Thirty-five C57BL/6J mice (5-month-old, 14 females and 21 males) were obtained from the Animal Facility of Aging Science in Korea Basic Science Institute (Gwangju, Republic of Korea). The mice were individually housed in a specific-pathogen-free (SPF) facility with ad libitum access to sterilized food and water. The housing environment was maintained at 22 ± 1°C and 40–50% humidity with a 12 h light/dark cycle. Microbiological monitoring in a specific-pathogen-free (SPF) facility was performed using Laser scanning confocal microscope (Leica TCS SP5 AOBS/Tandem, KJ302) at the Honam Regional Center of Korea Basic Science Institute (KBSI). The entire experiment was conducted within 1 week after the stabilization period. Following body weight measurement and blood collection, planned behavioral analysis experiments were conducted. All tests were performed during the light phase (09:00–18:00) to ensure consistency. After all behavioral analyses were completed, excreted feces (Feces) and gastrointestinal tract (GI) samples were collected for microbiome analysis. Freshly excreted feces were obtained from each mouse prior to sacrifice, representing the microbiota expelled from the gut. To collect GI samples, mice underwent a 24-h fasting period before sacrifice to minimize residual food contents in the gut. After sacrifice, the entire gastrointestinal tract, from the stomach to the end of the large intestine, was carefully dissected. The luminal contents and mucosal surfaces were collected by scraping the inner lining of the digestive tract, ensuring that the GI samples contained both luminal microbiota and potentially mucosa-associated microbiota.

The serum levels of total cholesterol (TCHO), triglyceride (TG), high-density lipoprotein-cholesterol (HDL-CHO) and blood glucose levels were determined using commercial assay kits (Asan Pharmaceutical, Seoul, Republic of Korea) following the manufacturer’s instructions, and the low-density lipoprotein-cholesterol (LDL-CHO) levels were calculated using Friedewald’s equation, as previously described (Baek et al., 2014; Chung et al., 2016).

This protocol describes four behavioral tasks for mice to assess motor function, cognitive function, and emotional function. To investigate the relationship between microbial diversity and host behavior, the rotarod test, wire suspension test, Y-maze spontaneous alternation test, and tail suspension test were conducted. Mice were categorized into high, medium, and low groups based on performance in behavioral tests.

Depression-like behavior was examined using the tail suspension test described previously (Ueno et al., 2022). The testing apparatus consisted of white acrylic walls (20 × 30 × 60 cm) and one open side for observation. Each mouse was suspended by its tail 60 cm above the chamber floor using adhesive tape placed near the tip of its tail (less than 1 cm). The mouse’s behavior was observed by an observer for 6 min and the time of immobility was recorded. Mice were classified into high (more movement), medium, and low (less movement, depressed) groups based on immobility time in the tail suspension test, a measure of emotional despair.

The DNA from each individual fecal and gastrointestinal (GI) sample was extracted using the phenol-chloroform-isoamyl alcohol extraction method, following the same procedure as previously described (Ahn et al., 2023a; Ahn et al., 2024; Zhai et al., 2022). To ensure DNA quality, both concentration and purity were measured using a BioSpec-nano spectrophotometer (Shimadzu Biotech, Japan), and DNA integrity was confirmed using 1% (w/v) agarose gel electrophoresis. The extracted DNA was sent to a specialized sequencing facility (Ebiogen, Inc., Republic of Korea) for Next-Generation Sequencing (NGS) using the Illumina 16S Metagenomic Sequencing Library protocol. This involved amplifying the V3–V4 region of the 16S rRNA gene using designed primers, as previously described (Ahn et al., 2023a; Ahn et al., 2024; Zhai et al., 2022). After amplification, additional steps were performed to incorporate a multiplexing index and Illumina sequencing adapters. Each sample underwent individual normalization and pooling once before sequencing, utilizing PicoGreen. Library size confirmation was performed using the Agilent TapeStation DNA ScreenTape D1000 system (Agilent Technologies, Santa Clara, CA, United States). The final pooled libraries underwent sequencing (2 × 300 paired-end reads) using the Illumina MiSeq platform (Illumina, San Diego, CA, United States). Amplicon error correction was performed by modeling from merged fastq files using DADA2 (Ver. 1.10.1). This process included filtering out noise sequences, correcting errors in marginal sequences, removing chimeric sequences and singletons, and de-replicating sequences (Callahan et al., 2016a).

All data and statistical analyses were conducted as previously described (Ahn et al., 2023a; Ahn et al., 2024; Zhai et al., 2022). Briefly, bacterial species were classified using Q2-Feature Classifier, a Naive Bayes classifier trained on the SILVA reference (V3–V4 region) database.¹ Following parameter configuration using the Denoise single function, the dataset was classified. Diversity calculations and statistical tests were conducted utilizing the q2 diversity option, specifically focusing on “sampling depth.” A minimum sequencing quality score threshold of 20 and a rarefaction depth of 11,510 were applied. Subsequently, upon validating the sequencing results’ quality, the “table.qzv” file was filtered using thresholds in QIIME 2. The metagenomic data OTU and taxonomic classification tables were imported into R version with the phyloseq (1.28.0) and MetagenomeSeq (version 1.16.0) packages. Following established guidelines, we imported metadata, OTU, and taxonomic classification tables into phyloseq, where the data was processed (Callahan et al., 2016b). To normalized the data specifically for metagenomic analysis, we converted phyloseq objects to MetagenomeSeq objects and applied Cumulative Sum Scaling (CSS) from the Bioconductor package metagenomeSeq (version 1.16.0). Finally, the normalized data was converted back to phyloseq class objects for further analysis and visualization in R (version 3.6.1). The α-diversity, β-diversity analyses and relative abundance evaluation were conducted as previously described (Ahn et al., 2023a; Ahn et al., 2024; Zhai et al., 2022). The α-diversity was statistically analyzed using the Kruskal–Wallis rank sum test. The β-diversity was computed using Bray–Curtis dissimilarity on log-transformed OTU data. Non-metric multidimensional scaling (NMDS) was employed using the “MetaMDS” function from the “vegan” package to assess relationships between samples by reducing dimensionality while retaining information on sample relationships. For abundance calculations, the normalized OTU data were utilized, and taxa were grouped at the phylum level for plotting. To enhance clarity in visualizing abundance data, classification groups with a relative abundance of less than 0.5% were aggregated as “Others.”

To identify differences in bacterial populations between the two groups following each behavioral test, differential abundance analysis was executed following established procedures (Ahn et al., 2023a; Ahn et al., 2024). Utilizing DESeq2 (version 1.24.0), we analyzed the differences in the types and amounts of bacteria found in different groups. This analysis focused on Feces and GI samples collected after each behavioral test (Love et al., 2014; Paulson et al., 2013). Taxonomic species present in less than 1% of the samples were disregarded in the DESeq2 analysis.

Microbiability was calculated to estimate the proportion of phenotypic variance explained by the abundance of specific microbial species (He et al., 2022). For each selected species, a linear model (lm) was fitted using the species abundance as the fixed effect and the host phenotype as the dependent variable. The total phenotypic variance was computed, and the contribution of each microbial species was determined based on the model’s R² value. Microbiability was then obtained by dividing the species-specific variance by the total phenotypic variance. To assess the statistical significance of the association between microbial species and host phenotypes, p-values were extracted from the regression model coefficients. The p-values corresponding to the microbial abundance variable were reported to evaluate whether the observed microbiability was statistically significant. All calculations were performed in R (version 4.3.3) using standard statistical packages.

All statistical analyses are reported as the mean ± S.E.M., and the differences in the relative abundance of bacterial populations containing feces and GI were analyzed using the Mann–Whitney sum rank tests in R software. Group comparisons in the NMDS plots were statistically analyzed using the PERMANOVA test, while most other group comparisons were performed using Welch’s t-test. The p-value less than 0.05 was considered significant.

In order to compare fecal microbiomes and gastrointestinal microorganisms and discover useful intestinal microbiomes, 16S rRNA sequencing of feces (Feces) and gastrointestinal tract (GI) was performed after behavioral analysis experiments on 35 6-month-old mice (Figure 1A). As a result of comparing the alpha diversity of Feces samples and GI samples, it was observed that the richness factors Fisher (p < 0.001) and ACE (p < 0.001) were larger in GI samples, and the evenness factors Shannon (p < 0.05), Simpson (p < 0.001), InvSimpson (p < 0.001), and Evenness (p < 0.001) were slightly smaller in GI samples than in Feces samples (Figure 1B and Supplementary Table S1). In more detail, as a result of checking the microbial composition of the Feces samples and the GI samples, it was confirmed that Actinobacteria (p < 0.001), Bacteroidetes (p < 0.001), and Patescibacteria (p < 0.001) were reduced, and Firmicutes (p < 0.01), Proteobacteria (p < 0.01) and Verrucomicrobia (p < 0.001) were significantly increased in the GI samples compared to the Feces samples (Figure 1C and Supplementary Table S2). In addition, through the NMDS plot drawn as a result of the analysis of the Feces samples and GI samples, it was seen that the microbial composition between the Feces samples and GI samples was clearly different (p = 0.001), and that there were clear differences between each individual in GI samples compared to Feces samples (Figure 1D and Supplementary Table S3). Through this, it is assumed that it will be possible to clearly distinguish differences in intestinal microorganisms according to phenotype in GI samples rather than Feces samples.

To investigate the relationship between gut microbiota obtained from two sampling methods and mouse phenotypes, we evaluated motor, cognitive, and emotional functions (Supplementary Table S4). Motor function was assessed using the rotarod and wire suspension tests. Based on performance records, mice were categorized into three groups: the high group (n = 9; rotarod record 30.81 ± 4.52 s, wire suspension record 79.7 ± 9.9 s), the medium group (n = 10; rotarod record 19 ± 3.12 s, wire suspension record 44.03 ± 3.96 s), and the low group (n = 9; rotarod record 15.78 ± 2.04 s, wire suspension record 29 ± 3.52 s). Body weight and biochemical analyses among these groups revealed that the high group had a lower body weight than the other groups. However, blood glucose, total cholesterol (TCHO), high-density lipoprotein cholesterol (HDL-CHO), low-density lipoprotein cholesterol (LDL-CHO), and triglyceride (TG) levels in the blood remained within the normal range, with no significant differences among the groups.

Cognitive function was evaluated using the Y-maze test, and groups were classified based on the calculated percentage alternation values. Mice were divided into the high group (n = 10; % alternation 68.67 ± 1.31%), the medium group (n = 11; % alternation 59.79 ± 0.83%), and the low group (n = 10; % alternation 50.68 ± 0.97%). Body weight and biochemical parameters among these groups showed no significant differences, and all values remained within the normal range.

Emotional function was assessed using the tail suspension test, and mice were grouped based on their immobility time. The high group (n = 13) had an immobility time of 115.31 ± 6.47 s, the medium group (n = 10) recorded 166.03 ± 4.28 s, and the low group (n = 12) exhibited 197.58 ± 2.95 s. Significant differences were observed among these groups. Additionally, body weight and biochemical analyses revealed that the high group had lower body weight, blood glucose, TCHO, HDL-CHO, LDL-CHO, and TG levels compared to the other groups.

We previously identified gut microbiota that were abundant in groups with enhanced and reduced functionality through gut microbiome analysis related to behavioral assessments. Since gut microbiota influence host phenotypes, the differences in the abundance of these specific microbes may have impacted motor function, cognitive function, and emotional function. Based on this, we hypothesized that microbial species enriched in groups with higher motor, cognitive, and emotional function levels may be beneficial bacteria (Table 1).

Analysis of Feces samples related to motor function revealed that Aestuariispira insulae, Parabacteroides goldsteinii, Ligilactobacillus animalis, Anaerocolumna jejuensis, Caecibacterium sporoformans, and Alistipes finegoldii were more abundant in the high group (Figure 2E and Table 1). Among these, P. goldsteinii (high vs. low, p < 0.01) and A. finegoldii (high vs. low, p < 0.05) showed significant differences in % abundance across groups, suggesting their potential role as beneficial bacteria for motor function. In contrast, the analysis using GI samples identified Anaerostipes butyraticus, Cellulosilyticum ruminicola, Caecibacterium sporoformans, Mogibacterium neglectum, Faecalicatena contorta, [Clostridium] polysaccharolyticum, Roseburia faecis, Pseudoflavonifractor phocaeensis, and Vallitalea pronyensis as more abundant in the high group (Figure 2F and Table 1), but their % abundance did not show statistical significance.

In terms of cognitive function, the analysis of Feces samples showed that Anaeroplasma abactoclasticum, Aestuariispira insulae, Prevotella oralis, Alistipes senegalensis, Ligilactobacillus animalis, Clostridium oryzae, Limosilactobacillus reuteri, Murimonas intestini, and Parabacteroides goldsteinii were enriched in the high group (Figure 3E and Table 1). Among these, A. abactoclasticum (high vs. low, p < 0.05), L. animalis (high vs. low, p < 0.01), C. oryzae (high vs. low, p < 0.01), and M. intestine (high vs. low, p < 0.05) exhibited significant differences in % abundance, indicating their potential association with cognitive function. In GI samples, Phocaeicola barnesiae, Eisenbergiella massiliensis, Anaerotignum lactatifermentans, Oscillibacter ruminantium, and Flintibacter butyricus were more abundant in the high group (Figure 3F and Table 1). However, no statistically significant differences were observed in % abundance.

In case of emotional function, Feces sample analysis revealed that Caecibacterium sporoformans, Anaerocolumna jejuensis, [Clostridium] populeti, Marvinbryantia formatexigens, and Limosilactobacillus reuteri were more abundant in the high group (Figure 4E and Table 1), though there were no significant differences in % abundance between the high and low groups. In GI samples, Phocaeicola barnesiae, Prevotella oralis, Lacrimispora aerotolerans, Acutalibacter muris, Turicibacter sanguinis, Clostridium amazonense, [Eubacterium] siraeum, Eisenbergiella massiliensis, Lacrimispora indolis, Hungatella xylanolytica, and Ruminiclostridium cellulolyticum were found to be enriched in the high group (Figure 4F and Table 1). Among these, [Eubacterium] siraeum (high vs. low, p < 0.05) exhibited statistical significance, suggesting a potential beneficial role in emotional function.

To further assess the potential impact of these selected beneficial microbes on host phenotypes, we quantified microbiability (He et al., 2022), which estimates the proportion of phenotypic variance explained by microbial abundance (Table 2). The microbiability analysis demonstrated that microbes identified from Feces and GI samples exhibited different levels of influence on motor, cognitive, and emotional functions. In particular, the microbial diversity of the microorganisms selected from the Feces sample was higher than the GI sample, and the microbial diversity of the microorganisms selected from the GI sample was higher than that from the Feces sample, indicating that the choice of sampling method can influence the identification of microbes that are more strongly associated with host phenotypic traits (Table 2). Overall, our findings highlight that the selection of beneficial gut microbes and their estimated contribution to host phenotypic variance are dependent on the sampling method. These results emphasize the importance of considering sample type when studying gut microbiota’s functional role in host physiology.