Abstract Background Gut microbiota (GMB) alteration has been reported to influence the Alzheimer’s disease (AD) pathogenesis through immune, endocrine, and metabolic pathways. This study aims to investigate metabolic output of the dysbiosis of GMB in AD pathogenesis. In this study, the fecal microbiota and metabolome from 21 AD participants and 44 cognitively normal control participants were measured. Untargeted GMB taxa was analyzed through 16S ribosomal RNA gene profiling based on next-generation sequencing and fecal metabolites were quantified by using ultrahigh performance liquid chromatography-mass spectrometry (UPLC-MS). Results Our analysis revealed that AD was characterized by 15 altered gut bacterial genera, of which 46.7% (7/15 general) was significantly associated with a series of metabolite markers. The predicted metabolic profile of altered gut microbial composition included steroid hormone biosynthesis, N-Acyl amino acid metabolism and piperidine metabolism. Moreover, a combination of 2 gut bacterial genera (Faecalibacterium and Pseudomonas) and 4 metabolites (N-Docosahexaenoyl GABA, 19-Oxoandrost-4-ene-3,17-dione, Trigofoenoside F and 22-Angeloylbarringtogenol C) was able to discriminate AD from NC with AUC of 0.955 in these 65 subjects. Conclusions These findings demonstrate that gut microbial alterations and related metabolic output changes may be associated with pathogenesis of AD, and suggest that fecal markers might be used as a non-invasive examination to assist screening and diagnosis of AD. Supplementary Information The online version contains supplementary material available at 10.1186/s12866-021-02286-z. Keywords: Alzheimer’s disease, Gut microbiota, Fecal metabolome, Inflammatory cytokines, Fecal markers Background Alzheimer’s disease (AD) is the most prevalent neurodegenerative disorder and is characterized by extracellular plaques composed of amyloid-β (Aβ) peptide and intracellular neurofibrillary tangles composed of hyperphosphorylated tau protein [[47]1]. Emerging evidences show that dysbiosis and alterations of the intestinal microbiota contribute to the development of neurodegenerative diseases, including Parkinson’ s disease, schizophrenia, stroke, epilepsy and depression via the “microbiota-gut-brain” axis [[48]2–[49]7]. A few studies have revealed the relationship between gut microbiota (GMB) and AD. Decreased diversity in microbiota is reported in AD and mild cognitive impairment (MCI) patients compared with normal controls, and changes in intestinal microbiota could be used for early detection of AD [[50]8–[51]10]. Some populations of enterotype bacteria are differential between demented and non-demented dementia [[52]11]. A recent study shows that the fecal and blood microbiota in MCI patients are similar to that of AD patients compared with normal controls, suggesting the bacterial dysbiosis preceded the onset of dementia [[53]12]. Fecal microbiota transplantation in mice also supports a possible connection between the GMB and AD [[54]13, [55]14]. The clarification of the gut-brain axis in AD pathogenesis would provide accessible markers from feces. Human intestinal tract harbors a complex community of microbes with the vast majority of the resident microbial population [[56]15], which plays an important role in the maintenance of human health, including energy extraction, vitamin biosynthesis, prevention of pathogen overgrowth and education of the immune system [[57]16]. Several studies suggest that GMB impact brain through a bidirectional communication system that is connected via neural, immune, endocrine, and metabolic pathways [[58]17–[59]20]. Although 16S ribosomal RNA (rRNA) gene sequencing allows for profiling of complex microbial composition and indicating the microbial entities [[60]21], annotation and the actual microbial activity of GMB is sparse as 16S rRNA gene sequence only provides a global composition analysis but cannot distinguish alive and dead microbes [[61]22]. Fecal metabolites, served as the metabolic output of both GMB and cellular metabolism occurring inside the human intestinal tract, could complement sequencing-based approaches with showing functional readout of microbial activity [[62]23]. Recent evidence indicated that the compounds secreted or influenced by bacteria may cause systemic inflammatory reactions, and then damage the blood-brain barrier (BBB) and promote neurodegeneration [[63]3, [64]24, [65]25]. A recent study reported that during AD progression, the alteration of GMB composition leads to the peripheral accumulation of phenylalanine and isoleucine and the increase of pro-inflammatory T helper 1 cell frequency in the blood in AD mouse models, which were also observed in MCI due to AD in human [[66]26]. Based on these findings, we sought to investigate potentially AD-relevant gut microbes and their metabolic outputs [[67]27], which has rarely been reported in AD. Therefore, in this study, we investigated the altered GMB composition and metabolic changes in the fecal samples that are relevant with AD, and examined their correlation by integrating the two omics findings. Based on the correlation analysis, we determined the potentially functional microbiota and its metabolic output, and further identified the potentially fecal markers of AD. Considering that compounds influenced by bacteria may cause systemic inflammatory reactions, and the altered cytokine levels had been reported as potential hallmark of AD, we also analyzed the association between differential fecal markers and peripheral inflammatory cytokines that linked with AD to further understand the function of altered GMB composition. Results Characteristics of study subjects The demographic and clinical information for 21 AD participants and 44 cognitively normal control (NC) participants was presented in Table [68]1. No statistically significant difference in proportion of sex was found (P < 0.05) between AD group and NC group. AD and NC groups also did not differ with respect to age at examination, education years and body mass index (BMI) (P < 0.05). The mean MMSE score was significantly lower in AD participants (18.0) than in NC participants (29.0), P < 0.001. APOE ε4 carrier rate was significantly higher in AD participants (50.0%) than in NC participants (11.9%), P < 0.001. Table 1. Clinical characteristics of AD subjects and cognitively normal control (NC) subjects Baseline characteristic AD NC P value^a Participants, n 21 44 Age, years (IQR) 76.2 (9.9) 78.4 (6.6) 0.262 Female, n (%) 8 (38.1) 24 (54.6) 0.215 Education, years (IQR) 12 (3) 12 (7) 0.889 MMSE scores, mean (IQR) 18 (8) 29 (2) < 0.001 APOE ε4, n (%) 10 (50.0)^b 5 (11.9)^c < 0.001 BMI, mean (IQR) 22.8 (3) 23.1 (5) 0.388 [69]Open in a new tab Values are shown as median (interquartile range, IQR) or number (percent) BMI Body mass index, AD Alzheimer’s disease, NC cognitively normal control ^a Mann-Whitney U test or x^2 test ^b Missing data of APOE genotype (n = 1) ^c Missing data of APOE genotype (n = 2) Genera of GMB between AD and NC participants After quality filtering and trimming, sequencing of the V3-V4 region of 16S rRNA gene generated 3.55 million sequence reads from 65 fecal samples with an average length of 419. At the 97% sequence similarity level, a total of 1014 OTUs were obtained and matched to 13 phyla, 20 classes, 35 orders, 71 families and 251 genera. The Venn diagram showed 53 unique OTUs in AD, 185 unique OTUs in NC and 776 shared OTUs between two groups (Additional file [70]1: Fig. S1A). The α-diversity did not differ between AD group and NC group (Additional file [71]2: Table S1). With respect to β-diversity, principal co-ordinates (PCoA) based on Bray-Curtis dissimilarity showed the significant differences of GMB composition between the fecal samples from AD participants and that from NC participants (PERMANOVA, R^2 = 0.025, P = 0.027) (Fig. [72]1a). Meanwhile, control analyses were performed to determine whether the potentially confounding variables influenced this global microbial phenotype [[73]28]. The results showed that the AD participants or NC participants were not clustered based on sex (Additional file [74]1: Fig. S1B), age (Additional file [75]1: Fig. S1C-D) or BMI (Additional file [76]1: Fig. S1E-F). Fig. 1. [77]Fig. 1 [78]Open in a new tab Gut microbial characteristics of Alzheimer’s disease (AD). A Principal co-ordinates (PCoA) analysis of microbiota in AD and NC participants (n = 21, AD; n = 44, NC, P = 0.027). B Taxonomic differences of gut microbiota bacterial between AD and NC participants analyzed by Linear discriminant analysis (LDA) effect size (LEfSe) (LDA > 2.5, P < 0.05). Histogram represents the LDA scores for differentially abundant genera. *P < 0.05; ** P < 0.01 Furthermore, LEfSe analysis (Fig. [79]1b) showed that 15 genera significantly differed between AD group and NC group (LDA values > 2.5, Wilcoxon rank-sum test, P < 0.05; Additional file [80]3: Table S2). Specifically, AD participants exhibited 13 increased genera compared to NC participants, which belonged to bacterial families Lachnospiraceae (genera Agathobacter, unclassified_f_Lachnospiraceae, Eubacterium_ventriosum_group, Lachnospiraceae_NC2004 and Coprococcus_1), Ruminococcaceae (genera Faecalibacterium and Ruminococcaceae_UCG-007), Prevotellaceae (genus Alloprevotella), Atopobiaceae (genus Atopobium), Clostridiales (genus Parvimonas), Synergistaceae (genus Cloacibacillus), Erysipelotrichaceae (genus Solobacterium) and Pseudomonadaceae (genus Pseudomonas). In contrast, AD participants exhibited 2 decreased genera compared to NC participants, which belonged to bacterial families Lachnospiraceae (genus Tyzzerella) and Erysipelotrichaceae (genus Erysipelatoclostridium). At phyla level, these 15 differential bacterial genera were classified to Firmicutes (11/15, 73.3%), Bacteroidetes (1/15, 6.7%), Actinobacteria (1/15, 6.7%), Proteobacteria (1/15, 6.7%) and Synergistetes (1/15, 6.7%). Fecal Metabolome between AD and NC participants Fecal metabolites, served as the metabolic output of both GMB and cellular metabolism occurring inside the human intestinal tract, is considered a functional readout of the GMB. As show in Fig. [81]2a, the OPLS-DA scores plot showed a clear variation between AD group and NC group (Permutation test: R^2Y = 0.70, Q^2Y = − 0.28 for positive ion, and R^2Y = 0.74, Q^2Y = − 0.15 for negative ion). Metabolome of AD participants showed 15 differential fecal metabolites relative to the Control group (P < 0.05, VIP > 1, Table [82]2), with decreased levels of 11 metabolites and increased level of 4 metabolites in AD. The implicated metabolites belonged to organic acids (3 metabolites: N-Docosahexaenoyl-GABA; Hypoglycin B and 12-Hydroxydodecanoic acid), lipids and lipid-like molecules (7 metabolites: 19-Oxoandrost-4-ene-3,17-dione; 5-Butyl-3,4-dimethyl-2-furanundecanoic acid; Trigofoenoside F; Sagittariol; 22-Angeloylbarringtogenol C; 1α,25-dihydroxy-3α-methyl-3-deoxyvitamin D3 and PG(16:0/0:0)[U]), Benzenoids (2 metabolites: (4E)-12-hydroxy-1-(4-hydroxy-3-methoxyphenyl)dodec-4-en-3-one and 5-3’, 5’-Dihydroxyphenyl-γ-valerolactone), Organic nitrogen (N,N-Dimethylsphingosine), Organoheterocyclic ((5α,8β,9β)-5,9-Epoxy-3,6-megastigmadien-8-ol) and Piperidine (1-ACETYLPIPERIDINE) (Table [83]2). KEGG pathway enrichment analysis revealed that the differentially expressed metabolites were involved in steroid hormone biosynthesis (19-Oxoandrost-4-ene-3,17-dione), N-Acyl amino acid metabolism (N-Docosahexaenoyl-GABA) and Piperidine metabolism (1-ACETYLPIPERIDINE) (Table [84]2). Fig. 2. [85]Fig. 2 [86]Open in a new tab Characteristics of metabolites between AD participants and NC participants and the correlation analyses. A The orthogonal partial least-squares discriminant analysis (OPLS-DA) plot of fecal metabolites in comparisons of AD and NC groups (left: positive ion; right: negative ion). B Pearson correlation between 15 most differential genera and 15 differential metabolites, inflammatory cytokines, clinical characteristics linked with AD. The results were presented as a heatmap. Red squares indicate positive associations; blue squares indicate negative associations. *P < 0.05, ** P < 0.01, *** P < 0.001 denoted statistical significance. C Associations of fecal metabolites with inflammatory cytokines and clinical parameters. Red squares indicate positive associations; blue squares indicate negative associations. *P < 0.05; ** P < 0.01 Table 2. Alteration of Metabolites in AD subjects compared with cognitively normal controls Compound Metabolite Predicted molecular pathways or biological functions^a VIP FC (AD/NC) P Lipids and lipid-like molecules  Androstane steroids 19-Oxoandrost-4-ene-3,17-dione Steroid hormone biosynthesis 2.4 0.73 0.023  Fatty acids 5-Butyl-3,4-dimethyl-2-furanundecanoic acid Antixoxidants and radical scavengers 2.1 0.80 0.034  Steroidal glycosides Trigofoenoside F A compound isolated from fenugreek seeds 1.5 1.07 0.011  Triterpenoids 22-Angeloylbarringtogenol C A constituent of C. sinensis 3.3 1.22 0.042  Diterpenoids Sagittariol – 2.6 0.87 0.042  Vitamin D derivatives 1α,25-dihydroxy-3α-methyl-3-deoxyvitamin D3 – 2.7 0.87 0.031 – PG(16:0/0:0)[U] – 1.3 0.95 0.019 Organic acids and derivatives  Amino acids N-Docosahexaenoyl GABA N-Acyl amino acid metabolism 2.3 0.81 0.001  Hydroxy acids 12-Hydroxydodecanoic acid Substrate of glutathione-dependent formaldehyde dehydrogenase 1.6 0.85 0.047  Amino acids Hypoglycin B Natural toxins 1.4 0.86 0.037 Benzenoids  Methoxyphenols (4E)-12-hydroxy-1-(4-hydroxy-3-methoxyphenyl)dodec-4-en-3-one A predicted metabolite generated by BioTransformer 2.1 0.67 0.020  Benzenediols 5-(3′,5′-Dihydroxyphenyl)-γ-valerolactone A tea polyphenol metabolite detected in biological fluids 2.3 0.73 0.041 Organic nitrogen compounds  Amines N,N-Dimethylsphingosine Regulation of sphingolipid-mediated functions 3.9 1.35 0.032 Organoheterocyclic compounds  Benzopyrans (5α,8β,9β)-5,9-Epoxy-3,6-megastigmadien-8-ol – 1.8 1.23 0.045 Piperidine  Piperidine 1-ACETYLPIPERIDINE Piperidine metabolism 1.9 0.80 0.041 [87]Open in a new tab AD Alzheimer’s disease, NC cognitively normal control, VIP variable importance in the projection, FC fold change ^a Information referencing KEGG pathway database and HDBM database Relationship between altered gut microbial genera, fecal metabolites, clinical characteristics and inflammatory cytokines linked with AD As shown in Fig. [88]2b, the Pearson correlation analyses indicated that 46.7% (7/15 genera) of altered bacterial genera were significantly associated with a series of altered fecal metabolites (P < 0.05, r > 0.30 or < − 0.30 [[89]29], associated with more than one metabolite), suggesting a general correlation between gut bacterial file and metabolites. The coefficient of correlation and the P value between bacterial genera and fecal metabolites were presented in Additional file [90]4: Table S3 and Additional file [91]5: Table S4, respectively. These genera mainly belonged to Lachnospiraceae (Lachnospiraceae_NC2004 and Agathobacter), Erysipelotrichaceae (Erysipelatoclostridium and Solobacterium), Ruminococcaceae (Faecalibacterium), Pseudomonadaceae (Pseudomonas) and Clostridiales (Parvimonas) at family level. Additionally, significant correlations were observed between some gut microbial genera and clinical parameters including sex, BMI, MMSE score, APOE genotype and G-CSF (Fig. [92]2b). These findings suggested that AD was simultaneously distinguished by changes in GMB and fecal metabolome. As shown in Table [93]3, AD participants presented significantly decreased expression of serum G-CSF and IFN-γ compared to NC participants (P = 0.008 and P = 0.007, respectively) (Table [94]3). A significant negative correlation of Trigofoenoside F metabolite with G-CSF (r = − 0.301, P = 0.019) and a significant positive correlation of 19-Oxoandrost-4-ene-3,17-dione metabolite with IFN-γ (r = 0.339, P = 0.006) were observed (Fig. [95]2c, Additional file [96]6: Table S5 and Additional file [97]7: Table S6). These findings suggested that some fecal metabolites might affect systemic inflammatory reactions during AD development. Table 3. Association of peripheral inflammatory cytokines and Alzheimer’s disease risk Cytokines Number of subjects Mean Median P value^a AD/NC AD NC AD NC IL-6 18/41 3.57 2.75 1.44 1.63 0.792 IL-8 18/43 7.38 6.77 6.09 5.99 0.794 IL-10 18/43 1.08 1.85 1.01 1.18 0.315 G-CSF 19/43 1.28 1.84 1.27 1.59 0.008 IL-7 19/43 2.14 2.22 2.05 2.05 0.667 IL-8 19/43 3.17 3.29 2.40 3.15 0.113 IL-12 19/42 1.32 1.50 0.95 0.95 0.849 IFN-γ 19/42 0.97 1.75 0.66 1.40 0.007 MCP-1 19/43 6.87 8.27 6.85 7.72 0.172 MIP-1β 19/43 8.33 10.95 6.45 8.28 0.194 TNF-α 19/39 2.27 2.67 1.61 2.12 0.314 [98]Open in a new tab AD Alzheimer’s disease, NC cognitively normal control ^a Comparison between AD group and NC group of median, using Mann-Whitney U test Potential contribution of fecal markers of AD Among the newly implicated 17 fecal markers in our study, including of differential 7 fecal genera and their 8 significantly correlated fecal metabolites, and 2 fecal metabolites that were correlated with differential inflammatory cytokines in AD, combination of 6 fecal markers, including of 2 genera (Faecalibacterium and Pseudomonas) and 4 metabolites (N-Docosahexaenoyl GABA, 19-Oxoandrost-4-ene-3,17-dione, Trigofoenoside F and 22-Angeloylbarringtogenol C), was able to discriminate AD from NC with an AUC of 0.955 in these 65 subjects. The AUC of each separate microbial (AUC = 0.798) or metabolic markers (AUC = 0.873) was significantly lower than that of combination of microbial and metabolic markers, P < 0.001 and P = 0.033, respectively (Fig. [99]3). Addition of inflammatory cytokines did not significantly improve the discriminative power of the combined model (P = 0.857). Fig. 3. [100]Fig. 3 [101]Open in a new tab Diagnostic markers for AD. Receiver operating characteristic curves for genetic models among Chinese subjects in the testing set. Genus group represents 2 microbial (Faecalibacterium and Pseudomonas). Metabolite group represents 4 metabolites (N-Docosahexaenoyl GABA, 19-Oxoandrost-4-ene-3,17-dione, 22-Angeloylbarringtogenol C and Trigofoenoside F). Cytokine group represents G-CSF and IFN-γ Discussion This study demonstrated that AD exhibited the disorder in microbial ecology, which may affect both metabolism and subsequent peripheral inflammatory cytokine in human subjects. Importantly, our study showed that the AUC for discriminating AD from NC was 0.955 for combination of fecal microbial markers and correlated metabolites (a total of six fecal markers), suggesting the potential of fecal markers as a non-invasive tool for screening or assistant diagnosis of AD. Further studies in large scale and longitudinal analysis can be justified to validate and improve the clinical utility of this fecal marker panel. In our study, AD was characterized by increased 13 genera, of which 5 genera belonged to family Lachnospiraceae. Our results confirmed the previous report by chi et al. that the higher abundance of family Lachnospiraceae (general Fusicatenibacter, Blautia, and Dorea) was negatively associated with MMSE score [[102]30]. In addition, a metagenomic study indicated the association of the family Lachnospiraceae with Type 2 diabetes and the intestinal colonization by the Lachnospiraceae contributes to the development of diabetes [[103]31]. It is well known that diabetes and insulin resistance were risk factors of AD [[104]32–[105]34], therefore, a potential mechanism by which gut Lachnospiraceae influences AD development might be resulted from promoting the development of diabetes. Our results showed that the abundance of genera Pseudomonas and Faecalibacterium increased in AD participants and these two genera contributed to fecal marker panel for effectively discrimination of AD from NC (Fig. [106]3). The genus Pseudomonas was found to be a possible risk factor of AD, as it increased both in fecal sample (our study results) and blood sample [[107]12] from AD subjects, compared with that in cognitive normal controls. Functional studies reported that P. aeruginosa infection could increase endothelial tau phosphorylation and permeability [[108]35, [109]36], which is a common pathophysiological mechanism in AD pathogenesis. In addition to the pathogenic P. aeruginosa strain, the genus Pseudomonas also comprises non-pahtogenic strains, such as P. syringae strain [[110]37]. So far, little is known about how interactions between different pathogenic and non-pathogenic Pseudomonas strains within the host impact. Future studies should furtherly identify the specific pathogenic strain responsible to AD and explore the interaction between pathogenic and non-pathogenic strains on the occurrence of AD. With regard to Faecalibacterium, F. prausnitzii (F. prausnitzii) is the founding strain and has been studied extensively. F. prausnitzii plays a beneficial role in anti-inflammatory and promoting gut health [[111]38, [112]39]. Several studies reported lower abundance of F. prausnitzii in patients with inflammatory bowel diseases [[113]40, [114]41], type 2 diabetes [[115]42] and colon cancer [[116]43]. However, higher abundance in F. prausnitzii was reported in pediatric patients with Crohn’s Disease compared with controls [[117]44, [118]45], indicating the detrimental role of F. prausnitzii, at least in some contexts. Lopez-Siles et al. [[119]46] has found that patients with Crohn’s Disease may harbor a unique set of F. prausnitzii OTUs, as compared with controls. A study with deep metaproteomics approach has defined five available strain proteomes of F. prausnitzii. Therefore, inconsistent role of F. prausnitzii may be due to the different function of different strains of Faecalibacterium genera, which might be the potential confounders for the ROC curves evaluated in our study. Future research is needed to determine the specific strains that relevant to AD. Our results reported a higher abundance of gut Faecalibacterium in AD than NC group, providing additional evidence of detrimental role of Faecalibacterium. One explanation of may be that Faecalibacterium could reduce secretion of proinflammatory cytokines IFN-γ [[120]38]. Then, reduced IFN-γ causes increase of Aβ deposits [[121]47]. In fact, we also found the lower peripheral IFN-γ expression in AD compared with NC group, which supports this hypothetical explanation. AD participants in our study showed alteration of fecal metabolome compared to NC participants. Correlation analysis provided a significant association of altered bacterial genera with altered metabolites. Furthermore, pathway analysis linked the alteration of metabolites to three metabolic pathways including Steroid hormone biosynthesis, N-Acyl amino acid metabolism and Piperidine metabolism. Together, these findings suggest the potential mechanism of GMB affecting occurrence or development of AD. Future functional studies are justified to formally verify that if and how these metabolic pathways participate in the occurrence or development of AD. Notably, among 15 altered metabolites in AD, 7 metabolites belong to lipid or lipid-like molecules, suggesting that lipid-mediated metabolic or synthetic processes in gut are linked with AD. The 19-Oxoandrost-4-ene-3, 17-dione, a decreased lipid-like molecular in AD compared with NC group in our study, was involved in estrogen biosynthesis. Estrogen is a potent factor that not only prevents vascular disease but also plays an important role in the preservation of neurons and reparation of neurons damaged by the disease [[122]48]. A population-based study indicated higher frequency of estrogen use in non-patients compared with Alzheimer’s disease patients [[123]49]. Thus, decreased 19-Oxoandrost-4-ene-3, 17-dione in gut may influence the AD pathology by downregulating the estrogen. 5-Butyl-3, 4-dimethyl-2-furanundecanoic acid, another decreased lipid-like metabolite in AD, is a furan fatty acid (F-acid). F-acids play an protective role for Alzheimer’s disease progression mainly by preventing lipid peroxidation [[124]50, [125]51] and protecting polyunsaturated fatty acids (PUFAs) [[126]52], suggesting possible mechanism for lower level of 5-Butyl-3, 4-dimethyl-2-furanundecanoic acid in AD than NC group. Regarding to amino acid metabolism, all three metabolites belonging to organic acids showed decreased level in AD, among which N-Docosahexaenoyl GABA was annotated to be involved in N-Acyl amino acid metabolism. Consistent with our results, N-acyl amino acids have anti-inflammatory and neuroprotective effects [[127]53], but their biosynthesis and catabolism have not been fully elucidated [[128]54]. Additionally, we observed decreased serum G-CSF and IFN-γ level in AD participants compared with NC participants. The consistent results were also observed in previous clinical and animal studies. Decreased plasma levels of G-CSF were previously reported in patients with early AD in comparison with healthy controls [[129]55]. Moreover, in AD participants, G-CSF levels showed a significant inverse correlation with amyloid-β (Aβ1 − 42) levels in cerebrospinal fluid [[130]55]. Mouse model experiment proved that subcutaneous administration of G-CSF into two different Aβ–induced AD mouse models could substantially rescue their cognitive/memory functions [[131]56]. Additionally, in the stroke mouse model, administration of G-CSF could lead to neurogenesis near the injured area and the neurological and functional recovery [[132]57–[133]59]. With respect to IFN-γ, although the relationship between IFN-γ level and AD has not been reported, IFN-γ overexpression was found to reduce Aβ deposits and peripheral monocytes infiltration in amyloid precursor protein (APP) transgenic mice [[134]47]. Additionally, in both 5XFAD and APP/PS1 mice model at intermediate stage of AD, reversion of decreased IFN-γ level by transient depletion of Treg cells could increase infiltration and recruitment of leukocyte to Aβ plaques and then alleviate the AD pathology [[135]60]. The findings of these functional studies support our observation that IFN-γ is decreased in AD compared with NC group. By integrating the correlation analyses of three groups of markers (GMB, fecal metabolites and peripheral inflammatory cytokines), we found two potential mechanistic pathways for the activity and function of GMB in AD pathogenesis. One pathway was that decrease of 19-Oxoandrost-4-ene-(3,17)-dione metabolite caused by low abundance of gut Erysipelatoclostridium might lead to reduction of peripheral IFN-γ level, which subsequently influenced AD pathogenesis. There are some evidences supporting the potential of this mechanism.19-Oxoandrost-4-ene-(3,17)-dione was involved in estrogen biosynthesis as we discussed above, besides, administration of estrogen to mouse could enhance the secretion of IFN-γ by mouse spleen cells [[136]61] and by CD4+ human T cell clones [[137]62]. The second pathway was that increase of Trigofoenoside F metabolite caused by high abundance of gut Cloacibacillus might lead to reduction of peripheral G-CSF level and then influenced AD pathogenesis. Though there are no other references supporting this mechanism for now, our findings have