All animal experiments were performed in accordance with the Directive 2010/63/EU of the European Parliament and of the Council and approved by the Lithuanian State Food and Veterinary Service (Project No G2-244). C57BL/6JRj (n = 8, Janvier Labs, France) B6.Cg-Tg (APPswe, PSEN1dE9)85Dbo/Mmjax, MMRRC ID: 34832 (APP/PS1) 12-week old mice (n = 22, The Jackson Laboratory, United States) were used for this study. APP/PS1 mice overexpress a chimeric mouse and human-mutated amyloid precursor protein (APP) with the Swedish mutation (APP695Swe) and a mutant human presenilin 1 gene with exon 9 deletion (PSEN1dE9). Genotyping of APP/PS1 mice was performed according to the official JAX protocol using the touchdown cycling protocol with 5′-GTGTGATCCATTCCATCAGC-3′, 5′-GGATCTCTGAGGGGTCCAGT-3′, 5′-ATGGTAGAGTAAGCGAGAACACG-3′ primer sequences (Supplementary Figure S1). All animals were maintained under controlled conditions in a 12-h light–dark cycle at a temperature of 22 ± 1°C and a humidity of 55 ± 3%. All animals were provided with regular chow and had ad libitum access to water. They remained under veterinary supervision for the entire duration of the experiment.

Before the experiment, the animals were divided into four groups: WT—C57BL/6JRj; APP/PS1—APPswe, PSEN1dE9 AD mice model; APP/PS1 + A. muciniphila—AD mice model supplemented with A. muciniphila; APP/PS1 + A. muciniphila + GOS AD mice model supplemented with A. muciniphila and GOS.

All animals received same standard feed (with 3.87 kcal/g energy value, Altromin, Germany) and autoclaved water. APP/PS1 + A. muciniphila + GOS received water supplemented with 3% GOS (King-Prebiotics^®, Yunfu City, Biotechnology Corporation Limited, China). The animals in the A. muciniphila and A. muciniphila + GOS groups were gavaged with the bacteria intermittently (3 times a week) for 7 months (Figure 1). After thawing on ice, A. muciniphila was centrifuged, diluted in sterile PBS and each animal received 100 μL of the solution containing 1 × 10⁹ CFU of the bacteria.

Bacterial cultivation for animal experiments: A. muciniphila CIP107961 was grown in GAM medium (Nissui Pharmaceutical Co.) and incubated at 37°C in an anaerobic chamber (Mac 500; Don Whitley Scientific) under a 10% (v/v) H2, 10% (v/v) CO₂ and 80% (v/v) N₂ atmosphere. Twenty-four hours culture was used to inoculate (1% v/v) pre-reduced GAM broth medium, which were incubated for another 24 h. Afterward, the culture was washed and concentrated with pre-reduced PBS plus 20% (v/v) glycerol to about 1 × 10¹⁰ CFU/mL and, stored at −80°C until administration (Higarza et al., 2021).

Bacterial cultivation for in vitro experiments: A. muciniphila was cultivated in stationary cultures anoxically at 37°C in liquid brain heart infusion (BHI) medium (37 g/L, Carl Roth, Germany, #X916.2) supplemented with porcine stomach mucin (4 g/L, Sigma-Aldrich, United States, #M1778). An appropriate amount of GOS was added to the medium as indicated in the Results section. To prepare agar plates, 1.5% agar was added to the medium. Anaerobic manipulation of the bacteria was carried out either in a 100% N₂ environment using the Unilab Pro glovebox (Mbraun, Germany) or by syringe using glass bottles with rubber stoppers flushed with sterile N₂. The growth curves were determined by spectrophotometric measurements with the NanoPhotometer (C40, Implen, Germany). For the growth curves, 2 mL of the bacterial suspension were grown as stationary cultures in 2 mL test tubes in a dry heating block. The presence of A. muciniphila was confirmed using end-point PCR (Supplementary Figure S2) for the species-specific sequence with 5′-CAGCACGTGAAGGTGGGGAC-3′ and 5′-CCTTGCGGTTGGCTTCAGAT-3′ pair of primers adapted from Kim et al. (2021).

Experiments with animals began at 3 months of age and concluded at 10 months. The study was conducted over a span of 7 months. Glucose tolerance test (GTT) was performed to check glucose metabolism impairment as previously described (Burokas et al., 2018). During the GTT, glucose level was measured at different time points up to 2 h. The animals were not fed for 14 h before the experiment but were provided with water. After measuring basal levels of glucose (0 timepoint), animals were injected with glucose 2 g/kg intraperitoneally (i.p.), and blood glucose levels were measured at 15, 30, 45, 60 and 120 min. Five microliters of blood was collected from the snipped tail to measure glucose levels using a blood glucose meter (CONTOUR^®NEXT GEN, United States).

Standard blinding and randomization procedures, following the PREPARE guidelines, were applied to experimental animals (Smith et al., 2018). Animal housing and cage placement were randomized at the start of the experiment. All behavior experiments were recorded and later analyzed by a researcher who was blinded to group allocation. Treatment administration, behavior testing and the behavioral recordings analysis were performed by different researchers to ensure blinding.

The battery of behavioral tests was structured according to standard practice, beginning with the least stressful and moving to increasingly long and potentially stressful experiments. Before each behavioral experiment, animals were randomized to minimize order effects. A rest period of 2 to 3 days was implemented between tests (with the exception of the GTT, after which the animals rested for 5 days) to allow animals to recover and reduce potential carryover effects. The animals were acclimatized in the experimental rooms for 1 h before behavioral experiments.

The animals were placed in the center of the open field arena (the walls of the maze were made of white opaque Plexiglas; 40 cm long × 40 cm wide × 30 cm high) at an illumination of 500 lux (measured at floor level in the center of the arena) for 10 min, as previously described by Kulesskaya and Voikar (2014). A video recording was carried out during the experiment. The time and number of entries were counted when the animal moved to the center of the maze, which was virtually marked as a “small central square” (10 cm × 10 cm). The videos were analyzed using the Biobserve Viewer (Biobserve GmbH, Germany) software.

The Novel object recognition test was conducted in a V-shaped maze, as previously reported (Busquets-Garcia et al., 2011; Burokas et al., 2014). The V-shaped maze consisted of two perpendicular arms (made of black opaque Plexiglas, dimensions of the arms: 25 cm long × 5 cm wide × 15 cm high) with dim lighting (15 lux, measured at the top of the maze). The experiment consisted of three phases: habituation, training, and testing. During habituation, the animals explored an empty maze, during the training, mice examined two identical objects and then conducted the testing phase either 3 h (for the short-term memory testing) or 24 h (for the long-term memory testing) after the training. This test is based on the spontaneous tendency of rodents to spend more time exploring a novel object than a familiar one, assigned as discrimination index (DI). The DI was calculated using the following formula: DI = (T_R − T _L)/(T_R + T_L), where T_R represented the exploration time devoted to the new object, and T_L represented the exploration time devoted to the old object. A DI of 0.4 and above was considered the standard for good memory, and a value below 0.2 was considered as a memory impairment.

The maze was a closed Y-shaped apparatus and was illuminated from the top (15 lux, measured at the end of the arms), as first reported by Kraeuter et al. (2019). The experimental animals were placed in a grey, opaque Plexiglas arena with 3 identical arms (40 cm long × 9 cm wide × 15 cm high). Ten-minute test sessions were performed, and the number of altering arms visited was counted. Mice were considered to have entered the maze when all four limbs of the mice were within an arm zone of the maze.

The marble test was performed according to the protocol of Angoa-Pérez et al. (2013). Standard cages were filled with regular unscented mouse bedding material to a depth of 7 cm and leveled evenly. Twenty unicolor glass marbles (15 mm in diameter) were carefully arranged on the bedding without pressing in five rows of four marbles. After, the animals were placed in the experimental cage, which was covered with upside-down grills. Mice were allowed to explore the marbles freely for 30 min. After each session, mice were carefully returned to their home cage, and the number of partially and completely buried marbles was evaluated. Marbles were considered buried if at least two-thirds of their surface area was covered by the bedding. After each session, marbles were washed with 70% ethanol. Animals that did not exhibit marble-burying behavior were determined to be non-responsive and were consequently excluded from subsequent analysis.

The procedure was developed based on the protocol of Steru et al. (1985). The mice were suspended by the tail using a clip placed approximately 2 cm from the tip of the tail attached to a metal bar elevated 30 cm from the surface. The duration of immobility was recorded during a single 6-min session. Mice were considered immobile only when hanging passively. All experiments were videotaped for subsequent analysis, which was performed manually by an experimenter blinded to the experimental conditions.

The Vsoc-maze test was performed in a V-shaped maze according to the protocol by Martínez-Torres et al. (2019) that consisted of two perpendicular arms ending in a metal bar-separated compartment. The experiment consisted of 3 phases that were conducted in a continuous sequence: habituation, sociability testing, and social novelty testing, with each phase lasting 5 min. During habituation, the animals explored an empty maze, while during the sociability testing, the experimental animals explored a neutral object or an unfamiliar mouse. The unfamiliar mice were of the same sex, similar age, and weight. During the social novelty testing, the neutral object was replaced by a novel, unfamiliar animal. The sociability index (SI) was calculated in the same manner as DI. SI = (T_A − T_O)/(T_A + T_O), where T_A represented the exploration time devoted to the unfamiliar animal, and T_O represented the exploration time devoted to an inanimate object.

The animals were housed individually in standard cages with a running wheel (metal wheel diameter 12 cm), provided with their respective food, and supplemented with water. Voluntary physical activity was measured by counting the number of rotations by the running wheel during a 24-h period: 12-h day cycle (cage lighting 80–100 lux, light intensity measured at the top of the cage) and 12-h night cycle (no lighting). Physical activity was recorded using a specialized recording system with PowerLab hardware, and LabChart-8 software (ADInstruments Ltd., United Kingdom) was used to measure the number of wheel rotations.

Whole gut transit time was evaluated utilizing a mixture of carmine and methylcellulose, as previously described by Golubeva et al. (2017). The animals were deprived of food overnight for 16 h prior to the experiment. A solution of 0.5% methylcellulose and 6% carmine was prepared and sterilized by autoclaving. The experimental animals were gavaged with 250 μL of the carmine solution, placed in empty cages under dim lighting, and provided with water. The cages were monitored for the appearance of red fecal pellets. The transit time was considered to be the time interval between administration of carmine and appearance of the first red pellet. A maximum period of 6 h was chosen for the experiment.

The number of microglia cells was determined by immunohistochemistry of coronal brain sections fixed in 4% paraformaldehyde (PFA) for 24 h, washed in PBS, and embedded in paraffin. The brain tissue was cut in 10 μm thick slices, placed on glass slides, and stored at +4°C. Microglia immunofluorescence was performed according to the protocol described by Zaqout et al. (2020). The slices were blocked with 10% goat serum (Thermo Fisher; United States) for 1 h. Rabbit anti-Iba1 primary polyclonal antibody was used (AB_2544912, 1:1000; Thermo Fisher; United States) overnight, followed by a secondary goat anti-rabbit antibody Alexa Fluor 488 (AB_143165, 1:5000; Thermo Fisher; United States) for 1 h. Tissues were then counterstained with DAPI (5 μg/mL), and fluorescence was visualized using a bright field scanning fluorescence microscope. The scanned unprocessed images were imported into the ImageJ software. The images were converted to 8-bit, and the region of interest was marked. The background was removed, and the rolling ball radius was set to 50 pixels. The number of cells was analyzed using find maxima (prominence >30, edge maxima excluded). The cells from the DAPI channel and microglia Iba1 channels were quantified, and the percentage of Iba1 positive microglia cells was calculated.

Cecum samples from each animal were collected split into two equal parts and kept at −80°C until processing. Before SCFA analyses, samples were weighed, diluted in PBS (1:5 w/v) and homogenized (5 min, full speed) in a stomacher (LabBlender 400). From 1 mL of the homogenate, cell free-supernatants was isolated by centrifugation (10,000 rpm, 15 min). Another part of cecum was used to extract DNA for gut microbiota analysis. The cecum was homogenized using a BeadBug^™ 6 microtube homogenizer (Merck) for three cycles of 30 s at 300 rpm, with a 30 s rest period on ice between each cycle. After homogenization, we used Quick-DNA Fecal/Soil Microbe Miniprep Kit (Zymo, D6010) to extract microbial DNA, according to the manufacturers protocol. Concentration of isolated DNA was evaluated using a spectrophotometer.

The determination of SCFA concentrations was performed by gas chromatography, as described by Higarza et al. (2021). Briefly, 250 μL of cell free-supernatants were mixed with 100 μL methanol, 50 μL internal standard solution (2-ethylbutyric 1.05 mg/mL), and 50 μL of 20% v/v formic acid. The mix was then centrifuged, and the supernatant was injected into a system composed of a 6890NGC injection module (Agilent Technologies Inc., United States) connected to a flame injection detector (FID) and a mass spectrometry detector (MS, 5973 N) (Agilent).

Library preparation and sequencing were performed by Novogene (United Kingdom) Co., Ltd. (Cambridge, United Kingdom) using the Pacbio platform, generating 10,000 clean CCS reads per sample. 16S rRNA/18S rRNA/ITS genes of distinct regions (16SV4/16SV3/16SV3-V4/16SV4-V5, 18SV4/18SV9, ITS1/ITS2, ArcV4) were amplified used specific primer (e.g., 16SV4: 515F-806R, 18SV4: 528F-706R, 18SV9: 1380F-1510R, et al.) with the barcode. All PCR reactions were carried out with 15 μL of Phusion^® High-Fidelity PCR Master Mix (New England Biolabs); 0.2 μM of forward and reverse primers, and about 10 ng template DNA. Thermal cycling consisted of initial denaturation at 98°C for 1 min, followed by 30 cycles of denaturation at 98°C for 10 s, annealing at 50°C for 30 s, and elongation at 72°C for 30 s and 72°C for 5 min.

Sequencing libraries were generated using NEB Next^® Ultra^™ II FS DNA PCR-free Library Prep Kit (New England Biolabs, United States, Catalog #: E7430L) following manufacturer’s recommendations and indexes were added. The library was checked with Qubit and real-time PCR for quantification and bioanalyzer for size distribution detection. Quantified libraries were pooled and sequenced on Illumina platforms, according to effective library concentration and data amount required.

Alpha diversity indices for the gut microbiota—Chao1, Shannon, and Simpson (Shannon, 1948; Simpson, 1949; Chao, 1987)—were computed based on the absolute amplicon sequence variant (ASV) counts, and their relationship to the study groups was assessed by linear regression. Gut motility is an important factor shaping gut microbiota (Tottey et al., 2017), thus we adjusted for whole gut intestinal transit time (in seconds) following Mingaila et al. (2023) data analysis work flow. Results with p < 0.05 were considered significant.

Beta diversity was assessed by calculating Euclidean distance on log-ratio transformed relative abundance genus counts (Aitchison et al., 2000), including only taxa with a relative abundance of ≥0.001 in at least 10% of samples. Using the “adonis2” function in the R vegan package (version 2.6-6.1) (available at https://CRAN.R-project.org/package=vegan), permutational multivariate analysis of variance (PERMANOVA) (Anderson, 2001) was conducted to assess differences in microbiota composition according to study groups. Model was adjusted for whole gut intestinal transit time and p < 0.05 as significant threshold.

Principal component analysis (PCA) was performed to illustrate the variation between study groups in a two-dimensional plot.

Differential abundance analysis was conducted using the MaAsLin2 R package (17) to test the association between specific gut microbiota taxa and study groups. General linear models were applied to log-ratio transformed and genus counts prior abundance and prevalence filtering, and accounting for carmine intestinal transit time as covariate. No further filtering was applied in the model. The analysis included Benjamini–Hochberg adjusted p-values, focusing comparisons on the primary predictor with covariate control. Findings with Benjamini–Hochberg adjusted p < 0.05 were reported.

The association between study groups and the log-ratio transformed relative abundance of the Akkermansia genus and A. muciniphila was tested using linear regression, adjusting for whole gut transit time (in seconds). Results with p < 0.05 were considered significant.

Spearman’s correlation was performed to assess the relationship between differentially abundant taxa and a dataset containing various variables. Since the dataset included continuous variables and it was assumed that these variables were missing completely at random (MCAR), multiple imputation using predictive mean matching (PMM) was applied (Rubin, 1987). PMM was considered an appropriate approach because it preserves the distribution of the variable (Morris et al., 2014; van Buuren, 2018) and imputation is performed by matching observed values with similar patterns (Schenker and Taylor, 1996). As the variables were measured on different scales (e.g., SCFA in mM, behavioral scores, and microglia percentages), standardization was conducted to prevent bias. The correlations are filtered based on the criteria absolute correlation coefficient |r| > 0.5 and adjusted p < 0.05 to identify statistically significant and practically meaningful relationships in the data.

Quantification of Aβ_(1–42) levels in the soluble and insoluble fractions, was performed according to the protocol of Bracko et al. (2020). Frozen mouse brain hippocampus and prefrontal cortex tissues were homogenized in 1 mL PBS with complete protease inhibitor cocktail (Roche), 1 mM PMSF (Sigma), and 2 mM sodium orthovanadate (Sigma) using a Dounce homogenizer on ice. The homogenized tissues were vortexed and centrifuged at 14,000 g for 30 min at 4°C. The soluble fraction was removed and stored at −80°C until further analysis. The remaining pellet was dissolved in 0.4 mL of 70% formic acid, shaken, and centrifuged at 14,000 g for 30 min at 4°C to dissolve the insoluble fraction. The supernatant was removed and neutralized with 1 M Tris-base buffer at pH 9. Protein concentrations were measured using Bradford Protein Assay (Thermo Fisher Scientific, United States). The soluble fraction was diluted 4-fold and the insoluble fraction was diluted 30-fold before the ELISA assay was performed. The solid-phase sandwich ELISA (KHB3441, Thermo Fisher Scientific, United States) was performed according to the protocol provided. The Aβ_(1–42) concentrations were calculated by comparing the optical density of the sample with the optical density of the Aβ_(1–42) standard calibration curve within the same plate. The data were standardized according to the extracted total protein concentration in the both soluble and insoluble fractions.

All grouped analysis data were presented as mean ± SEM. The data were analyzed and graphed using Prism 9 (GraphPad software), and Inkscape. Gut microbiota statistical analyses were carried out using R (version 4.4.1) and RStudio (version 2023.6.0.421). The Shapiro–Wilk test was applied to assess the normality of the data. Statistical significance was determined using Welch’s ANOVA, followed by Dunnett’s Multiple Comparisons Test, unless stated otherwise. Differences were considered significant at ^*p ≤ 0.05, ^**p ≤ 0.01, ^***p ≤ 0.001.

To confirm the ability of GOS to improve the growth of A. muciniphila, the bacteria were incubated in a liquid BHI medium with different concentrations of GOS (Figure 2). We observed a concentration-dependent increase in optical density with an exponential growth stage starting after 24 h. Significant growth differences were observed between all investigated groups at all time points, beginning at 36 h. 3 and 5% GOS provided the best conditions for the growth of A. muciniphila. Although the growth differences between 3 and 5% GOS were statistically significant, they remained minimal. Given our study’s focus on a mild intervention, we selected the 3% GOS concentration for in vivo experiments.

Obesity and elevated fasting glucose levels are important risk factors for AD. Metabolic syndrome increases the level of inflammation in the body and accelerates the accumulation of Aβ may play a role in the progression of AD (Gomez et al., 2018). To evaluate whether the administration of A. muciniphila could improve the metabolic status of APP/PS1 mice, we measured the weight and glucose levels of animals that were fasted for 14 h. No differences in weight were observed (Figure 3A). However, APP/PS1 mice had significantly higher levels of fasting glucose which were restored to WT levels by administration A. muciniphila and GOS (Figure 3B). The disruption of glucose metabolism was confirmed by the GTT in which APP/PS1 animals had higher glucose levels 1 h after glucose administration (Figure 3C) and a larger area under the curve overall (Figure 3D). The combination of A. muciniphila and GOS was effective in restoring glucose metabolism to WT levels (Figure 3D). Furthermore, APP/PS1 animals exhibited a faster intestinal transit time which was normalized with A. muciniphila and GOS supplementation (Figure 3E). These data show the potential of the treatment with A. muciniphila and GOS to improve the metabolic state in AD patients.

Lower levels of fasting glucose and slower glucose metabolism showed that the effects of A. muciniphila and GOS extend beyond the gut, manifesting systemically in the blood. However, it was still unclear whether the prebiotic-probiotic mixture could exert its effect on the CNS in an AD model. To investigate this, we conducted a series of behavioral tests designed to assess anxiety, activity levels, depressive behavior, memory, and sociability. Supplementation with A. muciniphila and GOS significantly increased both the time spent and the number of visits to the center in the open field test, indicating reduced anxiety (Figures 4A,B). Additionally, the open field test revealed a significantly lower activity in APP/PS1 animals with a trend (p = 0.150) towards partial recovery of activity in the A. muciniphila and GOS treated animals (Figure 4C). A similar trend (p = 0.076) was observed in the voluntary wheel running test (Supplementary Figure S3A), where APP/PS1 mice tended to show lower activity, which was fully restored to the WT level by the administration of A. muciniphila and GOS. Furthermore, APP/PS1 animals exhibited high immobility time in the tail-suspension test (Figure 4D), but the administration of A. muciniphila and GOS showed no effect towards the depression-like phenotype. Regarding the working memory, the Y-maze test revealed that APP/PS1 mice showed a dramatic decrease in the number of alterations (Figure 4E), which was partly reversed by the administration of A. muciniphila and GOS. The effect on memory was confirmed in the novel object recognition test with an increased discrimination index in the group supplemented with A. muciniphila and GOS (Figure 4F). The results of the short-term memory test supported these findings (Supplementary Figure S3B). This indicates the ability of A. muciniphila and GOS to prevent the memory deficits observed in the APP/PS1 AD model. APP/PS1 animals exhibited lower levels of sociability compared to WT animals and supplementation with A. muciniphila, either alone or in combination with GOS, did not result in a significant improvement in social behavior (Figure 4G). An increased frequency of repetitive behaviors is a common sign of anxiety in mice (Bailey and Crawley, 2009). A trend (p = 0.058) of an increased number of buried marbles was observed in APP/PS1 mice (Figure 4H), despite a reduction in overall activity levels (Figure 4C). Administration of A. muciniphila and GOS significantly reduced repetitive behavior to the WT levels (Figure 4H). Overall, treatment with A. muciniphila and GOS demonstrated the ability to mitigate AD-related behavioral traits in the APP/PS1 mice (summarized in Table 1).

The levels of soluble and insoluble Aβ_(1–42) monomers in extracted hippocampus and prefrontal cortex were measured after 7 months of A. muciniphila supplementation. As expected, APP/PS1 animals had significantly higher amount of Aβ_(1–42) in both soluble and insoluble fractions (Figures 5A,B). Supplementation with either A. muciniphila or combination of A. muciniphila and GOS had no effect on Aβ_(1–42) burden in the hippocampus. In the prefrontal cortex, Aβ (1–42) burden did not differ significantly between groups, with no notable reduction observed following A. muciniphila supplementation alone or in combination with GOS (Figure 5C).

A total of 2,500 ASVs were identified in 30 samples based on raw counts. After aggregation at the genus level, 67 taxa were identified in the dataset. We found that the Chao1 alpha diversity index was significantly higher in the WT group compared to the APP/PS1 group (p = 0.004) (Figure 6A, Supplementary Table S1). However, for the calculated alpha diversity indices, Shannon and Simpson, we found no significant differences between groups (Supplementary Tables S2, S3).

The PCA plot in Figure 6B shows that the top two axes, PC1 and PC2, account for 18.23 and 9.73% of the total variation, respectively. The results of the PERMANOVA indicated that the group variable significantly influenced the composition of the gut microbiota, with 25% of the variance explained by differences between the groups. This suggests that group membership was an important factor influencing the gut microbiota. The F-value of 3.075 and the p-value of 0.001 indicated that these differences were highly statistically significant, suggesting that supplementation contributed to the observed group differences (Table 2).

The results of the differential abundance analysis are summarized in Figure 6C and Supplementary Table S4. In summary, we identified a total of nine genera associated with APP/PS1 compared to WT (three positively associated), 10 genera associated with APP/PS1 + A. muciniphila compared to WT (three positively associated), and 14 genera associated with APP/PS1 + A. muciniphila + GOS compared to WT (seven positively associated). The abundance of an unknown genus from the Erysipelotrichaceae family was positively associated with APP/PS1 + A. muciniphila + GOS compared to APP/PS1 (β = 2.338, p = 0.043).

The correlation analysis between the detected bacterial genera and the obtained data variables revealed two significant correlations, both involving the genus NK4A214 group from the Oscillospiraceae family (Figure 7A). We found a moderate positive correlation (r_s = 0.626, adjusted p = 0.00002) between the abundance of the NK4A214 group and the soluble fraction of Aβ_(1–42) monomer in the hippocampus (Figure 7B). Furthermore, there was a strong negative correlation (r_s = −0.807, adjusted p = 0,040) between the same NK4A214 group genus and the number of alterations performed by mice during the Y-maze test (Figure 7C). Taken together, these results suggest a potential role of the NK4A214 group genus in the development of AD. Similar results were observed with genus UCG-010 from the Oscillospirales family (Figure 7A). Here, we observed a trend that UCG-010 abundance correlated positively (r_s = 0.564, adjusted p = 0.079) with the levels of Aβ_(1–42) monomer in the hippocampus and a trend that UCG-010 abundance negatively correlated (r_s = −0.560, adjusted p = 0.079) with the number of alterations performed in the Y-maze test (Supplementary Table S5).

We found no significant correlation between the study groups and the abundance of the Akkermansia genus or A. muciniphila (Supplementary Tables S6, S7). The lack of an increase in A. muciniphila abundance in the gut, even after long-term supplementation, suggests that higher colonization of A. muciniphila is not essential to exert a significant impact on the host microbial community.

The changes in the gut microbiota observed after the administration of A. muciniphila indicate changes in the metabolic activity of the bacterial community. The contents of the cecum were analyzed by gas chromatography and the concentrations of acetate, propionate, and butyrate were determined. We observed a trend (p = 0.052) towards increased acetate concentrations in APP/PS1 animals, which was significantly reduced by the administration of A. muciniphila and GOS (Figure 8A). In contrast, we observed no differences in concentration of propionate (Figure 8B). The changes in SCFA concentrations were particularly striking for butyrate, where APP/PS1 animals had significantly higher levels, and the administration of A. muciniphila and GOS brought these levels back to those of the WT controls (Figure 8C). These results indicate that long-term supplementation of A. muciniphila and GOS modulates the concentration of SCFA in the APP/PS1 AD model to WT levels.

Immunohistochemical analysis of the microglia marker Iba1 was performed on hippocampal slices (Figure 9). We observed a significantly increased percentage of microglia in the hippocampus of APP/PS1 animals, indicating a higher level of inflammation. Supplementation with A. muciniphila significantly prevented this increase and partially restored the phenotype to the WT levels. The addition of GOS in addition to the A. muciniphila treatment diminished the positive effect of the probiotic. These results suggest that supplementation with A. muciniphila alone is able to prevent pathological increase in microglia abundance in the hippocampus of APP/PS1 animals.