The ABC transporter A7 modulates neuroinflammation via NLRP3 inflammasome in Alzheimer’s disease mice

The Abca7^{tm1.1(ABCA7)Pahnk} allele (MGI: 6258226) was bred in homozygosity on the C57BL/6 background (B6J.Cg-Abca7^{tm1.1(ABCA7)Pahnk}). The constitutive knockouts were generated by crossing this line with the Cre-expressing line B6.C-Tg^CMV−cre1Cgn/J (JAX, strain 006054). The resulting heterozygous knockouts were bred back to homozygosity by negatively selecting the Tg^CMV−cre1Cgn allele until extinction. The floxed-allele and the knockout-bearing lines were independently crossed to the APPPS1-21 line (B6.Cg-Tg^{Thy1−APPSw, Thy1−PSEN1*L166P}/21JkcrPahnk). In both cases, the Abca7 allele (either floxed or knocked out, named APPPS1-hA7^flx and APPPS1-hA7^ko, respectively) was further bred back to homozygosity, whereas the APPPS1 transgene (Tg^{Thy1−APPSw, Thy1−PSEN1*L166P}/21Jckr) was maintained at hemizygosity (+/Tg). APPPS1-21 (named APPPS1) mice have combined APP (Swedish mutation) and PS1 (L166P mutation) transgenes under the control of the Thy1 promoter, leading primarily to pathological Aβ production in fronto-cortical neurons and the first cortical Aβ plaques at 45–50 days of age [29]. The constitutive knockouts were generated by crossing individuals with the APPPS1 transgene and the floxed Abca7^{tm1.1(ABCA7)Pahnk} allele (hA7) to the B6.129P2(C)-Cx3cr1^{tm2.1(creERT2)Jung/}J line (JAX, strain 020940) (named APPPS1-Cx3cr1-hA7^ko). Once more, the floxed Abca7^{tm1.1(ABCA7)Pahnk} allele was bred back to homozygosity. In experimental animals treated with tamoxifen, the Cx3cr1^creERT2 targeted mutation (Cx3cr1^{tm2.1(cre/ERT2)Jung}) was present in heterozygosity (+/tm). As breeders, the sires were always selected to carry the APPPS1 transgene, whereas the Cx3cr1^creERT2 mutation may be inherited from either the sires or the dams.

The animals were housed at the Department of Comparative Medicine (section Radium Hospital) at the Oslo University Hospital (Norway) at a temperature of 22 ± 1 °C, relative humidity of 62 ± 5%, 15 air changes per hour, and light cycles of 12 h dark/light (strength: 1 lx - night, 70 lx - day, 400 lx - working illumination) in Eurostandard type III cages (Makrolon^®) filled with aspen wood (Populus tremula, Tapvei^®, Estonia) as bedding substrate and provided with additional enrichment material (tissue paper, tunnel rods and occasionally gnawing sticks). The animals were grouped into cohorts of up to 8 individuals per cage. Mice were fed ad libitum (maintenance expanded pellets from SDS, Estonia) and offered water acidified to pH 3. Health monitoring was performed three times per year according to the FELASA guidelines, and opportunists were included in one of the tests. All experiments were conducted following the guidelines for animal experimentation of the European Union and Norwegian national laws.

To genotype the new ABCA7 locus, we designed primers for the identification of the recombinant, wild-type and knockout sequences (Table S1). PCR cycling was performed as follows: 5 min at 95 °C; 35 cycles of 45 s – 95 °C, 60 s – 62 °C, and 90 s – 72 °C; 5 min of elongation of the final products at 72 °C; and cooling at 4 °C until further use. The genotyping PCR showed a clear differentiation between all possible genotypes: homozygous ABCA7 recombinant (tm/tm), heterozygous Abca7/ABCA7 recombinant (+/tm), constitutive Abca7 knockout (−/−, ko), Abca7 haplodeficient (+/−), and wild-type (+/+) animals (Figure S1).

The tamoxifen induction technique involves the use of a microglia-specific Cre recombinase (Cx3cr1-CreERT2), which can be activated by tamoxifen through the known Cre/loxP site-specific recombination system and the generation of a specific microglial ABCA7 transporter. Tamoxifen was applied twice within 48 h at the age of 4 weeks. Hereby, an emulsion of 4 mg of tamoxifen (originator AstraZeneca, purchased at Merck KGa, Darmstadt, Germany) in 200 µL of corn oil (Merck KGaA) was injected subcutaneously in four 50 µL depots. The injection was carried out under general anesthesia (sevoflurane 3.5% in an oxygen/nitrogen mixture).

Mice were euthanized by a ketamine/xylazine overdose of 400 mg/kg ketamine (VetViva Richter GmbH, Wels, Austria) and 40 mg/kg xylazine (Bayer, Barmen, Germany) at 100 and 200 days of age. After intracardial perfusion with ice-cold PBS, the brains were removed and separated into two hemispheres. One hemisphere was kept in paraformaldehyde (PFA 4% in PBS for immunohistology), and the other hemisphere was snap frozen in liquid nitrogen and later transferred to − 80 °C (for protein and RNA extraction).

Brain tissue samples from APPPS1 and APPPS1-hA7^ko mice (n = 14 / 15) mice were used as biological replicates for proteomics analyses. For each replicate equal amounts of brain homogenate (appr. 20 µg of protein) were precipitated on amine beads as previously described [30]. The precipitated proteins on beads were dissolved in 50 mM ammonium bicarbonate, reduced, alkylated and digested with trypsin (1:50 enzyme: protein ratio; Promega, USA) at 37 °C overnight. Digested peptides were acidified and desalted on EVOTIPs using a standard protocol from EVOSEP (EVOSEP Biosystems, Denmark). Liquid chromatography with tandem mass spectrometry (LC-MS/MS) analysis was carried out using an EVOSEP one LC system coupled to a timsTOF pro2™ mass spectrometer, using a CaptiveSpray nanoelectrospray ion source (Bruker Daltonic GmbH, Bremen, Germany). 200 ng of digested peptides were loaded onto a capillary C18 column.15 cm length, 150 μm inner diameter, 1.5 μm particle size, EVOSEP, Odense Denmark).). Peptides were separated at 50 °C using the standard 30 sample/day method from EVOSEP. The timsTOF pro2™ mass spectrometer was operated in data-dependent Parallel Accumulation-Serial Fragmentation (PASEF^®) mode [31]. Mass spectra for MS and MS/MS scans were recorded between m/z 100 and 1700. Ion mobility resolution was set to 0.85–1.40 V·s/cm over a ramp time of 100 ms. Data-dependent acquisition was performed using four PASEF MS/MS scans per cycle with a near 100% duty cycle. A polygon filter was applied in the m/z and ion mobility space to exclude low m/z, singly charged ions from PASEF precursor selection. An active exclusion time of 0.4 min was applied to precursors that reached 20,000 intensity units. Collisional energy was ramped stepwise as a function of ion mobility. Raw data files from LC-MS/MS analyses were submitted to MaxQuant software (version 2.4.3.0, Max-Planck-Institute of Biochemistry, Martinsried, Germany) for protein identification and quantification [32]. The UniProt mouse database (UniProt Consortium, European Bioinformatics Institute, EMBL-EBI, UK) was used. Trypsin without proline restriction enzyme option was used, with two allowed miscleavage sites. Carbamidomethyl was set as a fixed modification and acetyl (protein N-term), carbamyl (N-term) and oxidation (M) were set as variable modifications. First search peptide tolerance of 20 ppm and main search error 4.5 ppm were used. The allowed FDR was 0.01 (1%) for peptide and protein identification. Label-free quantitation (LFQ) was employed with default settings.

Microglia were isolated from 100-day-old APPPS1 and APPPS1-hA7^ko mice. Briefly, brains were homogenized in a glass potter with dissection buffer (1x HBSS, 45% glucose, 1 M HEPES). After centrifugation, the pellet was resuspended in 70% Percoll, which was covered with a layer of 30% Percoll. The gradient was centrifuged (45 min, 800 × g, 4 °C), and myelin debris in the interphase between 25% Percoll (Cytiva Freiburg, Germany) and PBS was removed. The collected cells were resuspended in PB buffer (1x PBS, 0.5% BSA) and centrifuged again (10 min, 300 × g, 4 °C). Once the supernatant was discarded, the pellet was resuspended in PB buffer, and CD11b microbeads (Miltenyi Biotec Norden AB, Lund, Sweden) were added. After 15 min of incubation in the dark and under cold conditions, CD11b⁺ microglia were separated using a DynaMag-2 magnetic separator (Invitrogen, Waltham, MA, USA).

The frozen hemispheres were thawed on ice in 500 µL of RNAlater^® (Merck KGaA, Darmstadt, Germany) for one hour and homogenized for 60 s with four 2.8 mm ceramic beads (OMNI International, Kennesaw, GA, USA) using a SpeedMill PLUS (Analytik Jena GmbH, Jena, Germany). A total of 20 mg of homogenate was mixed with 10 µl of ice-cold Tris-buffered saline [TBS, pH 7.5, containing protease inhibitor (Roche, Mannheim, Germany)] per 1 mg of brain tissue. The samples were homogenized with 2.8 mm ceramic beads (SpeedMill PLUS, 30 s) and centrifuged (16,000 g, 4 °C, 20 min) to separate soluble and aggregated Aβ. The resulting supernatant (TBS fraction containing soluble Aβ) was collected and stored at − 20 °C until further use. The pellet was mixed with 8 µl of cold 5 M guanidine-HCl buffer (pH 8.0) per 1 mg of brain homogenate and homogenized (SpeedMill PLUS, 30 s). The samples were incubated at room temperature for 3 h under constant shaking (1,500 rpm) before centrifugation (16,000 × g, 4 °C, 20 min). The supernatant (the GuHCl fraction containing aggregated Aβ) was collected and stored at − 20 °C until further use. To quantify Aβ₄₂ in TBS/GuHCl fractions and CD11b⁺ cells (diluted in GuHCl buffer), we performed electrochemiluminescence immunoassays using the V-PLEX Plus Aβ₄₂ Peptide (4G8) Kit (detection limit above 516 fg/ml) and a MESO QuickPlex SQ120 machine according to the manufacturer's recommendations (Meso Scale Diagnostics LLC, Rockville, MD, USA) [33 –36]. The results were normalized to the sample weight. The brain Aβ₄₂ concentration was calculated as pg/mg brain.

Formalin-fixed hemispheres were embedded in paraffin and cut into 4-µm-thick coronal sections using a rotation microtome (HM355S, Leica Biosystems GmbH, Nussloch, Germany). Sections at bregma + 0.8 mm and − 1.8 mm were stained for Aβ (anti-human Aβ clone 4G8; 1:4000, BioLegend, USA), microglia (anti-IBA1, 1:1,000, FUJIFILM Wako Chemicals Europe GmbH, Germany, 019-19741) and astrocytes (anti-GFAP, 1:500, Agilent, USA, Z033401-2) using a BOND-III^® automated immunostaining system (Leica Biosystems GmbH, Nussloch, Germany) with a hematoxylin counterstain (provided with the staining kit, Bond Polymer Refine Detection, DS9800) [37, 38]. After staining, tissue sections were digitized at 230 nm resolution per pixel using a Pannoramic MIDI II slide scanner (3DHISTECH Ltd., Budapest, Hungary) [39]. Quantitative analysis was performed automatically using deep learning algorithms generated with DeePathology™ STUDIO (DeePathology Ltd., Ra'anana, Israel) using algorithms previously established by the laboratory [40, 41]. The cell/plaques density (number per ROI), relative area coverage (% cell area per total ROI), and cell size were determined for each animal.

Total RNA was extracted from primary cell culture pellets or brain tissue homogenates using an RNeasy Mini Kit and QIAzol as a lysis reagent (Qiagen, Hilden, Germany). The total amount of RNA extracted was quantified with a NanoDrop 1000 (Thermo Fisher, Waltham, MA, USA) by absorbance at 260 nm, and its purity was calculated as the ratio between the absorbance values at 260/280 nm and 230/260 nm. cDNA was synthesized from 1 µg of total RNA using the commercial RNeasy Mini QuantiTect Reverse Transcription Kit (Qiagen, Hilden, Germany). Quantitative real-time PCR assays were performed using TaqMan Gene Expression Assays (Applied Biosystems, Foster City, CA, USA) to quantify the mRNA levels of Nlrp3 (Mm00840904_m1), Pycard (Mm00445747_g1), Casp1 (Mm00438023_m1), Tlr4 (Mm00445273_m1), Cd36 (Mm00432403_m1), Nfkb1 (Mm00476361_m1), P2rx7 (Mm01199500_m1), Ctsb (Mm01310506_m1), and Gapdh (Mm99999915_g1) as endogenous controls for normalization. Quantitative PCR was performed using the StepOne Plus Real Time PCR System (Applied Biosystems, Foster City, CA, USA), and the threshold cycle (Ct) was calculated by the instrument's software (v2.4, Sequence Detection, Applied Biosystems, Foster City, CA, USA). The expression levels were calculated using the 2^−ΔΔCt method.

For protein extraction, 30 mg of brain homogenate tissue was lysed in ice-cold RIPA buffer (50 mM Tris-HCl, pH 8; 150 mM sodium chloride; 1 mM EDTA; 1% Triton X-100; 0.5% sodium deoxycholate; 0.1% sodium dodecyl sulfate) supplemented with complete protease inhibitor cocktail (Roche, Mannheim, Germany). The final protein concentration in the supernatants was determined using a Pierce BCA assay kit (Thermo Fisher, Waltham, MA, USA), and a total of 25 µg of protein was subjected to 4 − 15% Mini-PROTEAN polyacrylamide gel electrophoresis (Bio-Rad Laboratories, Hercules, CA, USA) and transferred onto a polyvinylidene difluoride (PVDF) membrane (Bio-Rad Laboratories) using a mini Trans-Blot Electrophoretic Transfer Cell (Bio-Rad Laboratories). The membranes were then blocked for 1 h at room temperature with Tris-buffered saline containing 5% nonfat dry milk and 0.1% Tween-20 and incubated with anti-caspase1 (1:1,000; clone EPR16883, Abcam, Cambridge, UK, ab179515) and anti-beta-tubulin (1:2,500; clone AA2, MerckMillipore, Burglinton, MA, USA, T5076) antibodies overnight at 4 °C. The membranes were finally incubated with an enhanced chemiluminescence (ECL) horseradish peroxidase-conjugated secondary antibody (1:5,000, Bethyl Laboratories, Montgomery, TX, USA, A90-216P) for 1 h at room temperature. Images were obtained by using the Octoplus QPLEX imaging system after the membranes were incubated with Clarity Max ECL substrate reagent (Bio-Rad Laboratories).

Brain homogenates were extracted using lysis buffer (phosphate-buffered saline + 1% Triton X-100 + phosphatase inhibitor + protease inhibitor) at 2 µL/mg brain homogenate. The samples were homogenized for 30 s (SpeedMill PLUS, Analytic Jena GmbH, Jena, Germany) and centrifuged at 16,000 × g for 40 min at 4 °C. Supernatants were diluted 4-fold (for brain tissue samples) or 2-fold (for cell culture medium) using a working solution (Meso Scale Diagnostics LLC, Rockville, MD, USA), and immunoassays were performed using the V-PLEX Proinflammatory Panel 1 (mouse) Kit (detection limit above 40 fg/ml) and a MESO QuickPlex SQ120 machine following the manufacturer's instructions.

Pure astrocyte or microglial cell cultures were obtained from APPPS1 and APPPS1-hA7^ko pups (P0-P5), respectively, using the protocol previously published by Güler et al. [42] with some modifications (Figure S4A). In brief, brains were collected in 1× HBSS buffer (Gibco, Grand Island, NY, USA) and homogenized with DNase I (Merck KGaA, Darmstadt, Germany) and 0.05% trypsin-EDTA (Thermo Fisher Scientific - Gibco, Grand Island, NY, USA) after meningeal removal. After centrifugation (10 min, 150 × g), the pellet was resuspended in DMEM (for astrocytes) or DMEM/F12 (for microglia) (Gibco, Grand Island, NY, USA) supplemented with 10% FBS (Cytiva, Freiburg, Germany) and 2% penicillin/streptomycin (Gibco, Grand Island, NY, USA) and transferred to a T-25 flask that was incubated in a humidified atmosphere of 5% CO₂ at 37 °C. When the cultures reached confluence (7 − 10 days), the OLs were removed by flask tapping, and the remaining cells were removed from the flask using 0.05% trypsin-EDTA. After centrifugation (10 min, 150 × g), the pellet containing a mixture of microglia and astrocytes was resuspended in complete DMEM or DMEM/F12, and the cells were separated following consecutive incubation steps on sterile bacterial-grade plates. When the derived APPPS1-hA7^ko microglial cultures reached confluence (≈ 2 × 10⁶ cells), they were incubated in DMEM/F12 supplemented with 1% FBS. Three hours later, the cells were treated with 100 ng/ml LPS (from E. coli 055:B5, Merck KGaA - Sigma‒Aldrich, San Louis, MS, USA) for 24 h. Then, the media was removed to be added to the APPPS1 astrocyte cultures (≈ 2 × 10⁶ cells) to induce an inflammatory reaction via the conditioned media. Twenty hours after the addition of conditioned media stimulated with LPS, astrocyte and microglia pellets and medium were collected for qPCR and cytokine ELISA, respectively.

All the statistical analyses were performed using GraphPad Prism software (v9, Dotmatics, Boston, MA, USA). We verified that the data had a normal Gaussian distribution by using the Shapiro‒Wilk normality test. If all groups were normally distributed (p > 0.05), we performed one-way ANOVA with Tukey's correction for multiple comparisons. Student's t test was also performed to determine the significant differences between the two experimental groups. The data are presented as the means ± standard deviations (SD). Differences were considered statistically significant when p < 0.05. The number of animals used in each experiment is reported in the figure legends. LFQ proteomic data were analyzed using Perseus (v2.0.11, https://cox-labs.github.io/coxdocs/perseus_instructions.html↗) [43] and VolcaNoseR (https://github.com/JoachimGoedhart/VolcaNoseR↗) [44], which plot the fold change (log₂ transformation) versus the p value (− log₁₀ transformation) for all the quantified proteins. This generates a so-called volcano plot, which shows a measure of effect size versus a measure of significance. The data points with the largest effect size and a statistical significance threshold > 1.3 (corresponding to p < 0.05) were considered hits. These protein hits are annotated in the plot with an abbreviated name (mouse nomenclature).

The design and generation of the new humanized ABCA7 mouse line were performed in close collaboration with genOway SA (Lyon, France). Humanization of the murine Abca7 gene locus was accomplished by inserting human type I ABCA7 cDNA starting from human exon 2 into the 5' end of mouse Abca7 exon 4 (Table S2). Thus, we avoided deregulation of neighboring genes while disrupting the Abca7 murine gene locus. Additionally, two loxP sites were inserted upstream of Abca7 exon 1 and downstream of the ABCA7 cDNA. The insertion of these sites enables the targeted deletion of ABCA7 cDNA using a Cre recombinase, generating conditional or constitutive knockout models (Fig. 1). The vector was purified and inserted into C57BL/6 N embryonic stem (ES) cells by electroporation. The allele was registered as Abca7^{tm1.1(ABCA7)Pahnk} in the Mouse Genome Informatics database (MGI:6258226).

ES cell clones were selected and assessed for correct homologous recombination and insertion of the ABCA7 cDNA. To validate the insertion of ABCA7 cDNA in the targeted Abca7 locus at exon 4 and assess the deletion of the ABCA7 insert in Cre-recombined knockout mice (quality control test), Southern blot analyses for homolog and induced recombination were performed with tissue samples from Abca7 wild-type controls, ABCA7^flx/flx, and ABCA7^ko mice (Fig. 2).

To investigate the impact of the deletion of the humanized ABCA7 transporter in our AD model, we analyzed the brain proteome of chimeric APPPS1-hA7^ko (n = 8 males, 7 females) and APPPS1 mice carrying the wild-type mouse Abca7 gene (n = 7 males, 7 females) from 50 to 200 days of age by LC‒MS/MS (DDA). Principal component analysis (PCA), which defines the variation within experimental groups, revealed that 29.3% of the variability between APPPS1-hA7^ko and APPPS1 was due to the lack of the ABCA7 transporter, separating the mice into two distinct clusters (Fig. 3A). Based on the volcano plot data, we found a total of 166 differentially expressed proteins in APPPS1-hA7^ko mice compared to controls, with log₂-fold change values between ± 1.5 and significance values (− log₁₀) greater than 1.3 (corresponding to p = 0.05). Among all hits, 84 proteins were upregulated and 82 proteins were downregulated in the absence of the humanized ABCA7 transporter (Fig. 3B).

Proteomic hits were classified according to pathway involvement using the KEGG 2021 Human database embedded in the Enrichr bioinformatics resource [45 –47]. We found that hABCA7^ko induced the upregulation of several proteins involved in NFκB signaling, one of the canonical pathways involved in the control of inflammation, among other processes. Dysregulation of ubiquitin-related proteolysis and the TGFβ signaling pathway was also found, which might be closely related to the toxic aggregation of Aβ and the consequent apoptotic process (Fig. 3C).

Next, we performed unbiased functional enrichment analysis (Gene Ontology, GO) of the differentially expressed proteins found in the volcano plot using the Enrichr database. In this case, the absence of the ABCA7 transporter in the APPPS1 mouse model was associated with changes in biological processes related to inflammation control and regulation. Positive regulation of cellular senescence (GO: 2000774; p value = 0.00029) was the most enriched biological process. The inflammatory response is one of the hallmarks of cellular senescence, and it serves to both propagate the senescence process and facilitate its clearance [48, 49]. A fairly recent publication identified the expression of senescence-related genes specifically in TREM2⁺ microglia in the 5xFAD β-amyloidosis model, highlighting the link between these two degenerative processes [50]. Other biological processes enriched and directly related to the modulation of the immune response in the CNS were the IL1-mediated signaling pathway (GO: 0070498; p value = 0.00105) and the positive regulation of T-cell cytokine production (GO: 0002726; p value = 0.00128). The enrichment of these biological processes could be related to increased cytokine release, which also affects other cellular functions. One example is the negative regulation of mTOR signaling (GO: 0032007; p value = 0.00100), which is involved in the control of other key cellular mechanisms that are altered in AD, such as autophagy or reactive oxygen species (ROS) generation (Table 1).

Based on the proteome findings, we investigated whether ABCA7 is a key player in neuroinflammation control in AD. To this end, we measured Aβ₄₂ levels in CD11b⁺ microglia isolated from the brains of 100-day-old APPPS1 expressing the wild-type ABCA7 protein and chimeric APPPS1-hA7^ko mice. While the absence of the transporter in female mice leads to a significant decrease in Aβ₄₂ levels, in APPPS1-hA7^ko males, the levels of Aβ₄₂ were increased compared to those in APPPS1 control mice (Figure S2). This data show how ABCA7 excerts sex-specific roles in microglial Aβ clearance. Whereas in males its absence seems to be protective by increasing the uptake of the toxic peptide, in females there is a loss of the transporter's protective role when it is functional.

To further investigate the link between microglia and the ABCA7 transporter, we developed a conditional knockout model of the ABCA7 transporter in Cx3cr1⁺ microglia, inducible by tamoxifen treatment, crossbred with the APPPS1 model. From these breedings, we finally got APPPS1-hA7^flx as the control group, the chimeric APPPS1-hA7^ko mice and the conditional APPPS1-Cx3cr1-hA7^ko mice. First, we checked the status of β-amyloidosis by measuring Aβ₄₂ levels. Human Aβ₄₂ is preferentially generated over Aβ₄₀ in our AD mice model [29]. By immunoassay, we observed that soluble Aβ₄₂ increased in APPPS1-hA7^ko mice at 100 days of age compared to that in age-matched APPPS1-hA7^flx controls, but in the conditional Cx3cr1-hA7^ko group peptide levels returned to control APPPS1-hA7^flx levels in both males and females (Fig. 4A-B). This indicate that selective loss of the ABCA7 transporter in microglia does not affect total soluble Aβ₄₂ levels, even when microglial Aβ clearance it is affected, as has been described before (Figure S2). Sex differences are also evident in insoluble peptide levels. In females, aggregated Aβ₄₂ increased in APPPS1-hA7^ko mice and returned to control levels in APPPS1-Cx3cr1-hA7^ko mice, whereas in males, there were no significant differences between the experimental groups (Fig. 4A-B). Analysing highly aggregated Aβ in plaques, the number and coverage is significantly different in male mice (Fig. 4G). The complete absence of the ABCA7 transporter didn't change Aβ plaque coverage and number in comparision to APPPS1-hA7^flx control mice, whereas conditional knockout on Cx3CR1 microglia cells showed a significant reduction in both parameters demonstrating the functional role that microglial ABCA7 may exert for Aβ load and the formation of Aβ plaques (Fig. 4F-G). It is noteworthy that the decrease on insoluble Aβ₄₂ quantify on conditional Cx3cr1-hA7^ko mice by immunoassay also correlates to the number and to the density of the plaques observed by IHC (Fig. 4E). At 200 days of age, when β-amyloidosis is widespread throughout the brain (the total amount of insoluble Aβ₄₂ detected is approximately 4-fold greater than that detected at 100 days), immunoassay measurements showed that complete ABCA7^ko protects against soluble Aβ₄₂ burden but not conditional knockout in microglia in females (Fig. 4C). Concerning insoluble Aβ, which at this stage is more toxic and physically associated with the microglia that surround the plaques, APPPS1-Cx3cr1-hA7^ko was again able to control the levels of these peptides in both male and female mice (Fig. 4C-D).

The status of microglia and astrocytes in the brain was examined by IHC at 100 days of age, at which point the changes in β-amyloidosis status were pronounced and physiologically translated to the onset and development of the disease in AD patients. Regarding microglial IBA1⁺ cells, we observed that IBA1 coverage and cell size in APPPS1-hA7^ko female mice were the same as those in APPPS1-hA7^flx control mice, whereas ABCA7 transporter knockout in CX3CR1⁺ cells reduced microglial cell coverage and their possible activation state (Fig. 5B). In males, the complete absence of the ABCA7 transporter leads to a large increase in IBA1⁺ cell coverage and soma size, which could be related to the development of a neuroinflammatory phenotype. This effect was restored to control levels when the knockout was restricted to microglial CX3CR1⁺ cells in the conditional model in both males and females (Fig. 5A-C). We also checked whether there was a link between microgliosis in male mice and β-amyloidosis. A clear positive correlation existed between soluble Aβ₄₂ (not an aggregated peptide), IBA1 coverage and the size of microglia (Fig. 5D), revealing how the microglial phenotype could change in response to increased amyloid production. Higher levels of Aβ₄₂ in APPPS1-hA7^ko male mice are associated with the presence of brain microgliosis, whereas lower levels of Aβ₄₂ are associated with reduced microglial coverage and soma size in the APPPS1-hA7^flx and APPPS1-Cx3cr1-hA7^ko groups. This positive correlation was also detected in males (but not in females) because higher levels of soluble Aβ₄₂ peptide are associated with pronounced astrogliosis, which was more evident in the APPPS1-hA7^ko group (Fig. 6D).

GFAP changes are restricted to males; in females, we did not observe any significant differences. This could be related to the general increase in all parameters evaluated in the female APPPS1-hA7^flx control group compared to the male group (Fig. 6B). Both APPPS1-hA7^ko and APPPS1-Cx3cr1-hA7^ko males showed a greater percentage of astrocytic brain coverage and a greater total number of GFAP⁺ astrocytes than did the APPPS1-hA7^flx controls (Fig. 6C). These parameters are the main hallmarks of astrogliosis in vivo and could be caused by morphological changes such as hypertrophy of the main cellular processes or loss of the filamentous shape of the cell, among others. The IHC results suggest that neuroinflammation observed in the APPPS1 model could be driven, at least in part, by the ABCA7 transporter, which is expressed not only in astrocytes but also in microglia. ABCA7-related communication could be essential for the modulation of this pathogenic mechanism.

One of the most studied mechanisms of inflammatory signaling in glial cells is the NLR-based inflammasome, whose main cellular response is the release of proinflammatory cytokines and pyroptosis [51]. We used qRT‒PCR to determine the mRNA expression levels of the different players involved in the sequential steps required for the triggering of this complex, including activation, priming, and oligomerization genes, in brain tissue samples from 100-day-old mice. As we have previously observed in the IHC results, most mRNA changes occur in male mice. In females, we did not detect differences between the experimental groups (Fig. 7A, C, E). In contrast, there was a clear increase in the expression of the P2rx7, Cd36 and Nfkb1 genes in APPPS1-hA7^ko male mice compared to APPPS1-hA7^flx controls. However, conditional knockout of the ABCA7 transporter in CX3CR1⁺ microglia restored the mRNA expression of all these markers to the APPPS1-hA7^flx control group levels (Fig. 7B-D) which indicates that selective loss of the ABCA7 transporter in microglia does not affect NLRP3 activation and priming, while complete loss of ABCA7 does.

According to the literature, Aβ can trigger the activation of CD36 and P2X7 receptors [19]. We verified that in male mice, we found a significant positive correlation between soluble Aβ₄₂ levels quantified by immunoassay and the mRNA expression levels of these genes (Figure S4A-B). This is an indirect way of showing how the β-amyloidosis deposition process might influence the neuroinflammatory process in the APPPS1 model. Regarding the genes directly involved in the assembly of the inflammasome complex, we detected increased mRNA expression of Nlrp3 and Pycard, but not of Casp1, in APPPS1-hA7^ko male mice, which was reversed to control levels in the absence of the transporter on microglia (Fig. 7F). One of the signals that initiates the assembly of NLRP3 into the ASC protein is the activation of the NFκB pathway. There was a positive correlation between the mRNA expression levels of Nfkb1 and Nlrp3 in both males and females, with the highest levels in APPPS1-Cx3cr1-h7^ko mice (Figure S4C-D).

Even if the expression of the Casp1 gene was not altered, we wanted to determine whether the levels of the functional protein responsible for converting pro-IL1β to the active form of this cytokine were dysregulated. In females, we did not observe any change in the qPCR data (Fig. 7G), but in males, we noticed the same trend as in the qPCR results: an increase in caspase-1 protein in the APPPS1-hA7^ko group, which returned to control levels in the APPPS1-Cx3cr1-hA7^ko mice. (Fig. 7H) We also detected NLRP3 inflammasome activation at 200 days of age, but in this case, we did not observe any differences (Figure S3). These results are consistent with our previous findings and suggest that ABCA7 functionality may be essential for regulating the inflammatory state in the brain via modulation of the NLRP3 inflammasome complex at early stages of the disease.

To gain a complete overview of the inflammasome complex and its activation, we also measured the protein levels of selected cytokines in brain tissue samples using immunoassays. At 100 days of age, we did not observe any marked changes in females, except for an increase in IL12p70 (Fig. 8A). In males, the results are consistent with the qRT‒PCR data. We detected a decrease in the levels of IL1β and TNFα in the APPPS1-Cx3cr1-hA7^ko mice, which was consistent with the decrease in caspase-1 protein levels and the expression of NLRP3-related genes (Fig. 8B). No changes in the levels of anti-inflammatory cytokines were detected at this age (Fig. 8C-D). At 200 days, when the system collapsed due to the burden and deposition of Aβ, changes in cytokine levels became more evident. In females, there was an increase in the anti-inflammatory cytokines IL-4, IL-5 and IL-10 in the APPPS1-hA7^ko group, which was restored to control levels by the absence of the transporter only in microglia (Fig. 8G>). This increase in anti-inflammatory cytokines may be related to the greater ability of female mice to counteract neuroinflammation than male mice, in whom ABCA7 deficiency appears to have a detrimental effect. Concerning proinflammatory cytokines, we quantified a decrease in the levels of IL1β and TNFα in both male and female APPPS1-Cx3cr1-hA7^ko mice (Fig. 8E-F).

We were also interested in how microglia lacking the functional ABCA7 transporter were able to induce changes in astrocytes exposed to the expression of the APP transgene under the control of the Thy1 promoter. For this purpose, we performed an in vitro conditioned culture assay on postnatal pups (P0-P5) in which we treated APPPS1 astrocytes with conditioned media from APPPS1-hA7^ko microglia to assess the communication between these two cell types. First, we verified the purity of both primary cell cultures by qPCR for Gfap (enriched in astrocytes) and Cx3cr1 (enriched in microglia) markers, which were within the expected levels (Figure S5B).

We established two experimental groups of APPPS1-hA7^ko mice to mimic in vitro the same inflammatory conditions observed in vivo: (a) untreated microglia or (b) microglia stimulated with 100 ng/mL LPS for 24 h. LPS stimulation led to increased expression of NLRP3 inflammasome assembly genes (Figure S5C) as well as a significant increase in proinflammatory cytokine release (IL1β, TNFα, IL-6 and IL12p70) in APPPS1-hA7^ko microglia (Figure S5D). Conditioned media from both groups of microglia was used to stimulate APPPS1 astrocytes for another 24 h. qRT‒PCR measurements showed that APPPS1-treated astrocytes incubated in the presence of LPS(+)-conditioned media (named APPPS1 (+ LPS)) expressed higher levels of Nlrp3, Casp1 and Nfkb1 than control DMEM-treated astrocytes or astrocytes incubated with conditioned media from nontreated microglia (named APPPS1 (− LPS)). In this case, there was a trend that did not reach statistical significance. (Fig. 9A). These results were consistent with the cytokine measurements. APPPS1 astrocytes (+ LPS) release increased levels of the proinflammatory cytokines IL1β, TNFα, IL12p70 and IL-6 as well as some anti-inflammatory cytokines, such as IL-4, IL-5 and IL10, into the media. The incubation of astrocytes with untreated microglial cell media significantly increased the levels of some of these cytokines (Fig. 9B), indicating that LPS is required in these early developmental conditions to induce the inflammatory conditions observed in adult mice. The in vitro data mimic the results obtained in vivo and support the existence of an interplay between microglia and astrocytes in the absence of the ABCA7 transporter, which may be involved in modulating the neuroinflammatory response.