This study was conducted in a subtropical evergreen broad-leaved forest of Dujiangyan City (altitude 600–1,000 m, 31°04′N-31°05′N, 103°42′E-103°42′E), Sichuan Province, China. The local annual mean temperature is 15.2°C, and annual precipitation is 1,200–1,800 mm. The climate is often cloudy with few sunny days. The major rodent species include Apodemus draco, Apodemus chevrieri, Niviventer fulvescens, Niviventer niviventer, and Leopoldamys edwardsi (Yang et al., 2018, 2022b).

Wire live traps (L × W × H = 30 cm × 13 cm × 12 cm), baited with fresh chestnuts, were used to capture small rodents. Following Yang et al. (2018), we set 4 × 10 trapping grids with an interval of 10 m in each plot. Traps were settled at 14:00–16:00 in the afternoon and checked at 08:00–10:00 the next morning. Captured rodents were identified to species and age. Then we recorded sex, body mass, and reproductive status (females pregnant, lactating or not; males with testes descended or not).

Forty-eight males and twenty-eight females of captured A. draco were selected. Prior to the experiments, A. draco was housed in plastic cages individually (L × W × H = 30 cm × 15 cm × 20 cm) with sawdust bedding and provided with rat chow (Shenyang Maohua Biotechnology Co. Ltd., Liaoning, China) and tap water ad libitum. The temperature of the feeding chamber was maintained at 20–25°C. Cotton was added for insulation in winter. All subjects were exposed to a 14 h light / 10 h dark photoperiod. About 4 months passed from the day of the capture of mice until the end of the tests. All animals were housed in cages under laboratory conditions and fed with rat chow once captured from the field. The numbers of animals used in each experiment and category were shown in Supplementary Table S1.

In this study, there were five larger experimental enclosures (L × W × H = 10 m × 10 m × 2.5 m; see Yi X et al., 2021 for details). The top of the enclosures was covered by iron sheets to prevent any predators from entering the enclosure. We divided each of the five larger enclosures by using 0.6 m high iron sheets to obtain four enclosures (Supplementary Figure S1). Thus, a total of 20 enclosures (L × W × H = 5 m × 5 m × 0.6 m) were used to evaluate the scatter-hoarding intensity of the rodents. We reinforced the separation walls with bricks settled on the wall bottom and iron wires linked to the ceiling. In each enclosure, the floor was built of bricks, and contained 16 shallow pits (L × W × D = 24 cm × 12 m × 6 cm) separated by 1 m to simplify the experimental design by referring to Yi X et al. (2021). All pits were filled with sand for rodents to cache seeds. One nesting box (L × W × H = 40 cm × 40 cm × 40 cm) and a plate were placed at the same corner of each enclosure to allow rodents to rest and drink freely. A central area of 0.5 m × 0.5 m of each enclosure was settled as the seed station.

All subjects were transported individually to the enclosure for the scatter-hoarding intensity test. An obligatory acclimation for one night was conducted for all animals before the tests. Animals were provided with chestnuts and water ad libitum. After acclimatization, we cleared away the uneaten food at 08:00 to ensure that the animals were starved for 8 h. Chestnuts were used in this study and tagged with a numbered plastic tag for relocation and identification. Before seed release, we drilled a 0.3-mm hole through the husk of chestnuts and tied the numbered plastic tag with a thin 10 cm long steel thread through the hole. At 16:00, 20 labeled chestnuts were placed in the seed station. Subjects were allowed to interact freely with the seeds until the next morning. At 08:00 the next morning, subjects were moved out of enclosures, and all seed fates were recorded (intact in situ, IIS; eaten in situ, EIS; scatter hoarded, SH; larder hoarded, LH; eaten after removal, EAR; eaten in the nest, EIN; intact after removal, IAR) (Wang and Yi, 2022). After each experiment, the enclosures were cleaned and ventilated for at least one night. The enclosure test was performed three times for each subject to achieve reliable data.

By referring to a previous study (Pan et al., 2013), the subjects were divided into scatter-hoarding group (SG) or non-scatter-hoarding group (NG) according to the result of the scatter-hoarding intensity test. The SG is comprised of the rodents that scatter-hoarded ≥5% of seeds at least once in three-time tests, while the NG is composed of rodents that scatter-hoarded <5% of the seeds. The scatter-hoarding intensity was evaluated according to the number of SH seeds. Because scatter hoarding behavior is often disrupted in enclosure tests, thus, we also referred to the other behaviors in defining the individuals of SG and NG groups. In this work, we observed that individuals adopting scatter hoarding strategy also had a larger hoarding intensity, but less eating intensity in situ or after seed removal as compared to the NG group.

Following Denninger et al. (2018), we set four testing arenas (L × W × H = 40 cm × 40 m × 40 cm) in a 2-by-2 manner on the floor with environmental cues arranged on trilateral sides (Supplementary Figure S2). A camera was settled above the arenas, about 1.7 m high to the floor. All procedures were recorded on video.

To evaluate the memory, we performed a memory test consisting of an object location test (OLT) and a novel object recognition test (NORT). Animal habituation, training, and testing of the OLT and the NORT were performed under the instruction of Denninger et al. (2018). All animals under test were numbered and the test order was arranged randomly. The inter-trial intervals (ITI) were extended to 24 h to assess the long-term memory (contrary to short-term memory which lasts for about 20 s, the retention of long-term memory lasts from minutes to multiple years) of rodents. The 4-day experiment was conducted from 08:00 to 18:00. We brought animals to the test room for a 30-min acclimation before starting the test.

During the habituation session, mice were placed in the arenas (one mouse per arena) faced to the release corner (Figure 1A) and allowed to explore the arenas freely for 10 min. In the training trial, two objects were affixed 12 cm × 12 cm away from 2 non-release corners, and mice were placed and faced the same release corner (Figure 1B) and were able to explore the environment for 10 min. In the OLT, one of the two objects in the training trial was moved to another non-release corner (Figure 1C), and mice were allowed for a 10-min exploration. In the NORT, we used a novel object to replace the object that was not moved in the OLT (Figure 1D) and repeated the 10-min exploration test.

After each test, mice were sent back to their housing cages individually and all arenas were cleaned with 75% ethanol to minimize olfactory cues. All procedures were video recorded by an overhead camera. The movements were tracked and assessed with EthoVision® XT (Noldus, Wageningen, Netherlands). The valid investigation time was scored when the subject pointed its nose at the object at a maximum distance of 2 cm from that object. The time that the subject climbed on or jumped off the object was excluded. Moreover, the mice that did not investigate the objects were excluded from the analyses since the memory test was designed by allowing the mice to explore and remember the original objects in previous trials and recognize which object was removed or replaced in the following trials. The percentage of investigation time was calculated as 100×tnovel/tnovel+tsimilar . Values above 50 indicate preferences for object location or novel object.

To investigate whether the gut microbiota could mediate memory and the intensity of scatter-hoarding of rodents, FMT was conducted to alter the structure and function of the gut microbiota of the recipient animals. The SG and NG groups were selected as the donors and recipients, respectively. Fresh feces from donors (100 mg) were resuspended in 1 mL of sterile 0.9% saline by vortexing for 5 min. Then we centrifuged the supernatant (500 g, 1 min) to prepare a bacterial suspension. Forty-three NG mice were used as recipients (22 males and 21 females), which were randomly divided into two groups: the FMT treatment group NG-SG (11 males and 10 females) received bacterial suspension from SG donors, while the control group NG-NS (11 males and 11 females) received sterile 0.9% saline. Animals of both recipient groups were first treated with 200 μL antibiotics (100 μg/mL neomycin, 50 μg/mL streptomycin, and 100 U/mL penicillin) via intragastric gavage for successive 3 days of administration. After antibiotic treatment, the NG-SG and NG-NS groups were administered by gavage with 200 μL bacterial suspension from SG donors and sterile saline for 30 days every 3 days, respectively. After FMT, the scatter-hoarding intensity test and memory test were conducted to evaluate the differences in hoarding behavior and memory among the NG-SG and NG-NS groups. For fecal collection, mice were placed in sterilized cages (L × W × H = 30 cm × 15 cm × 20 cm). At the same time, fresh feces were collected from each mouse within 10 min with forceps and then stored at −80°C for subsequent 16S rRNA analysis (Zhu et al., 2022). After sample collection, the mice were returned to their cages.

Total genome DNA from samples was extracted using the cetyltrimethylammonium bromide (CTAB) method. The purity and concentration of DNA were evaluated on 1% agarose gels. Then DNA was diluted to 1 ng/μL with sterile water. Forward primer 341F (5’-CCTAYGGGRBGCASCAG-3′) and reverse primer 806R (5’-GGACTACNNGGGTATCTAAT-3′) were adopted for amplification of the V3–V4 regions of 16S rRNA gene. The library was generated using NEB Next^® Ultra DNA Library Prep Kit (Illumina, United States) and added with index codes. The quality of the library was assessed using Agilent 5,400 (Agilent Technologies Co Ltd., United States). Finally, the library was sequenced on the Illumina NovaSeq platform and 250 bp paired-end reads were generated. Sequencing services were provided by Wekemo Tech Group Co., Ltd. Shenzhen China.

Demultiplexed sequences were filtered for quality, then trimmed, de-noised, and merged with the Quantitative Insights Into Microbial Ecology (QIIME, version 2) software suite (Callahan et al., 2016). DADA2 plugin was used to obtain the feature table of the amplicon sequence variants (ASVs) by removing the chimeric sequences. Taxonomic classification of ASVs was conducted in QIIME2 using classify-sklearn with a pre-trained Greengenes (13_8 release) 99% database with a confidence threshold of 0.7 (Bokulich et al., 2018). Diversity metrics were generated from the ASVs feature table using the core-diversity plugin within QIIME2. The minimum sampling depth is 60,217 reads in our study. Microbial community of different relative abundance among samples and groups was identified with appropriate methods, including analysis of the composition of microbiome (ANCOM), and linear discriminant analysis (LDA) Effect Size (LEfSe) while the DESeq2 analysis was based on absolute abundance (Segata et al., 2011; Love et al., 2014; Mandal et al., 2015).

Data were analyzed using R software. Kolmogorov–Smirnov tests were used to assess the variance and normality of data on seed fates and memory tests. Data on seed fates were not normally distributed, therefore we used generalized estimating equations (GEE; R package “gee”; category × sex; Zeger et al., 1988) to analyze the differences in seed fates between groups. As the data on memory were normally distributed, the difference of groups in memory tests was evaluated using two-way ANOVA (category × sex). Alpha diversity was assessed in QIIME2 by the Kruskal-Wallis test applied to the Chao1, Observed, Shannon, and Simpson indices calculated from the obtained ASVs. Rarefaction curves were calculated using the Observed index at the ASV level with OriginPro (Version 2023, OriginLab Corporation, Northampton, MA, United States). Based on the Bray-Curtis distance, permutational multivariate ANOVA (PERMANOVA) was used to evaluate the differences between groups in beta diversity (adonis, permutation = 9,999). The results of beta diversity were visualized via nonmetric multidimensional scaling (NMDS) and partial least squares discriminant analysis (PLS-DA) with the “mixOmics” package in R software (Rohart et al., 2017). The level of statistical significance was set at p < 0.05. We focused on analyses of differences between SG and NG, and between NG-SG and NG-NS groups considering the treatment they experienced.

Animal raising and handling were in accordance with the guidelines of Animal Use and Care Committee, Institute of Zoology, Chinese Academy of Sciences. Pregnant and immature animals were excluded from the tests, and released immediately on-site. Animals were acclimated to the facility for at least one week before any experiment. All animals were released to the wild after our experiments.

There was a large variation in seed fates between SG and NG. Specifically, compared with the NG, scatter-hoarded intensity (SH; χ² = 14.004, df = 1, p < 0.001) and the numbers of eaten after removal (EAR; χ² = 7.496, df = 1, p = 0.006) seeds were significantly increased in SG, while the numbers of intact in situ (IIS; χ² = 3.868, df = 1, p = 0.049) and eaten in the nest (EIN; χ² = 29.122, df = 1, p < 0.001) seeds were significantly decreased (Figure 2A). There was no significant difference in the numbers of eaten in situ (EIS), larder hoarding (LH), and intact after removal (IAR) between SG and NG (all p > 0.05). No significant difference between genders and sex-by-group interaction was found in all seed fates (all p > 0.05). In total, mice can be divided into two distinct groups in hoarding seeds: SG mice showed higher intensity of scatter hoarding (SH), eaten after removal (EAR), and intact after removal (IAR), but lower intensity of intact in situ (IIS), eaten in situ (EIS), eaten in the nest (EIN) and larder hoarding (LH) than NG mice.

Results of both the object location test (OLT) and novel object recognition test (NORT) showed that the memory of non-scatter-hoarding animals is inferior to that of scatter-hoarding animals (Figures 2B,C). The OLT showed a marginally significant higher preference for the new location in the SG than NG (F_(1,51) = 3.549, p = 0.065; Figure 2B). The NORT results showed that SG spent more time exploring the novel object than the NG (F_(1,45) = 6.098, p = 0.017; Figure 2C). These results indicated scatter-hoarding rodents owned an elevated spatial memory compared to the non-scatter-hoarding rodents. We found there were no significant sex differences and effects of sex-by-group interactions in both OLT and NORT tests (all p > 0.05).

We found that the NG-SG group (i.e., NG recipients received bacterial suspension from SG donors) larder-hoarded more seeds than the NG-NS group (NG recipients received sterile 0.9% saline) (χ² = 5.418, df = 1, p = 0.018; Figure 3A). Moreover, the scatter-hoarding intensity was statistically higher in NG-SG group (p > 0.05). No significant effects of gender and interaction between sex and group were found in all seed fates (all p > 0.05).

A significant FMT effect on memory was observed in the NORT. Compared to the NG-NS group, the members of the NG-SG group showed a significant preference towards the novel object (F_(1,31) = 4.538, p = 0.041; Figure 3C). No FMT effect was found in the OLT (Figure 3B). The results did correspond to the results in the above experiments before FMT and we only found significant variability in the NORT. No sex difference and sex-by-group interaction were found in the OLT and the NORT (all p > 0.05).

The rarefaction curve analysis showed that the microbial diversity across all samples had reached stable (Supplementary Figure S3), indicating that most fecal microbial diversity was captured in our study. Alpha diversity increased in the NG-SG group compared to the NG-NS group (Chao1 index, p = 0.027; Observed index, p = 0.030; Shannon index, p = 0.002; Simpson index, p = 0.006; Figures 4A–D). Beta diversity analysis based on Bray-Crutis distance indicated that FMT altered the structure of gut microbiota between the NG-SG group and the NG-NS group (PERMANOVA, p = 0.027). The NMDS analysis provided a visual demonstration of the effects of FMT on gut microbiota (Figure 4E), indicating that there were significant differences in the structure of gut microbiota between the NG-SG group and the NG-NS group. Moreover, partial least squares discrimination analysis (PLS-DA) showed that samples were separated with the samples of the NG-SG group occupying the upper left region and the NG-NS group located in the right axis (Figure 4F).

Firmicutes was the most abundant bacterial phylum in all groups (NG-SG group, mean = 58.75%; NG-NS group, mean = 54.41%) (Supplementary Figure S4). Compared to the NG-NS group, ANCOM analysis indicated that Deferribacteres was enriched at the phylum level (W = 10), and Deferribacteres was increased at the class level (W = 9) in the NG-SG group. LEfSE method analysis with LDA score > 2 identified that there were significant differences in the microbial communities among groups (Figure 4G). At the genus level, Lactobacillus and Mucispirillum were enriched in the NG-SG group. At the species level, Lactobacillus reuteri and Mucispirillum schaedleri were increased in the NG-SG group. No species were found enriched in the NG-NS group as compared to the NG-SG group.