Gut lining enzyme Dicer1 influences gut bacteria and Alzheimer's signs in genetically modified mice

App^NL−G−F knock-in (App^ki/ki) mice were kindly provided by T.C. Saido, RIKEN Brain Science Institute, Japan [30], Dicer1-floxed (Dicer1^fl/fl) mice by A. Domanskyi, University of Helsinki, Finland [31] and Villin-Cre mice over-expressing CreERT2 under control of the Villin gene promoter by E. Batlle, The Barcelona Institute of Science and Technology (BIST), Spain [32]. App^ki/ki mice express humanized APP in the endogenous App locus harboring Swedish, Beyreuther/Iberian and Arctic mutations, avoiding the transgenic mouse artifacts that Aβ is generated along with overproduction of other APP fragments. The mice show typical Aβ pathology, neuroinflammation and memory impairment in an age-dependent manner [30]. App^ki/ki mice with and without expression of Dicer1 in epithelial cells of the intestine (App^ki/kiDicer1^del and App^ki/kiDicer1^wt) were created by crossing App^ki/ki, Dicer1^fl/fl and Villin-Cre mice to obtain the genotypes App^ki/ki/Dicer1^fl/fl/Cre^tg and App^ki/ki/Dicer1^fl/fl/Cre^wt, followed by feeding littermate AD mice with 600 mg/kg tamoxifen (Carbolution Chemicals GmbH, St. Ingbert, Germany) supplements in diet from 6 month of age for 1 month. AD phenotypes were analyzed at 9 months before the end of the experiments. During experiments, the experimenters did not know which genotype the mice had.

All App^ki/kiDicer1^del and App^ki/kiDicer1^wt mice on a C57BL/6 genetic background born between June 2018 and December 2019 were included in this study. As we only compared AD phenotypes between Dicer1-deficient and wild-type brothers or sisters, all mouse litters without suitable comparison pairs were excluded. Both male and female mice were used. The exact number of mice used in each experiment was listed in Supplementary Table 1. The experimental littermate mice with different expression of CreERT2 were kept in the same cage and received tamoxifen-supplemented diet.

To exclude a sex-related bias in AD pathology, we performed a preliminary experiment to compare Aβ42 levels in brain homogenates from 9-month-old male and female App^ki/ki mice. We found no significant effect of sex on cerebral Aβ level (see Supplementary Fig. 1). We also made heterozygous breeding to get App^ki/ki mice and murine App wild-type (App^wt/wt) littermate controls, and then compared the gene transcription in the intestinal wall between these two groups of mice.

Animal experiments were conducted in accordance with national rules and ARRIVE guidelines, and authorized by Landesamt für Verbraucherschutz, Saarland, Germany (registration numbers: 36/2018).

An open field test and a Y-maze test were performed. In the open field test, each mouse was allowed to run freely for 10 min in a bare and open chamber. Initially, the mice often stayed close to the walls, and later explored the center area. The total distance, duration in the center area, frequency of visits to the center area and velocity were monitored with a camera placed above the open field. The open field test was conducted once a day for three consecutive days. The Y-maze was made of gray plastic and consisted of three compartments radiating out from the center platform. In this test, each mouse was placed in the center of the maze facing toward one of the arms and was then allowed to explore freely for 5 min. The number of arm entries and the number of triads were recorded to calculate the percentage of alternation. An entry occurred when all four limbs were within the arm. The Y-maze test was repeated three times per day in three consecutive days.

Animals were euthanized by inhalation of overdose isoflurane. After intracardial perfusion with ice-cold phosphate-buffered saline (PBS), the brain was removed and divided along the anterior-posterior axis. The left hemisphere was immediately fixed in 4% paraformaldehyde (Sigma-Aldrich Chemie GmbH, Taufkirchen, Germany) in PBS and embedded in paraffin for immunohistochemistry. The cortex and hippocampus were carefully separated from the right hemisphere. A roughly 0.5-mm thick sagittal section of tissue was cut from the medial side and homogenized in TRIzol (Thermo Fisher Scientific, Darmstadt, Germany) for RNA isolation. The remainder was snap-frozen in liquid nitrogen and stored at -80 °C until biochemical analysis. The cecum together with a 0.5-cm-long segment of the neighboring colon was also collected, snap-frozen and stored at -80 °C for isolation of intestinal bacteria. After the collection of gut bacteria, the cecum was washed in cold PBS and homogenized in TRIzol to extract total RNA of intestinal wall.

To isolate brain blood vessels, the cortex and hippocampus from 9-month-old App^ki/kiDicer1^del and App^ki/kiDicer1^wt mice were dissected and brain vessel fragments were isolated as we did previously [33]. Briefly, brain tissues were homogenized in HEPES-contained Hanks' balanced salt solution (HBSS) and centrifuged at 4,400 g in HEPES-HBSS buffer supplemented with dextran from Leuconostoc spp. (molecular weight ∼70,000; Sigma-Aldrich) to delete myelin. The vessel pellet was re-suspended in HEPES-HBSS buffer supplemented with 1% bovine serum albumin (Sigma-Aldrich) and filtered with 20 μm-mesh. The blood vessel fragments were collected on the top of filter and stored at -80 °C for biochemical analysis.

Bacterial DNA was extracted from intestinal content (100 mg) in the frozen cecum and neighboring colon with QIAamp Fast DNA Stool Mini Kit (Qiagen, Hilden, Germany). The amount of bacteria was evaluated by quantifying 16S rDNA with SYBR Green-based real-time PCR using the universal bacterial r16S gene primers (16S-V2-101F: 5-AGYGGCGIACGGGTGAGTAA-3, and 16S-V2-361R: 5-CYIACTGCTGCCTCCCGTAG-3) as we did in a previous study [13]. The amount Gapdh DNA derived from the intestinal cells was also measured and used as an internal control with the primers: sense, 5'-ACAACTTTGGCATTGTGGAA-3' and antisense, 5'-GATGCAGGGATGATGTTCTG-3'. The V3 - V4 region of the 16S rRNA-encoding gene was then amplified with the barcode fusion primers (338 F and 806R). After purification, PCR products were used for constructing libraries and sequenced on the Illumina MiSeq platform at Majorbio Co. Ltd. (Shanghai, China). The raw data was processed on Qiime2 (https://qiime2.org/↗), reducing sequencing and PCR errors, and denoising to get the operational taxonomic unit (OTU) consensus sequences, which were mapped to the 16S Mothur-Silva SEED r119 database (http://www.mothur.org/↗). Alpha diversity including Sobs, Shannon, Ace, Chao and Simpson indexes were used for the analysis of bacterial richness and diversity in a single mouse. Principal coordinate analysis (PCoA) and analysis of similarity (ANOSIM) were used for β-diversity analysis to compare bacterial compositions on genus level between App^ki/kiDicer1^del and App^ki/kiDicer1^wt mice. The difference of bacterial compositions on genus level between these two groups were also compared with Wilcoxon rink sum test. Additionally, we employed the Phylogenetic Investigation of Communities by Reconstruction of Unobserved States (PICRUSt2) software [34] to predict microbial functions by annotating the gene catalog based on the Kyoto Encyclopedia of Genes and Genomes (KEGG) modules [35]. All the analysis was performed using cloud-based tools with default analysis parameters (https://cloud.majorbio.com/page/tools.html↗).

Total DNA was isolated from mouse brains using TRIzol. Real-time PCR was used as described above to quantify the bacterial 16S rDNA in the brain tissue with murine Gapdh DNA as an internal control. Subsequently, the V3-V4 region of bacterial 16S rDNA was sequenced and analyzed for the microbiome at Novogene GmbH, Munich, Germany. Amplicon sequence variants (ASVs) instead of OUT were generated from the denoised raw data, and annotated with the database Silva 138.1. The α- and β-diversity analysis were performed. The difference in bacterial abundance in the brains of App^ki/kiDicer1^del and App^ki/kiDicer1^wt mice at the genus level was determined by MetagenomeSeq analysis [36].

The spleen of 9-month-old AD mice were prepared for single-cell suspensions using Spleen Dissociation Kit (mouse) (Miltenyi Biotec GmbH, Bergisch Gladbach, Germany). After blocking with 50 µg/ml CD16/CD32 antibody (clone 2.4G2; BioXCell, Lebanon, USA), CD4-positive spleen cells from the spleen with Dynabeads magnetic beads-conjugated CD4 antibody (clone L3T4; Thermo Fisher Scientific). Lysis buffer was immediately added to selected cells and total RNA was isolated using RNeasy Plus Mini Kit (Qiagen).

Serial 50-µm-thick sagittal sections were cut from the paraffin-embedded hemisphere. Four neighboring sections with 300 μm of interval were deparaffinized, labeled with rabbit anti-ionized calcium-binding adapter molecule (Iba)-1 antibody (Wako Chemicals, Neuss, Germany) and VectaStain Elite ABC-HRP kit (Cat.-No.: PK-6100, Vector Laboratories, Burlingame, USA), and visualized with diaminobenzidine (Sigma-Aldrich). Iba-1-positive microglia/brain macrophages were counted with Optical Fractionator in the hippocampus on a Zeiss AxioImager.Z2 microscope (Carl Zeiss Microscopy GmbH, Göttingen, Germany) equipped with a Stereo Investigator system (MBF Bioscience, Williston, USA), as we did previously [37].

To evaluate the cerebral Aβ level, after deparaffinization, 4 serial brain sections from each animal were stained with mouse monoclonal Aβ antibody (clone MOAB-2; BioXCell, Lebanon, USA), biotin-conjugated goat anti-mouse IgG and Cy3-conjuagted streptavidin (Jackson ImmunoResearch Europe Ltd. Cambridge, UK). After mounting, the whole section including hippocampus and cortex was imaged with Microlucida (MBF Bioscience) and merged. The positive staining and brain region analyzed were measured for the area with Image J tool "Analyse Particles" (https://imagej.nih.gov/ij/↗). The threshold for all compared samples was manually set and kept constantly. The percentage of Aβ coverage in the brain was calculated.

To investigate the microglial clearance of Aβ, brain tissue was incubated with a rabbit monoclonal antibody against CD68 (clone E3O7V; Cell Signaling Technology) after Aβ staining and visualized with Alexa488-conjugated goat anti-rabbit IgG (Thermo Fisher Scientific). The stained hippocampus and cortex were imaged under the 40× objective first in the Cy3 channel and then in the Alexa488 channel for more than 3 regions in the hippocampus and 7 regions in the cortex with ≥ 1 Aβ plaque per region. The ratio of CD68- and Aβ-immunoreactive areas measured with Image J was calculated.

Frozen brain tissues and blood vessel isolates were homogenized in RIPA buffer (50mM Tris [pH 8.0], 150mM NaCl, 0.1% SDS, 0.5% sodiumdeoxy-cholate, 1% NP-40, and 5mM EDTA) supplemented with protease inhibitor cocktail (Roche Applied Science, Mannheim, Germany) on ice. The proteins were separated by 10% or 12% Tris-glycine SDS/PAGE. Before loading on PAGE gel, vessel preparations were sonicated. Western blots were performed using rabbit monoclonal antibodies against ABCB1 (clone E1Y7S), PDGFRβ (clone 28E1), CD13/APN (clone D6V1W), Munc18-1 (clone D4O6V), SNAP25 (clone D110), PSD95 (clone D74D3), and synaptophysin (clone D8F6H), and rabbit polyclonal antibody against LRP1 (Cat.-No.: 64099) (all antibodies above were bought from Cell Signaling Technology). The detected proteins were visualized via the Plus-ECL method (PerkinElmer, Waltham, USA). To quantify proteins of interest, rabbit monoclonal antibody against β-actin (clone 13E5), Vinculin (clone E1E9V), or GAPDH (clone 14C10) (these three antibodies were from Cell Signaling Technology) was used as a protein loading control. Densitometric analysis of band densities was performed with Image-Pro Plus software version 6.0.0.260 (Media Cybernetics, Rockville, USA). For each sample, the protein level was calculated as a ratio of target protein/loading control per sample.

The frozen brain hemispheres were homogenized and extracted serially in Tris-buffered saline (TBS), TBS plus 1% Triton X-100 (TBS-T), guanidine buffer (5 M guanidine HCl/50 mM Tris, pH 8.0) as described in our previous study [37]. Aβ concentrations in three separate fractions of brain homogenates were determined by Human Amyloid β42 and β40 Brain ELISA kits (Cat.-No.: EZBRAIN42 and EZBRAIN40, respectively; both from Merck Chemicals GmbH, Darmstadt, Germany). Results were normalized on the basis of the sample's protein concentration.

Membrane components were prepared from brains of 9-month-old App^ki/kiDicer1^del and App^ki/kiDicer1^wt littermate mice and β- and γ-secretase activities were measured by incubating the crude membrane fraction with secretase-specific FRET substrates according to our established methods [37].

Total RNA was isolated from mouse brains using TRIzol. RNAseq libraries were prepared using 100 ng of total RNA per sample with the NEBnext Ultra II Directional RNA library kit for Illumina (New England Biolabs [NEB]; Ipswich, USA) and the NEBNext Poly(A) mRNA Magnetic Isolation Module (NEB) according to the manufacturer's recommendations. Libraries were quantified with the NEBnext library quantification kit for Illumina (NEB) and then sequenced on an Aviti sequencing system (Element Biosciences; San Diego, USA) for 30–40 million 2 × 75 nt reads per sample. Raw data was processed with the nf-core RNAseq pipeline version 3.17.0 (https://nf-co.re/rnaseq/3.17.0/↗) and nextflow v23.04.1 applying the default settings and using mm10 reference assembly. Differential analysis was performed in R using the DESeq2 package. Gene ontology (GO) enrichment analysis was performed using the GO Consortium-provided website-based free service (https://geneontology.org↗) with a 0.05 adjusted p-value cut-off and a minimum absolute log2fold change of 1, and also using the GOstats package (https://www.bioconductor.org/packages/release/bioc/html/GOstats.html↗) with a 0.05 p-value cut-off and a minimum absolute log2fold change of 1.

Total RNA was isolated from mouse brains or the wall of cecum with TRIzol or from selected CD4-positive spleen cells with RNeasy Plus Mini Kit (Qiagen) and reverse-transcribed. Gene transcripts were quantified with established protocols [37, 38] and TaqMan gene expression assays of mouse Tnf-α, Il-1β, Chemokine (C–C motif) ligand 2 (Ccl-2), Il-10, Chitinase-like 3 (Chi3l3), Allograft inflammatory factor 1 (Aif-1, encoding Iba-1 protein), Il-17a, RAR-related orphan receptor C (Rorc), Interferon-γ (Ifn-γ), Il-4, Neprilysin and Insulin-degrading enzyme (Ide), Gapdh and 18S ribosomal RNA (18S) (Thermo Fisher Scientific). The transcription of mt-Nd4, mt-Nd4l, mt-Atp6, Atp5e, Atp6v1c2, Ndufa2, Ndufa5, Romo1, Lrrc17, and Agrp in brain tissues, and Zo1, Occludin1, Claudin1, Claudin2, Claudin5 and Dicer1 genes in intestinal walls was determined using the SYBR green binding technique with the primers showed in Table 1, part of which were copied from the published studies [39, 40].

Data were presented as mean ± SEM and displayed using a scatter plot with a bar overlay in the figure, with the scatter plot representing individual data points. Continuous variables between two groups of cases were compared with two independent-samples Students t-test or Mann-Whitney-U-test, depending on whether the variables were normally distributed. To exclude the possibility that the reduction of cerebral Aβ was co-caused by the deletion of Dicer1 in the gut and sex, we performed a linear regression analysis with Aβ as the dependent variable and the deletion of epithelial Dicer1 and sex as independent variables. For multiple comparisons, we used one-way or two-way ANOVA followed by Bonferroni, Tukey, or Dunnett T3 post hoc test (dependent on the result of Levene's test to determine the equality of variances). All statistical analyses were performed with GraphPad Prism 8 version 8.0.2. for Windows (GraphPad Software, San Diego, USA) or SPSS software for Windows (Version 26.0, IBM, Armonk, USA). Other statistical methods in microbiome analysis offered by external companies have been described above. Statistical significance was set at p < 0.05.

Intestinal epithelial cells are the major source of miRNA in the intestinal lumen [27]. Six-month-old App^ki/kiDicer1^del and App^ki/kiDicer1^wt mice were administered with oral tamoxifen for one month to delete Dicer1 in the intestinal epithelial cells, and analyzed for the phenotype in the gut at 9 months of age. Due to technical difficulties, we were unable to directly measure miRNA amount in the lumen of gut. However, we observed that the ratio of Dicer1/18S gene transcripts in the cecum wall was significantly lower in App^ki/kiDicer1^del mice than in App^ki/kiDicer1^wt mice (Fig. 1, A; 4.522 ± 0.851 × 10^{− 6} vs. 9.167 ± 1.776 × 10^{− 6}; t test, p < 0.05). We also observed a significant reduction in transcription of Claudin 1 gene but no other tight junction protein (TJP)-encoding genes in the cecum (Fig. 1, B - F; t test, p < 0.05). Our results were consistent with previous studies on the gut miRNA [27, 32], suggesting successful deletion of Dicer1 in the intestinal epithelial cells of our AD mice.

Reduced TJP expression may impair the intestinal barrier, therefore, we examined the structure and inflammatory activation of the gut. Intestinal morphology after ZO1 staining did not differ between App^ki/kiDicer1^del and App^ki/kiDicer1^wt mice (Fig. 1, G). The transcription level of inflammatory genes, e.g., Tnf-α, Il-1β, Inos, and Aif-1 that encodes Iba-1 protein, but not that of Ccl-2 and Il-10 genes, was significantly lower in App^ki/kiDicer1^del mice than in App^ki/kiDicer1^wt mice (Fig. 1, H - M; t test, p < 0.05). Moreover, deletion of Dicer1 gene in the gut did not alter gene transcription of Il-17a, Rorc, Ifn-γ, Il-4 and Il-10 in CD4-positive splenocytes of App^ki/kiDicer1^del mice compared to App^ki/kiDicer1^wt mice (Fig. 1, N - R; t test, p > 0.05). All these results suggest that deletion of Dicer1 in epithelial cells does not result in damage to intestinal barrier integrity, as leakiness of the gut barrier should trigger inflammatory activation and immune response in the gut. Our Dicer1-deleted App^ki/ki mice allowed us to investigate the effects of miRNA on bacteria in the gut and subsequent AD pathologies in the brain.

In additional experiments, we detected the transcription of Dicer1, and various inflammatory genes and TJP-encoding genes in the cecum wall of App^ki/ki and App^wt/wt littermates. These two groups of mice did not differ in the expression of Dicer1 (Supplementary Fig. 2, A; Dicer1/18S, 1.850 ± 0.108 × 10^{− 5} vs. 1.742 ± 0.247 × 10^{− 5}; t test, p = 0.699). Surprisingly, the transcriptional level of Il-1β and Ccl-2 genes, but not Tnf-α, Inos, Il-10 and Aif1 genes was significantly lower in App^ki/ki mice than in App^wt/wt mice (Supplementary Fig. 2, B - G; t test, p < 0.05), suggesting that the knock-in expression of pathogenic App alters the inflammatory status in the intestine, although the underlying mechanism remains to be investigated. The transcription of Zo1, Occludin1, Claudin 1, 2 and 5 genes was not different between App^ki/ki and App^wt/wt mice (Supplementary Fig. 2, H - L; t test, p > 0.05).

We knocked out Dicer1 gene specifically in epithelial cells in App^ki/kiDicer1^del mice, which in principle should reduce miRNA in the gut compared with App^ki/kiDicer1^wt mice. Indeed, 2 months after tamoxifen induction, the number of bacteria as shown by the level of bacterial 16S rDNA was significantly reduced in cecum contents of 9-month-old App^ki/kiDicer1^del mice compared with App^ki/kiDicer1^wt littermates (Fig. 2, A; t test, p < 0.05). In α-diversity analysis of bacterial species, Sobs, Ace, Chao, and Shannon indexes were significantly smaller in App^ki/kiDicer1^del mice than in App^ki/kiDicer1^wt littermates (Fig. 2, B - E; t test, p < 0.05). The Simpson index was not altered (Fig. 2, F). The Sobs, Chao, Ace were used to describe the bacterial species richness, i.e., the number of OUTs. The Shannon index and Simpson indexes were used to describe community diversity, including species richness and species evenness. Therefore, deletion of epithelial Dicer1 reduces gut bacterial richness and diversity in AD mice.

In the PCoA, we clearly observed the difference in the architecture of gut bacteria between App^ki/kiDicer1^del and App^ki/kiDicer1^wt mice (Fig. 2, G; ANOSIM, R = 0.8850, p = 0.002), which was in line with the observation in mice with constitutive knockout of Dicer1 gene in intestinal epithelial cells [27]. At the genus level, the abundance of bacteria of Helicobacter, Parasutterella, Bacteroides and Lachnoclostrium increased, while that of Prevotellaceas, Muribaculaceae, and Lachnospiraceae decreased in App^ki/kiDicer1^del mice compared with App^ki/kiDicer1^wt controls (Fig. 2, H; Wilcoxon rank-sum test, p < 0.01).

To investigate the impact of deletion of intestinal Dicer1 on the brain, we used the PICRUSt2 software [34] and annotated the metagenomic reads on the KEGG database [35] to predict the function of changed microbiota. Deletion of Dicer1 had the potential to influence various pathophysiological functions of the host. Notably, it might promote neurodegenerative disease and adversely affect the nervous system compared with Dicer1 wild-type AD mice (Fig. 2, I; Wilcoxon rank-sum test, p < 0.05), although the prediction needs to be validated.

In a further experiment, we pooled all mice and compared the bacterial composition between male and female mice by PCoA analysis. We observed that sex had no effect on the β-diversity of gut bacteria in our App^ki/ki mice, which was consistent with our preliminary experiment on the effect of sex on cerebral Aβ levels (see Supplementary Fig. 1).

Our recent study has shown that bacterial components, i.e., bacterial DNA, can circulate to the brain [13]. We detected a lower DNA amount of the bacterial 16S rRNA gene in the brain tissue of 9-month-old App^ki/kiDicer1^del mice compared with App^ki/kiDicer1^wt littermates (Fig. 3, A; t test, p < 0.05), which is consistent with bacterial reduction in the gut (Fig. 2, A). Alpha-diversity analysis showed that deletion of Dicer1 reduced both the Chao1 index and the observed features (Fig. 3, B and C; t test, p < 0.05), but did not significantly affect Shannon and Simpson indexes (data not shown). Chao1 index and observed features were positively and strongly correlated (Fig. 3, D; Spearman correlation test, p < 0.001). Thus, deletion of Dicer1 in the intestine has the potential to increase the richness and diversity of bacteria in the brain, even if it decreases the absolute number of bacteria.

In the brains of 9-month-old App^ki/kiDicer1^del and App^ki/kiDicer1^wt littermate mice, we observed no difference in β-diversity analysis at the genus level as shown by PCoA (Fig. 3, E; ANOSIM analysis, p > 0.05). However, MetagenomeSeq analysis had shown that deletion of epithelial Dicer1 significantly increased the abundance of Agathobacter, Akkermansia, Bradyrhizobium, Escherichia-Shigella, Desulfovibrio, Diaphorobacter, Novosphingobium, Paenibacillus and Prevotella bacteria, and decreased that of Lachnospiraceae_UCG-001 bacteria (Fig. 3, F - O; non-parametric tests, p < 0.05). These bacteria have been found in the gut in association with AD pathology [41 –43], or in the blood or cerebrospinal fluid in correlation with inflammatory lesions [44, 45].

It should be noted that the detection and sequencing of bacterial DNA in the brain is still a great challenge. Most reads in our 16S rDNA sequencing were mapped to DNA sequences from bacteria, e.g., Sphingomonas and Acinetobacter (see Supplementary Fig. 3), which are widely found in the environment (water and soil). We need further experiments to exclude the possibility that the environmental bacteria contaminated our bacterial samples from brain tissues.

Accumulation of Aβ in the brain is the pathological hallmark of AD patients and leads to neurodegeneration in AD [46]. After observing that deletion of epithelial Dicer1 altered the microbiome in both the gut and brain of AD mice, we asked whether Dicer1 deletion also altered AD pathology in the brain of App^ki/ki mice. Four serial brain sections were stained with human Aβ antibodies. Deletion of epithelial Dicer1 significantly reduced Aβ coverage in both the hippocampus and cortex of App^ki/kiDicer1^del mice compared to App^ki/kiDicer1^wt littermates (Hippocampus: 10.53% ± 0.44% vs. 15.29% ± 0.55%, n = 11–12 per group, and Cortex: 9.11% ± 0.84% vs. 13.05% ± 0.36%, n = 4–5 per group; Fig. 4, A - C; t test, p < 0.01).

We also homogenized brain tissue from 9-month-old tamoxifen-treated App^ki/kiDicer1^del and App^ki/kiDicer1^wt mice serially in TBS, TBS-T and guanidine-HCl buffer and measured the concentrations of Aβ42 and Aβ40 in the brain homogenates with ELISA, as we had done previously [37]. We observed that the concentrations of both Aβ42 and Aβ40 in the TBS- and TBS-T-soluble fractions were significantly lower in App^ki/kiDicer1^del than in App^ki/kiDicer1^wt mice (Fig. 4, D, E, G and H; t test, p < 0.05). Deletion of epithelial Dicer1 tended to decrease Aβ40, but not Aβ42 in guanidine-HCl-soluble fractions of brain homogenates (Fig. 4, F and I; t test, p > 0.05). To exclude the possibility that sex plays a role in the reduction of Aβ40 and Aβ42 induced by miRNA deletion in the gut, we performed a linear regression analysis and confirmed that deletion of epithelial Dicer1 significantly and independently reduced Aβ in the brain (see Table 2; p < 0.05). Sex did not affect Aβ levels in the brain, which was consistent with the result of our preliminary experiment (see Supplementary Fig. 1).

We recently found that depletion of bacteria in the gut decreases the activity of BACE1, a rate-limiting enzyme in Aβ production, and increases the expression of LRP1 and ABCB1, which facilitates the efflux of cerebral Aβ across the blood-brain barrier (BBB) [13]. Therefore, we asked whether these Aβ-reducing mechanisms were also present in App^ki/ki mice with deletion of intestinal epithelial Dicer1. Interestingly, the activity of BACE1, but not γ-secretase was slightly but significantly lower in brain tissues of 9-month-old App^ki/kiDicer1^del mice than in App^ki/kiDicer1^wt mice after tamoxifen treatment (Fig. 5, A and B, two-way ANOVA, p < 0.05). Deletion of intestinal epithelial Dicer1 also significantly increased protein levels of LRP1 and ABCB1 in cerebral capillaries isolated from App^ki/kiDicer1^del mice compared with App^ki/kiDicer1^wt mice (Fig. 5, C - E; t test, p < 0.05). However, deletion of Dicer1 did not alter the expression of PDGFRβ or CD13 in the capillaries of App^ki/ki mice (Fig. 5, F - H; t test, p > 0.05). PDGFRβ and CD13 are two protein markers of pericytes that decrease in AD brains [47]. Deletion of intestinal epithelial Dicer1 appeared not to increase microglial clearance of Aβ in the brain, as CD68-immunoreactive cells around Aβ deposits did not differ between 9-month-old App^ki/kiDicer1^del and App^ki/kiDicer1^wt mice (Fig. 5, I and J; t test, p > 0.05). The expression of CD68 as a lysosomal protein suggests the activity of microglial phagocytosis [13]. Moreover, depletion of intestinal epithelial Dicer1 did not affect the transcription of Neprilysin and Ide genes (Fig. 5, K and L; t test, p > 0.05), which encode two important enzymes in the brain digesting Aβ extracellularly [48, 49].

Inflammatory activation is a pathological feature of AD in the brain and influences the progression of the disease [50]. In our App-knock-in mice, deletion of Dicer1 in intestinal epithelial cells did not alter the numbers of Iba-1-positive microglia in both hippocampus and cortex (Fig. 6, A - C; t test, p > 0.05). By detecting transcription of individual inflammatory genes, we found that intestinal deficiency of Dicer1 only significantly reduced Ccl-2 transcripts and increased the transcription of Il-10 gene, without altering the transcription of other tested genes (Tnf-α, Il-1β and Chi3l3) (Fig. 6, D - H; t test, p < 0.05). In the following transcriptomic analysis of brain tissues (Fig. 7), there was no robust difference in inflammatory activation between 9-month-old App^ki/kiDicer1^del mice and in App^ki/kiDicer1^wt mice, except that the GO enrichment analysis of differentially expressed genes (DEGs), identified with p-value (instead of adjusted p-value) cut-off at 0.05 and a minimum absolute log2fold change ≥1, showed that Dicer1 deletion activated the IL-10 signaling pathway in the brain (p < 0.05; see Supplementary Fig. 4).

To investigate the effects of intestinal Dicer1 on the brain systemically, we performed a transcriptome analysis of the hippocampus and cortex of 9-month-old App^ki/kiDicer1^del and App^ki/kiDicer1^wt mice. Next-generation sequencing identified 15,549 genes, of which 21 and 2 genes were significantly altered by a log2 fold change < -1 and > 1 (adjusted p < 0.05) in transcription in brain tissue from App^ki/kiDicer1^del mice compared to App^ki/kiDicer1^wt mice (Fig. 7, A). Interestingly, most of these DEGs involved mitochondrial metabolism. The only characterized gene with upregulated transcription was a mitochondrial gene mt-DN4; and the genes with down-regulated transcription were located in the nucleus and included Ndufa2 and Ndufa5, encoding NADH dehydrogenase [ubiquinone] 1 alpha subunit 2 and 5, respectively, in mitochondrial complex I [51, 52], Atp5e, encoding a subunit of ATP synthase, also known as complex V [53], Romo1, encoding reactive oxygen species modulator 1 at the inner mitochondrial membrane, which recognizes reactive oxygen species and regulates mitochondrial dynamics [54], and Agrp, encoding Agouti-related protein, which increases appetite and decreases metabolism and energy expenditure [55]. Using real-time PCR, we validated that the transcript levels of Ndufa2 and Ndufa5, but not other DEGs, e.g., Atp5e, Atp6v1c2, mt-Nd4, mt-Nd4l, mt-Atp6, Romo1, Lrrc17, and Agrp, were significantly downregulated in the brains of 9-month-old App^ki/kiDicer1^del mice compared to App^ki/kiDicer1^wt mice (Fig. 7, B - K; t test, p < 0.05). Not surprisingly, GO enrichment analysis of DEGs predicted that depletion of miRNAs in the gut has the potential to reduce ATP production by the mitochondrial respiratory chain (Fig. 7, L; adjusted t test, p < 0.05).

After finding that deletion of intestinal epithelial Dicer1 gene reduced cerebral Aβ and altered neuroinflammation in App^ki/ki mice, we further analyzed the effects of intestinal Dicer1 on neuronal function. We performed open field test to probe anxiety and Y-maze test for the cognitive function in 9-month-old tamoxifen-treated App^ki/kiDicer1^del and App^ki/kiDicer1^wt mice. App^ki/ki mice have been reported to display anxiety-related symptoms, even before the occurrence of cognitive dysfunction [56, 57]. In the open field test, App^ki/kiDicer1^del mice travelled significantly shorter distance and at a lower speed than App^ki/kiDicer1^wt mice (Fig. 8, A - C; t test, p < 0.05); while, we did not observe any problems in motor function of App^ki/kiDicer1^del mice (data not shown). We subsequently divided the arena into an outer zone and a center zone (Fig. 8, A) and observed that deletion of epithelial Dicer1 gene significantly reduced the entries to the center zone (Fig. 8, D; t test, p < 0.05) and tended to decrease the traveling duration in the center area (Fig. 1, E; t test, p = 0.093), suggesting an anxiogenic phenotype. In Y-maze test, App^ki/kiDicer1^del and App^ki/kiDicer1^wt mice did not differ in the alteration of arm entries (Fig. 8, F and G; t test, p > 0.05), indicating that deletion of intestinal Dicer1 did not improve the cognitive function.

We also homogenized brains from 9-month-old App^ki/kiDicer1^del and App^ki/kiDicer1^wt mice and determined synaptic proteins with quantitative Western blot (Fig. 8, H). Deletion of epithelial Dicer1 gene did not alter the protein levels of PSD-95, Munc18-1, synaptophysin and SNAP25 (Fig. 8, I - L; t test, p > 0.05), which was in accordance with the results of the Y-maze test.