Exogenous lipase administration alters gut microbiota composition and ameliorates Alzheimer’s disease-like pathology in APP/PS1 mice

Alzheimer’s disease (AD) represents the most common form of dementia in the elderly with no available disease modifying treatments. Altered gut microbial composition has been widely acknowledged as a common feature of AD, which potentially contributes to progression or onset of AD. To assess the hypothesis that Candida rugosa lipase (CRL), which has been shown to enhance gut microbiome and metabolite composition, can rebalance the gut microbiome composition and reduce AD pathology, the treatment effects in APPswe/PS1de9 (APP/PS1) mice were investigated. The analysis revealed an increased abundance of Acetatifactor and Clostridiales vadin BB60 genera in the gut; increased lipid hydrolysis in the gut lumen, normalization of peripheral unsaturated fatty acids, and reduction of neuroinflammation and memory deficits post treatment. Finally, we demonstrated that the evoked benefits on memory could be transferred via fecal matter transplant (FMT) into antibiotic-induced microbiome-depleted (AIMD) wildtype mice, ameliorating their memory deficits. The findings herein contributed to improve our understanding of the role of the gut microbiome in AD’s complex networks and suggested that targeted modification of the gut could contribute to amelioration of AD neuropathology.

www.nature.com/scientificreports/ treated APP/PS1 was reshaped towards Wt taxa composition (Fig. 1D). In addition, two months of CRL treatment increased Proteobacteria levels in both genotypes but appeared to be more pronounced in treated Wt mice in fecal samples (Supplementary Fig. 5D; Fig. 1D). Finally, Linear discriminant analysis Effect Size (LEfSe) was applied for analysis of microbes in fecal samples of APP/PS1 mice after 2 months of treatment, which identified Clostridiales vadin BB60 group and Acetatifactor genera to be significantly increased in treated animals (Fig. 1E). Next, gas-chromatography mass-spectrometry (GCMS) was used to examine whether the fecal microbial shifts were associated with alteration in the cecal gut metabolome after two months of treatment. The identified increase of the Acetatifactor genus in treated APP/PS1 mice (which has been discovered in an obese mouse 27 ) was hypothesized to have occurred in response to lipase-dependent fatty acid release and other potential alterations of the gut metabolome (Fig. 1F). Therefore, treatment-dependent changes of the gut metabolite composition were investigated in APP/PS1 mice. Three metabolites were significantly decreased (5-hydroxytryptophan: Figure 1. Gut-related changes through CRL treatment in APP/PS1 and Wt mice. A Study design: Wt (n = 13-15, 8 months of age) and APP/PS1 (n = 12, 8 months of age) received 5000 FIP/kg CRL for two consecutive months or regular water before sacrifice. B α-diversity analysis in fecal matter analyzed by Shannon and Simpson index. C. β-diversity analysis by Bray-Curtis dissimilarity and distances of fecal matter. D Taxonomy analysis of phyla abundance in fecal matter by alignment with Greengenes database. E Biomarker identification via LEfSe on genus level identified significant genera driving microbial changes in APP/PS1 treated versus untreated animals. F Metabolomics analysis of cecal matter of APP/PS1 groups showed increased abundance of metabolites that are associated with lipid digestion. G Pathway analysis of identified metabolites to determine key pathways activated through lipase administration revealing pathways associated with lipid degradation. H. Gut integrity analysis by transition measurement of oral administered FITC-dextran dye into plasma did not show a significant difference between treatment groups. Significance for α-diversity was assessed by Kruskal-Wallis H test and pairwise comparisons, for β-diversity with PERMANOVA. Cecal metabolomic data and LEfSe results were analyzed using 2-way ANOVA and FDR correction. Gut integrity was evaluated using 2-way ANOVA. Significance: *p < 0.05, **p < 0.01. www.nature.com/scientificreports/ fold change (FC) = 0.280, p false discovery rate (FDR) = 0.071; allocholic acid FC = 0.498, p FDR = 0.065; and phenylalanine FC = 0.522, p FDR = 0.071), while five metabolites were significantly increased (malic acid, FC = 1.988, p FDR = 0.018; ethanolamine, FC = 1.732, p FDR = 0.017; linoleic acid, FC = 1.690, p FDR = 0.052; oleic acid, FC = 1.650, p FDR = 0.029; and glycerol, FC = 3.393, p FDR = 0.017; Fig. 1F). The associated metabolome pathway analysis further identified pathways associated with unsaturated fatty acids and lipid metabolism supporting the hypothesis that CRL administration elevates fatty acid release in the gut lumen of APP/PS1 mice (Fig. 1G). Finally, gut barrier integrity was assessed (Fig. 1H). No differences between treated and untreated animals or APP/PS1 and Wt animals were observed. Therefore, the gut analysis supported the hypothesis that CRL treatment induces changes in the gut lumen that may be beneficial to the host. CRL treatment normalizes peripheral levels of unsaturated fatty acids. Subsequently, it was investigated whether the identified gut alterations were reflected in the host's peripheral circulation, which has been shown to be in direct exchange with the gut, through the gut-brain axes 28 . First, immune cell populations (T-helper: CD3+CD4+ and cytotoxic T cells: CD3+CD8+) were examined by flow cytometry (Fig. 2A). While no difference was measured between untreated Wt and APP/PS1 mice, a trend for decreased CD3+CD4+ and CD3+CD8+ populations was observed in treated APP/PS1 mice when compared to untreated animals. But, since the ratio was unaffected by treatment, compared to control groups the analysis suggested no treatment-dependent effects ( Fig. 2A). Measurement of plasma cytokine levels assessed treatment-dependent activity changes of immune cells ( Fig. 2B; Supplementary Fig. 2B). Although multiple cytokines showed treatment-dependent trends, the results had negligible biological relevance due to the overall low concentrations. Hence, the immunological axis was excluded from further investigations. Next, the metabolic axis was analyzed to determine the treatment-dependent changes in plasma by untargeted metabolomics. The pathway analysis revealed, for both genotypes, significant treatment-dependent alterations in pathways associated with unsaturated fatty acid metabolism (Fig. 2C). To identify further the direction of the identified alterations, the data was sub-grouped in total saturated (SFA) and unsaturated fatty acids (UFA, Fig. 2D). While no significant genotype-dependent and treatment-dependent effects in SFAs and UFAs-to-SFAs ratio were observed, UFA levels were significantly increased in untreated compared to treated APP/PS1 mice (p = 0.001, Fig. 2D). Further subdivision of UFAs in ω-3 and ω-6 fatty acids displayed the same trend in both sub-groups (ω-3: p = 0.0008; ω-6: p = 0.021), but the effect was more pronounced in ω-3 fatty acids (Supplementary Fig. 2A). These, results warranted the subsequent analysis of specific unsaturated fatty acid levels and their downstream products, which showed significant reduction of the associated metabolites in treated APP/PS1 mice in comparison to untreated APP/PS1 levels including linoleic (FC APP/PS1 = 0.651; p = 0.0002), arachidonic (FC APP/PS1 = 0.618; p = 0.0001) and docosahexaenoic acid (FC APP/PS1 = 0.569; p < 0.0001; Fig. 2E). Finally, it was investigated whether alterations in lipid transport favored the peripheral differences of UFA in APP/PS1 mice measured by high-density lipoproteins (HDL) and very lowdensity and low-density lipoprotein (VLDL/LDL) cholesterol plasma levels (Fig. 2F). Peripheral total cholesterol and cholesterol ester levels were unchanged, while cholesterol and cholesterol esters in HDL and VLDL/LDL fractions showed treatment-dependent divergent differences (Fig. 2F). Only VLDL/LDL cholesterol and cholesterol ester levels were significantly increased post treatment (cholesterol: FC = 1.459, p = 0.031; cholesterol ester: FC = 1.529, p = 0.031; Fig. 2F).
In summary, we showed that CRL treatment altered the microbial β-diversity in fecal and cecal specimen and elevated two Clostridiales vadin BB60 genera as well as Acetatifactor abundance. Furthermore, we showed that cecal UFA levels of treated APP/PS1 mice were elevated, while peripheral UFA levels were normalized, and VLDL/LDL cholesterol and cholesterol esters were increased.
CRL treatment affects glial cell marker levels and memory. Next, CRL's treatment effect on memory and learning (Barnes maze), neuroinflammation (Glial Fibrillary Acidic Protein antibody (GFAP)/ ionized calcium-binding adapter molecule 1 (Iba1)/Congo Red staining), and gene expression (RNAseq) was examined. The Barnes maze acquisition trials mostly showed significant differences between untreated Wt and APP/PS1 animals and trends for treatment-dependent improvements ( Fig. 3A; Supplementary Fig. 3B). The probe trial showed significant genotype-dependent differences for frequency to find the target hole and a trend for latency to find the target hole (latency: p = 0.114; frequency: p = 0.043; Fig. 3A; Supplementary Fig. 3A). Significant improvement of latency to find the target hole and a trend for improvement in frequency of target hole investigations was determined in treated APP/PS1 mice when compared to untreated littermates (latency: p = 0.025; frequency: p = 0.478; Fig. 3A; Supplementary Fig. 3A). Furthermore, CRL treated Wt mice showed a trend for improvement when compared to untreated animals (latency: p = 0.158; frequency: p = 0.325; Fig. 3A; Supplementary Fig. 3A/B). In addition, untreated APP/PS1 mice exhibited decreased velocity results in both acquisition and probe days, which was normalized upon treatment (velocity: p = 0.045, Fig. 3A/B) and was attributed to diminished exploratory behavior.
Next, it was examined whether improvement of memory was associated with decreased neuroinflammation in the brain. The degree of microgliosis and astrocytosis were analyzed in cortex and hippocampus by Iba1/Congo Red and GFAP staining ( Fig. 3B; Supplementary Fig. 4, respectively). As expected, significant genotype-dependent differences were observed between Wt and APP/PS1 animals (Iba1: p < 0.0001; GFAP: p < 0.0001; Fig. 3B, Supplementary Fig. 4, respectively) and no differences between treated and untreated Wt mice. A significant reduction of GFAP as well as Iba1 staining was observed in the cerebral cortex in treated APP/PS1 mice (Iba1: p = 0.0094; GFAP: p < 0.0001; Fig. 3B; Supplementary Fig. 4, respectively). In addition, no differences in the ratio of Iba1 stained microglia counts surrounding amyloid plaques to the area of the respective Congo Red stained amyloid plaques was found between treated and untreated APP/PS1 mice in the cerebral cortex (Fig. 3C). In the hippocampal area Iba1 and GFAP levels were significant between Wt and APP/PS1 animals (Iba1: p < 0.0001; www.nature.com/scientificreports/ GFAP: p < 0.0001; Fig. 3B; Supplementary Fig. 4, respectively). In untreated and treated Wt animals no differences were observed, while untreated and treated APP/PS1 animals showed significant improvement which was less pronounced when compared to the cortical analyses (GFAP: p = 0.0003; Fig. 3B; Supplementary Fig. 4, www.nature.com/scientificreports/ respectively). However, when investigating the ratio of microglia counts-to-proximal amyloid plaque area in the hippocampus of APP/PS1 animals, a significant reduction in the ratio was examined in treated animals (p = 0.015; Fig. 3C). Due to the stronger impact of treatment on microglia and astrocyte marker in cortical tissue, the treatmentdependent effect on cortical cytokine and chemokine levels and transcriptome was measured. Although IL-1β and TNFα showed significant differences between APP/PS1 treatment groups, the results had no biological relevance due to the overall low levels of cytokines and chemokines ( Supplementary Fig. 2C). Next, the total RNA expression of whole cortical tissue was analyzed by a focused hypothesis-driven analysis (gene enrichment) comparing treated and untreated APP/PS1 mice although the EdgeR analysis did not show differential expressed transcripts after FDR ( Fig. 3D/E). Three pathways were significantly altered: endocytosis, lysosome, and glycerophospholipid metabolism (p = 0.032, p = 0.048 and p = 0.051, respectively; Fig. 3D). As endocytosis and lysosomal activity are particularly associated with glial cells, the dataset was reanalyzed for specific brain cell markers. Microglial and astrocytic cell marker GFAP, TMEM119, Aif1, Olfml3 and Ccl3 were shown to be differentially expressed in treated APP/PS1 mice (p FDR = 0.046, p FDR = 0.036, p FDR = 0.005, p FDR = 0.010 and p FDR = 0.002, respectively; Fig. 3E).

Multi-Omics integration for identification of crucial parameters. A total of three Omics datasets
including metagenomic (fecal 16S rRNA), metabolomic (plasma untargeted metabolomics) and transcriptomic data (cortical RNAseq) were analyzed to investigate global treatment effects in APP/PS1 mice. The focus of our integrated omics analysis was guided by the significant variables determined in the gut metagenomic, plasma metabolomic, and cortical transcriptomic datasets described above. However, to assess the variables' dependencies and strength in relation to treatment, repeated double-cross validation with random forest (rdCV-RF) was applied on the identified significant features to prevent statistical overfitting as described by Liu et al. (Fig. 4A) 29,30 . The rdCV-RF analysis revealed that 100% (determined by division of misclassified parameters and total number of parameters) of amplicon sequence variants (ASVs), 75% of metabolites and 20% for transcripts were correctly classified highlighting the weakness of the selected transcriptomic parameters (Fig. 4A). This was further emphasized by the under the curve (AUC) values, which were used to evaluate model performance post rdCV-RF analysis. ASVs and metabolite datasets resulted in an AUC of 0.8 to 1, suggesting good model performance, while the transcriptomics dataset only reached 0.62 suggesting again weak performance in accordance with the EdgeR analysis ( Fig. 4A When determining the correlation between the three datasets, the strongest correlation was observed between ASVs and metabolites, while metabolites and transcripts showed the weakest interaction (Fig. 4B). The correlation between variables was examined by Data Integration Analysis for Biomarker discovery using a latent component method for Omics (DIABLO, correlation cut-off of 0.95), which revealed two separated correlated pathways: (1) Clostridiales vadin BB60 group uncultured bacterium 2 showed significant negative correlations with linoleic, arachidonic and docosahexaenoic acid and genes associated with microglial gene expression; while (2) Acetatifactor was negatively correlated with GFAP and oleic acid (Fig. 4C). Nonetheless, the interpretation of correlations to RNA transcripts must be further validated due to the weakness of the transcriptomic dataset.
In summary, CRL treatment altered the gut metabolome and microbial composition, normalized peripheral UFA levels and increased fatty acid transport, while reducing AD-like pathology and improving aberrant behavior.

Validation of CRL's induced gut changes and their contribution to memory improvement.
To categorially analyze whether the observed changes of AD-like pathology in APP/PS1 mice from Study 1 were primarily driven by the gut-induced changes through CRL treatment; fecal samples of all groups of Study 1 were collected from treatment week 5 to 8. The fecal matter of each group was pooled and stored as live stocks. These live stocks were then transplanted into AIMD Wt mice, which exhibited 50% reduced amounts of observed ASVs in fecal samples prior to the fecal transplants ( Fig. 5A/B). After transplantation, the α-diversity index (observed ASVs) at day 42 showed that animals receiving FMTs from Wt animals (untreated) and APP/PS1 animals (treated and untreated) exhibited comparable ASVs when compared to the Sham group (Fig. 5B). In contrast, the ABX group, which received sham FMTs, maintained a reduction of 30 to 40% observed ASV lev- Figure 3. Brain pathological impact of CRL treatment. A Probe and acquisition trial parameters of Barnes maze to assess memory and learning. The probe trial showed a significant improvement for latency to find target hole as well as frequency of target hole investigations for treated mice and a trend for treated WT mice. The acquisition data suggested trends for improved learning in treated APP/PS1 mice. B Immunohistochemical analysis of treatment-dependent effects on microglia activation. Iba1/Congo Red stained cortex showed reduction in astrocytosis and microgliosis (× 20 magnification: four animals per group, four slides per animal, three images per tissue area). C Assessment of amyloid plaque size to associated microglia counts to determine changes in inflammatory response and plaque burden (× 40 magnification: four animals per group, four slides per animal, six images per tissue area). D Enrichment analysis of significant uncorrected cortical transcriptomic data of treated and untreated APP/PS1 mice suggesting microglia and astrocyte activity being altered as well as glycerophospholipid metabolism. E Transcriptomics analysis of brain cell markers of whole cortical tissue revealing treatment-dependent changes in microglia-specific genes in APP/PS1 mice. Significance of probe trial was determined by 1-or 2-way ANOVA and post-hoc Tukey analysis, dependent on the respective parameter, while acquisition data was analyzed using 3-way ANOVA and post-hoc Tukey analysis. 2-way ANOVA and Tukey correction were further applied for immunohistochemical results and 2-way ANOVA and FDR correction for transcriptomics analysis. Significance: *p < 0.05, **p < 0.01, ***p < 0.001. www.nature.com/scientificreports/ els after 42 days. This suggested that the gut microbiome was successfully disturbed after treatment (Fig. 5B). Fecal β-diversity of the Sham group at 0, 15 and 42 days co-located, while the ABX and PEG group (day 14 and 15, respectively) showed a significant dimensional separation of the Bray-Curtis distance from the Sham group in accordance with the previous results (Fig. 5C). After 42 days, β-diversity in mice that received antibiotics and sham FMT (ABX) or antibiotics and FMT from treated Wt mice (Wt + CRL) was separated from the Sham group, while antibiotic-depleted mice receiving fecal matter from APP/PS1, APP/PS1 + CRL and Wt mice shifted back towards the Sham group. This suggested partial recovery of these groupsConsequently, this group was excluded and labelled by light grey data points/bars in the microbial analysis ( Fig. 5B/C/D). from AIMD. The most prominent taxonomical difference between ABX, WT + CRL and APP/PS1 groups compared to the other three groups was the increased abundance of Verrucomicrobia and Firmicutes, and decreased Bacteroidetes (Fig. 5D). The overall unexpected results of the Wt + CRL group led to additional analyses that showed abnormal phyla levels, enlarged ceca and increased levels of δ-Proteobacteria ( Donor and recipient fecal material did not exhibit major differences between phyla except for the excluded Wt + CRL recipient group ( Supplementary Fig. 5C). Altogether, this suggested a present infection in this group that may have occurred during FMT administration. www.nature.com/scientificreports/ After microbial depletion and recolonization, the respective microbiomes were analyzed in relation to cognitive performance since treatment with antibiotics and the associated dysbiosis have been reported to inhibit learning and long-term memory 31 . All groups were normalized to Sham and compared to the ABX group. None of the groups showed overt differences during acquisition trials in the Barnes maze ( Fig. 5E; Supplementary Fig. 5E). During the probe trial AIMD Wt mice receiving FMT's from untreated APP/PS1 mice showed impaired memory (APP/PS1; frequency: 68.33%, p = 0.919; latency: 140.96%, p = 0.661; Fig. 5E; Supplementary Fig. 5D) like AIMD Wt mice not receiving FMTs (ABX; frequency: 55%; latency: 136.77%; Fig. 5E; Supplementary Fig. 5D). Mice receiving FMTs from CRL treated APP/PS1 mice exhibited a trend for improvement (APP/PS1 + CRL; frequency: 100%, p = 0.116; latency: 60.24%, p = 0.257; Fig. 5E; Supplementary Fig. 5D) like mice receiving fecal matter from untreated animals (Wt; frequency: 108.47%, p = 0.041; latency: 58.83%, p = 0.067). Despite the infection, mice receiving FMTs from Wt + CRL mice performed as well as untreated Wt mice in the Barnes maze (Wt + CRL; frequency: 108.47%, p = 0.041; latency: 35.71%, p = 0.022; Fig. 5E; Supplementary Fig. 5D). Hence, we established that alteration of the microbial and metabolic composition by CRL treatment contributed to mitigation of ADlike pathology and normalized aberrant behavior in treated APP/PS1 mice.

Discussion
Previous studies have shown that dietary interventions can lead to an altered gut microbial composition and shifts in gut metabolites with potential benefits to the host's health 32 . However, it is still unclear how these alterations can evoke benefits at sites distant to the gut including brain-related functions. Understanding the underlying mechanisms is an urgent need in microbiome research to determine the potential for microbiome-targeting treatment approaches in central nervous system diseases. To distinguish between cause-and-effect observations in this area of research, which will enhance clinical translation, the complex system's networks and pathways must be understood in depth. Therefore, Study 1 examined the evoked treatment-dependent alterations in the gut microbial composition and metabolome, and propagated effects to distal body sites in APP/PS1 compared to Wt mice. In addition, we assessed in Study 2 the causal relationship between CRL-induced gut environmental changes and improvement in memory via FMT.
In Study 1, CRL treatment altered the microbial and metabolic composition in the gut without affecting gut barrier integrity as we previously showed 15 . The selected CRL dose was determined by translation of an approved human Creon dose, which is administered in patients with cystic fibrosis or pancreatic insufficiency 33 . Three microbial genera distinguishing CRL treated from untreated APP/PS1 mice were observed: two Clostridiales vadin BB60 uncultured and Acetatifactor genera. The cecal metabolomics analysis showed elevated fatty acid release upon treatment. Acetatifactor was discovered in an obese mouse and was seen to occur in response to high-fat diets further supporting CRL's activity in the gut 27,34 . This genus has been suggested to counteract harmful obesity effects by stimulating the GLP-1 secretion, which can improve liver function and enhance glucose and insulin sensitivity as well as host energy metabolism 35,36 . The increased abundance of Acetatifactor could have therefore counteracted features that are known to frequently occur in APP/PS1 mice as well as AD patients prior to onset of plaque pathogenesis and cognitive decline [37][38][39] . The genus Clostridiales vadin BB60 has been less explored but is associated with homovanillic acid and 5-hydroxyindole acetic acid levels in the brain stem 40 . The increased abundance of all three genera might, therefore, suggest increased communication of the gut-brain axes and together with the elevation of lipid hydrolysis may have contributed to the observed amelioration of AD-like pathology 14,41,42 .
Next, peripheral and brain alterations were investigated to assess whether the treatment-dependent changes were affecting AD-like pathology. While the immunological axis was shown to be treatment-independent, the metabolic axis revealed treatment-dependent changes in certain unsaturated fatty acid levels, which were elevated on average by 65% in untreated APP/PS1 mice but mitigated in treated APP/PS1 mice. However, elevation of ω-3 fatty acids levels in particular had been shown to evoke anti-inflammatory effects and might have been an endogenous mechanism to counteract progressing APP/PS1 pathology 43 . Hence, one might hypothesize that reduction of those levels might harm the host and facilitate progression of the disease post CRL treatment 44 . But since these mice exhibited ameliorated brain pathology, we hypothesized that increased peripheral UFA levels in untreated APP/PS1 mice might have been induced by increased lipolysis and/or reduced liver metabolism rather than an endogenous protective mechanism. It has been shown that increased levels of peripheral FFAs-as observed in untreated APP/PS1 animals-by increased lipolysis reduces BBB integrity, increases insulin resistance, disturbs energy metabolism and can exacerbate neuroinflammation 37,45,46 . Normalization of UFA levels would have therefore contributed to reduction of parameters contributing to AD-like pathology. However, while we reported increased UFA levels in APP/PS1 mice, levels in AD patients have been shown to be decreased 47 . Nevertheless, a study of patients with probable AD reported reduced VLDL cholesterol levels in these patients supporting our finding and emphasizing the potential treatment mechanism of CRL at the selected time-point 48 .
In conclusion, a longitudinal study of peripheral and brain fatty acid levels will be necessary to understand whether the elevation in untreated APP/PS1 animals was based on a disease-relevant mechanism at a specific time-point in the disease progression or whether this represented a specific occurrence in the selected mouse model. This will enhance the understanding of whether the treatment-dependent reduction of ω-3 and ω-6 fatty acids contributed to the observed amelioration of brain pathology.
The observed treatment-dependent reduction of microgliosis and astrocytosis supports this hypothesis. Reactive astroglia and microglia populations have been reported to increase synaptic loss in cortical and hippocampal tissue and cognitive decline [49][50][51][52] . Interestingly, it has been reported that defective lipid metabolism of astrocytes is reversed by a high-fat diet counteracting neurological deficits in sterol regulatory element-binding protein (SREBP)-cleavage-activating protein (SCAP) depleted mice supporting our selected treatment strategy 53  www.nature.com/scientificreports/ environment improving neuroinflammation and cognitive performance as determined in the Barnes maze. Finally, to identify potential drivers of the observed effects and the associated pathways, we conducted a cortical transcriptomic analysis. No significant differences post FDR correction were found. When applying gene enrichment analysis on the significant transcripts prior to FDR correction, endocytosis, lysosomal degradation and glycerophospholipid metabolism were identified as significant pathways. Endocytosis and lysosomal degradation are protective mechanisms of microglial and astrocyte neuroimmune activity, while the glycerophospholipid metabolism was associated with the peripheral results of altered lipid metabolism. The subsequent analysis of cell specific markers of the transcriptomic dataset revealed gene expression differences of microglia and astrocyte transcripts. The reduction in astrocyte marker GFAP suggested a potential reduction in reactive astroglia, while microglia marker TMEM119 (homeostatic marker), Aif1 (activated microglia), Ccl3 (inflammatory marker), and Olfmfl3 indicated a reduction of the reactive microglia population in the cortex 49,[54][55][56][57] . In future studies, transcriptomic analysis of isolated cell populations such as microglia or astrocytes of both cortex and hippocampus should be applied to prevent transcript dilution by, for instance, neurons, which will enable stronger conclusions. This shortfall was further emphasized in the multi-omics integration analysis, which exposed the transcriptomics dataset's weak performance, while highlighting the vital role of the determined ASVs and metabolites in the treatment-dependent effects.
In summary, CRL administration in the gut might have reduced AD-like pathology by increased release of a variety of fatty acids, which altered the gut microbiome and metabolome and rebalanced energy metabolism. This might have led to normalization of peripheral fatty acid levels by enhancing liver homeostasis and thereby lipid transport. Consequently, this might have rebalanced the fatty acid composition in the brain and reduced oxidative stress in neurons, which resulted in reprogramming of astrocytes and microglia enhancing neuronal activity and brain performance.
As an alternative hypothesis, we proposed that UFA level reduction represented a consequence of treatmentdependent amelioration of brain pathology. Hence, microbial, and metabolic alterations in the gut might have propagated through a different axis in the brain-such as the neuronal axis. The treatment-dependent increase in cecal short-chain fatty acids might have activated enteric vagal signaling. Enteric vagal afferents project into the nucleus tractus solitarius and dorsal motor nucleus, which can signal via dopaminergic or serotonergic neurons in different areas of the brain such as the hippocampus and cortex [58][59][60] . As consequence, astrocytes and microglia might have been reprogrammed into the anti-inflammatory subtype by neuronal signaling or respective metabolites travelling to the brain via the vagal route 61,62 . The reduction of neuroinflammation might have in turn recovered lipid transport and normalized peripheral levels of ω-3 and ω-6 fatty acids in APP/PS1 mice.
To finally evaluate whether the beneficial treatment-dependent changes in the gut caused the reduction in AD-like pathology, we investigated whether these beneficial changes were transferable. Wt mice receiving FMT's from treated APP/PS1 mice exhibited improved memory when compared to ABX mice receiving FMT from untreated APP/PS1 mice or none. This supported the study's hypothesis that CRL treatment could recover a healthy gut environment, which reduced AD-like pathology in APP/PS1 mice.
These findings demonstrate the potential of digestive enzymes as clinically relevant agents to potentially treat diseases such as AD. CRL treatment in diseases of ageing are particularly interesting since enzyme activity has been shown to decline with increasing age, potentially contributing to dysfunction of glucose and lipid metabolism 63,64 . More experiments need to be conducted to further exploit the pathway leading to the determined beneficial effects and to validate the findings in other AD mouse models and human studies to enhance translatability. Nevertheless, this series of studies may have presented a novel strategy to target and treat neuropathology of AD but also pathologies of other neurodegenerative diseases due to their common association with ageing and inflammation.

Materials and methods
Animal models. Animals were kept in the Association for Assessment and Accreditation of Laboratory Animal Care International (AAALAC) accredited vivarium of the Roskamp Institute. Experiments with mice were reviewed and approved by the Institutional Animal Care and Use Committee (IACUC) of the Roskamp Institute before implementation and conducted in compliance with the National Institutes of Health Guidelines for the Care and Use of Laboratory Animals. The design of both animal studies and their connection have been visualized in a study overview figure (Supplementary Fig. 6). Wt and APP/PS1 for Study 1: CRL treatment study were obtained from in-house breeding. Wt mice for the Study 2: FMT study were also obtained from in-house breeding. Mice were maintained on a 12 h/12 h light/dark cycle and received food and water ad-libitum. All studies have been reported in accordance with the ARRIVE guidelines. Figure 5. Transferability of results via FMTs in AIMD WT mice. A Study design. B α-diversity measured as observed ASVs indicated depletion post antibiotics and bowel cleanse and reshaping post FMT. C β-diversity measured by Bray-Curtis distance. While Sham groups co-located and ABX and PEG groups co-located, animals receiving FMTs were distinct but located separately. The ABX group recovered partially. For analysis, the group receiving Wt + CRL FMT (abnormal α-and β-diversity) was fainted throughout microbial analysis, although no impact on behavioral data was observed. D Taxonomic analysis of phyla levels in fecal samples post FMT. Besides in animals receiving Wt + CRL, no major differences were found between groups. E Barnes maze experiment for examination of differences in memory and learning. Results were normalized to sham group to reduce complexity of analysis. Analysis revealed that AIMD WT mice receiving none, or FMT from untreated APP/PS1 mice, performed similar or worse than all other groups. Significance was determined as in prior experiments of microbiome and Barnes maze analysis. The probe trial results were compared to ABX group only. Significance: *p < 0.05, **p < 0.01, ***p < 0.001. from an internal breeding protocol, and randomly subdivided into CRL treated and untreated groups. Mice received 5000 Fédération Internationale Pharmaceutique (FIP)/kg body weight CRL (300,000 FIP/g, Enzymedica, FL, USA) in drinking water for 2 months, or regular water. The used dose was extrapolated from Creon, a human digestive enzyme treatment for pancreatic insufficiency 33 . CRL has been shown to be stable in water for up to 7 days and hence, water, treated or untreated, was changed twice a week 65 . Beginning four weeks before Study 1 started nesting and bedding materials were exchanged between all groups and throughout the study within groups twice a week, to normalize endogenous gut microbial variance and to prevent cage effects. Four groups were assessed in Study 1 (APP/PS1 n = 12, APP/PS1 + CRL n = 12, Wt n = 13, Wt + CRL n = 15). Fecal samples were collected at 0, 4, and 8 weeks of treatment and in addition, three times per week during treatment week 5 and 8 for the subsequent Study 2. In the last two weeks of treatment, mice were trained and assessed for spatial learning and memory in the Barnes maze. On the last day of treatment, gut integrity was assessed. Animals were humanely euthanized after anesthesia with 2% isoflurane (Patterson veterinary, Greeley, CO) in oxygen via cardiac puncture-induced exsanguination and subsequent perfusion with phosphate buffered saline (PBS) in accordance with the approved IACUC protocol. Right hemispheres of the brain were fixed in 4% paraformaldehyde (Sigma, St. Louis, MO) for 24 h for immunohistochemical analysis. The left-brain hemisphere, gastrointestinal tract, and plasma were collected and immediately flash frozen in liquid nitrogen and stored at − 80 °C.  68,69 . Subsequently, all mice fasted for 4 h. Finally, mice received the first FMT with 5.00 × 10 6 cells in 50 μL per FMT or 10% glycerol in 0.9% NaCl (Sham, ABX) via oral gavage 66,70 . The three subsequent gavages were administered once weekly thereafter. One week after the last FMT, mice were trained and evaluated for Barnes Maze to determine the effect of FMTs on learning and memory. After behavioral testing, animals were humanely euthanized after anesthesia with 2% isoflurane (Patterson veterinary, Greeley, CO) in oxygen via cardiac puncture-induced exsanguination in accordance with the approved IACUC protocol. Subsequently tissues were collected for 16S sequencing analysis.

Scientific Reports
Gut integrity. Prior to euthanasia, mice (APP/PS1 n = 6, APP/PS1 + CRL n = 5, Wt n = 3 and Wt + CRL n = 3 from Study 1) were starved (food and water) for 5 h. Next, 4 kDa fluorescein isothiocyanate (FITC)-dextran (0.6 mg/g body weight, Sigma, St. Louis, MO) was orally administered via oral gavage. 60 min later mice were humanely euthanized. Following euthanasia, plasma was collected, flash frozen in liquid nitrogen, and then stored at − 80 °C. For analysis plasma samples were diluted 1:5 in PBS pH 7.4. 100 μL of water and 50 μL of sample-or standard (0-40 μg/mL 4 kDa FITC-dextran)-were added to a black 96-well μ-clear bottom plate (Greiner Bio-One, Monroe, NC). Fluorescence was immediately measured at 485/528 nm and FITC dextran concentration was examined in each sample. Data was analyzed with 2-way ANOVA and Tukey multiple comparisons correction and plotted in GraphPad Prism 8 (GraphPad, San Diego, CA).
Microbiome analysis. DNA (n = 6/group for Study 1 and for Study 2) was extracted from feces (20 − 100 mg/ pellet) and cecum (30 mg) samples using the Fast DNA stool mini kit (Qiagen, Germantown, MD) according to the manufacturer's protocol. Volumes were adjusted to account for lower sample weight. DNA extracts were analyzed by V3-V4 region 16S rRNA gene amplicon sequencing using the Illumina MiSeq sequencing platform (Illumina, San Diego, CA) in collaboration with Prof. Klatt, University of Minnesota, following the Earth Microbiome Project protocols (http:// press. igsb. anl. gov/ earth micro biome/ proto cols-and-stand ards/ 16s/) and our recent publication 15 . In addition, the MicrobiomeAnalyst interface was used to identify differential abundant microbial genera with the LEfSe application 71 . Data was plotted in GraphPad Prism 8 (GraphPad, San Diego, CA), β-diversity plots were acquired in Qiime2 using the Emperor suite 72 . Flow cytometry. Flow cytometry analysis of T lymphocytes was conducted according to the protocol of Joshi et al. 75 . In brief, 200 μL of whole blood (n = 6/group for Study 1) were diluted 1:10 in red blood cell lysis buffer (RBC, Fisher Scientific, Waltham, MA), centrifuged and pellet resuspended in 500 μL 95% PBS and 5% fetal bovine serum (FBS). 100 μL suspensions were labeled with 0.25 μg of α-CD4-FITC antibody, 0.25 μg of α-CD8a-PE and/or 0.5 μg α-CD3-Cy7 antibody (11- www.nature.com/scientificreports/ surrounding Congo Red stained plaques in both hippocampus and cortex within the same area were taken and plaque size area analyzed with ImageJ. In addition, microglia that directly surround the associated plaque area were counted and the ratio between plaque area and count assessed. Data was plotted and analyzed via 2-way ANOVA and Tukey correction in GraphPad Prism 8 (GraphPad, San Diego, CA).

Barnes maze.
To analyze memory and learning the Barnes maze was used with 18 equally spaced holes around the outer perimeter of the maze (n = 11-15 mice/group for Study 1; n = 11-12 mice/group for Study 2). The target hole had a box positioned directly beneath it that allowed the mice to exit the maze. Distinct visual cues were positioned on each of the four walls. Mice were trained for 4 days (3 min/trial/day) to use the cues to locate the target hole and escape the maze, which was achieved when the mouse entered the target box positioned under the target hole. On day 5, the target box was removed to evaluate for learning and spatial memory in a 90-s probe trial. Each trial was tracked and recorded using EthoVision XT 14 software. Data plotting and statistical analysis was performed in GraphPad Prism 8 (GraphPad, San Diego, CA) with 3-way ANOVA testing for acquisition and one-or two-way ANOVA dependent on the respective parameter on the probe day with post-hoc multiple comparisons.
Transcriptomics. Cortex (n = 4/group for Study 1) was separated from one frozen hemisphere and immediately transferred into 1 mL of TRIzol (Fisher Scientific, Waltham, MA) under RNase free conditions. Samples were disrupted by ultrasonication. One hundred μL of 1-bromo-3-cholorpropane reagent (Sigma, St. Louis, MO) were subsequently added and samples mixed and centrifuged at 12,000 g for 15 min. Finally, RNA was precipitated by addition of 500 μL ice-cold isopropanol and washed with 75% ethanol in DEPC water (Fisher Scientific, Waltham, MA). The air-dried pellet was resuspended in DEPC water and RNA concentration measured at Cytation 3 (BioTek, Winooski, VT). At least 500 ng RNA was subsequently sent to GENEWIZ LLC (South Plainfield, NJ), for sample processing and total RNA sequencing (20-30 million reads). The sequencing data was analyzed using the following applications from the Galaxy platform 76  Multi-omics analysis. Multi-omics analysis was performed according to the Liu et al. protocol 29 . Briefly, a rdCV-RF analysis method was applied with an inner tuning and outer testing loop of 100 repetitions and subsequent cross-validation with 1000 permutations 30 . This was followed by DIABLO in the R package mixOmics 81 , which was employed for multi-Omics integration to determine variable correlation throughout the three datasets.
Ethics approval. Experiments with mice were reviewed and approved by the IACUC of the Roskamp Institute before implementation and conducted in compliance with the National Institutes of Health Guidelines for the Care and Use of Laboratory Animals.

Data availability
Authors did not use custom code or software for the described analysis of this manuscript. All data analyzed during this study are included in this published article. The sequencing data (paired end reads in FASTQ) of microbiome sequencing, the respective manifest and metadata files as well as unfiltered taxonomic classification of each study and specimen are available in the Mendeley data repository 82 . www.nature.com/scientificreports/