protein extraction

Whole hemispheres were lysed following a previously published protocol [65] and kept at 4 °C during all steps. Briefly, hemispheres were lysed in DEA buffer (0.2% diethylamine in 50 mM NaCl, pH 10, and protease inhibitor mix (Sigma, P8340) using the Precellys homogeniser in 2-mL Tissue Homogenizing CKmix tubes (Precellys, P000918-LYSK0-A). Lysate was centrifugated 10 min at 4000 g and supernatants were ultracentrifugated at 100 000 g before collection. Samples were neutralised by adding 10% of 0.5 M Tris-HCl buffer (pH 6.8) to each sample (DEA fraction). Remaining pellets in Precellys tubes were lysed in RIPA buffer (20 mM Tris-HCl (pH 7.5), 150 mM NaCl, 1 mM EDTA, 1% NP-40, 1% sodium deoxycholate, 0.1% SDS, and protease inhibitor mix). RIPA lysates were centrifuged 10 min at 4000 g, and the supernatants were ultracentrifuged at 100 000 g for 60 min before collection (RIPA fraction). The remaining material in Precellys tubes was resuspended in 70% formic acid with protease inhibitor mix and sonicated for 7 min. Samples were centrifuged at 20 000 g for 20 min and collected supernatant was diluted 1:20 in pH-neutralizing 1 M Tris-HCl buffer (pH 9.5) (FA fraction). Protein concentration was measured using Pierce Bicinchoninic acid (BCA) assay (Thermo Scientific, 23225).

Enzyme-linked immunosorbent assay (ELISA)

Aβ levels in FA fraction and CSF were determined using the Meso Scale Discovery (MSD) platform and the V-PLEX Plus Aβ Peptide Panel 1 (6E10) Kit (Meso Scale Discovery, K15200G). FA samples were diluted 1:10 in dilution buffer (Diluent Assembly 9), CSF was diluted 1:60. Cxcl10/IP-10 levels in DEA fraction were measured using the MSD U-PLEX Mouse IP-10 Assay (Meso Scale Discovery, K152UFK) at a dilution of 1:2.

TREM2 levels in DEA and RIPA fractions, as well as CSF, were measured using the MSD platform as described previously [66]. Briefly, MSD-gold Streptavidin-coated 96-well plates (Meso Scale Discovery, L15SA-1) were coated in 3% bovine serum albumin (BSA) + 0.05% Tween 20 in PBS (blocking buffer) overnight at 4 °C. Sample is diluted in 1% BSA + 0.05% Tween 20 in PBS + protease inhibitor mix (Sigma, P8340), 1:10 for DEA, 1:2 for RIPA, and 1:35 for CSF. The plate is incubated with 25 µL/well capture antibody in blocking buffer for 90 min, followed by 120 µL/well sample for 2 h, detection antibody 50 µL/well for 60 min and SulfoTAG antibody 25 µL/well for 60 min at 600 rpm at RT. In between each incubation the plate is washed 3x with 0.05% Tween 20 in PBS. Before read-out, the plate is washed 2x in PBS, 150 µL/well MSD Read buffer T (Meso Scale Discovery, R92TC-1) is added and is read immediately.

Western blot

4x Laemmli Buffer (Biorad, 1610747) + 10% β-mercaptoethanol was added to all samples. For Aβ immunoblotting DEA and FA lysates were loaded on Novex WedgeWell 10 to 20%, Tris-Tricine, 1.0 mm gels (Thermo Fisher, EC66255) and run in 1x Tris-Tricine-SDS buffer. For Trem2 and APP analysis, samples were run on 12% freshly cast Tris-Glycine gels in Tris-Glycine-SDS buffer. Protein was transferred to nitrocellulose membrane using wet transfer in Tris-Glycine buffer (25 mM Tris, 192 mM glycine, pH 7.5). Membranes were boiled in PBS for 15 min before blocking 1–2 h in 0.2% I-Block Protein-Based Blocking Reagent (Applied Biosystems, T2015) and 0.1% Tween 20 in Tris-buffered saline (TBS) (blocking buffer). Membranes were incubated in primary antibody in blocking buffer O/N at 4 °C while shaking. After 3 × 10 min washes in TBS + 0.05% Tween 20 (TBS-T) membranes were incubated with secondary antibody in blocking solution for 1 h at RT while shaking. Membranes were developed using Pierce ECL Western Blotting-Substrate (Thermo Scientific, 32106) and signals visualised using autoradiographic development using Fujifilm Medical X-ray Film Super RX-N (Fujifilm, 47410) or using the Amersham ImageQuant 800.

Magnetic-activated microglia sorting (MACS) from mouse brain

Prior to microglia isolation, meninges were removed by gently rolling brains on a clean piece of Whatman paper. Cerebellum, pons and olfactory bulb were removed, the two hemispheres were split and any remaining meninges were removed with Dumont forceps using a dissection microscope. Each hemisphere was cut into pieces using a scalpel and brain tissue was dissociated following manufacturer’s instructions using the Neural Tissue Dissociation Kit (P) (Miltenyi, 130-092-628) supplemented with 5 µM Actinomycin D (Cell Signaling Technology, 15021) and 2 µM Anisomycin (Cell Signaling Technology, 2222) in gentleMACS C-tubes (Miltenyi, 130-096-334) using a gentleMACS Dissociator (Miltenyi). Homogenised tissue was run through a 40-µm cell strainer (Corning, 352340) and pelleted by centrifugation. Pellets were resuspended in HBSS with 0.25% fatty acid-free BSA (Sigma-Aldrich, A8806), incubated with magnetic Cd11b⁺ MicroBeads (Miltenyi, 130-093-634) and run twice over MS columns (Miltenyi, 130-042-201). Viable cells were counted using trypan blue, aliquoted into tubes, centrifuged, and snap frozen in liquid nitrogen until further processing.

Sample preparation for mass spectrometry

CSF samples were prepared as described previously [57]. Briefly, a volume of 5 µL CSF was used for proteolytic digestion. Proteins were reduced by the addition of 2 µL of 10 mM dithiothreitol (Biozol, Germany) in 50 mM ammonium bicarbonate and incubated for 30 min at 37 °C. Cysteine residues were alkylated by the addition of 2 µL 55 mM iodoacetamide (Sigma Aldrich, US) and incubated for 30 min at room temperature in the dark. Afterwards, the reaction was quenched by adding another 2 µL of 10 mM dithiothreitol. Proteolytic digestion was performed using a modified protocol for single-pot solid-phase enhanced sample preparation (SP3) [67]. After binding proteins to 40 µg of a 1:1 mixture of hydrophilic and hydrophobic magnetic Sera-Mag SpeedBeads (GE Healthcare, US) with a final concentration of 70% acetonitrile for 30 min at room temperature, the beads were washed four times with 200 µL 80% ethanol. For proteolytic digestion, 0.1 µg LysC and 0.1 µg trypsin (Promega, Germany) were added in 20 µL 50 mM ammonium bicarbonate followed by an incubation for 16 h at room temperature. The magnetic beads were retained in a magnetic rack and the supernatants were filtered with 0.22 μm spin filters (Spin-X, Costar) to remove remaining beads and dried by vacuum centrifugation.

Liquid chromatography tandem mass spectrometry (LC-MS/MS) of CSF

Dried peptides were dissolved in 20 µL 0.1% formic and 5.5 µL were separated on a nanoElute nanoHPLC system (Bruker, Germany) on an in-house packed C18 analytical column (15 cm × 75 μm ID, ReproSil-Pur 120 C18-AQ, 1.9 μm, Dr. Maisch GmbH) using a binary gradient of water and acetonitrile (B) containing 0.1% formic acid at flow rate of 300 nL/min (0 min, 2% B; 2 min, 5% B; 62 min, 24% B; 72 min, 35% B; 75 min, 60% B) and a column temperature of 50 °C. The nanoHPLC was online coupled to a timsTOF Pro mass spectrometer (Bruker, Germany) with a CaptiveSpray ion source (Bruker, Germany). A Data-Independent Acquisition Parallel Accumulation-Serial Fragmentation (diaPASEF) method was used for spectrum acquisition. Ion accumulation and separation using Trapped Ion Mobility Spectrometry (TIMS) was set to a ramp time of 100 ms. One scan cycle included one TIMS full MS scan with 26 windows with a width of 27 m/z covering a m/z range of 350–1001 m/z. Two windows were recorded per PASEF scan. This resulted in a cycle time of 1.4 s.

Mass spectrometry data analysis

The software DIA-NN version 1.8.1 was used to analyse the data [68]. The raw data was searched against a one-protein-per-gene database from Mus musculus (UniProt, 21709 entries, download: 2024-02-19) combined with a database of common human contaminations (123 entries) using a library-free search. Trypsin was defined as protease and two missed cleavages were allowed. Oxidation of methionines and acetylation of protein N-termini were defined as variable modifications, whereas carbamidomethylation of cysteines was defined as fixed modification. The precursor and fragment ion m/z ranges were limited from 350 to 1001 and 200 to 1700, respectively. An FDR threshold of 1% was applied for peptide and protein identifications. The mass accuracy and ion mobility windows were automatically adjusted by the software. The match between runs option was enabled.

The statistical analysis was performed with the software Perseus version 1.6.2.3 [69]. First, a one-way ANOVA was used to determine statistically significant differences between the means of the groups. Afterwards, individual Student’s t-tests were applied to evaluate proteins with a significant abundance difference between 1, 3, and 10 mg/kg anti-Aβ compared to isotype control treatment. Additionally, isotype control samples were compared with sample from 3-month-old untreated mice. A permutation-based false discovery rate estimation was used with a FDR of 5% at s0 = 0.1 as threshold [70].

RNA isolation, RT‑qPCR, and library preparation

To prepare for RNA-seq, approximately 100 000 CD11b⁺ microglia isolated by MACS were used for RNA extraction by the RNeasy Plus Micro Kit (Qiagen, #74034). The extracted RNA was then resuspended in nuclease-free water for RNA-seq library preparation. Libraries for 30 total RNA samples were prepared using the Lexogen QuantSeq 3′ mRNA-Seq V2 Library Prep Kit FWD with Unique Dual Indices (Lexogen 193.384) and the UMI Second Strand Synthesis Module, following the manufacturer’s protocol to identify and remove PCR duplicates. In brief, total RNA was used as input for oligo(dT) priming during reverse transcription, followed by RNA removal. Unique Molecular Identifiers (UMIs) were incorporated during second-strand synthesis. The cDNA was purified using magnetic beads, amplified with 18 cycles of PCR, and subsequently purified again. Library quantity and quality were assessed using a TapeStation D1000 ScreenTape (Agilent 5067–5582). Equimolar pooling of libraries was performed, and sequencing reads were generated on one lane of an Illumina NovaSeq X 10B cartridge (75 bp single-end) by SeqMatic (Fremont, CA, USA).

RNA-seq data analysis

RNA-seq data was processed using nf-core/rnaseq v3.11.2 (https://doi.org/10.5281/zenodo.1400710) of the nf-core collection of workflows [71]. Reads were aligned to the GRCm39 release of the mouse genome, and gene annotations were obtained from Gencode M31. To account for the use of UMIs in the library preparation protocol, the following arguments were passed to the STAR aligner (version 2.7.9a [72]),: –alignIntronMax 1,000,000 –alignIntronMin 20 –alignMatesGapMax 1,000,000 –alignSJoverhangMin 8 –outFilterMismatchNmax 999 –outFilterType BySJout –outFilterMismatchNoverLmax 0.1 –clip3pAdapterSeq AAAAAAAA. After alignment, UMIs were extracted with the following regular expression: ^(?P.{6})(?P.{4}).*. As each transcript is only represented by a single sequence, the –noLengthCorrection parameter was passed to the salmon (version 1.10.1 [73]) gene-level quantitation step. The pipeline was executed with Nextflow v23.10.0 [74]. Downstream analysis was performed using R (version 4.4.0) using the limma/voom workflow [75] to fit linear models for each quantifiable gene. Library sizes were estimated using the TMM method [76] and we fit a linear model with treatment group and sex as fixed covariates, and takedown-batch as a random effect with the voomLmFit function from the edgeR R package (version 4.2.0 [75]),. Sample weights were included by setting the sample.weights argument to TRUE. Differentially expressed genes were identified with the eBayes function from the limma R package (version 3.60.0 [77]),, setting the robust = TRUE argument. P-values were corrected for multiple-testing according to [78]. Gene set enrichment analyses were performed with the fgsea R package (version 1.30.0 [79]),, with gene sets obtained via the msigdbr package (7.5.1, doi: https://doi.org/10.32614/CRAN.package.msigdbr).

Lipid extraction

Cell pellets (100 000 MACS-sorted cells) were suspended in 400 µL of a 3:1 butanol/methanol extraction buffer with stable isotope-labeled internal standards and mixed for 5 min at 600 rpm on a plate shaker at room temperature. Plates were stored for one hour at -20 °C and centrifuged at 21 000 g for 5 min at 4 °C. After centrifugation, 200 µL of the supernatant was collected and dried under a continuous stream of nitrogen gas. The dried extracts were reconstituted in 200 µL of LC-MS-grade methanol for subsequent analysis.

LC-MS analysis of lipids

Lipid analysis was performed using an Agilent Infinity II 1290 UHPLC coupled with a QTRAP 6500 + mass spectrometer. Lipids were analysed in both positive and negative ionization modes and resolved on a UPLC BEH C18 column (150 × 2.1 mm, 1.7 μm, Waters Corp.) at 55 °C with a 0.25 mL/min flow rate, following the buffer and gradient schedule as described previously [80]. Data acquisition, peak integration, and quantification were conducted using MultiQuant (version 3.3, ABSciex) with a minimum signal-to-noise ratio of 5 and at least 8 points across the baseline.

Statistical analysis

Unless indicated otherwise in the methods, statistical analysis was performed in R studio (R version 4.2.3) [62]. Data are shown with the mean and standard error of the mean (± SEM), unless indicated otherwise. For normally distributed data a one-way ANOVA was applied. Statistical evaluations are displayed as follows: *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001. Graphs were plotted using the tidyverse package and statistical significance was plotted using the ggsignif package [81, 82].