Soil is a key factor influencing gut microbiota and its effect is comparable to that exerted by diet for mice

Soil-associated changes in gut microbiota at phylum level

We analyzed the bacterial community composition using MiSeq sequencing targeted to the 16S rRNA gene in up to 180 fecal samples (Supplementary Table 2). For each sample, we PCR-amplified the V4 hypervariable region using the 515F-907R primer set²⁹. The sequencing data set consisted of 10,937,871 high-quality, classifiable 16S rRNA gene sequences, with at least 45,816 sequences per sample (Supplementary Table 2).

On the basis of the obtained sequences, we discovered that sterile and unsterile soil significantly altered gut bacterial community composition (Figure 1 and Supplementary Table 3) in a diet-dependent manner. Mice fed a normal diet and living on sterile soil after weaning exhibited significantly more Bacteroidetes (P < 0.001) and fewer Firmicutes (P < 0.001) (Figure 1A and Supplementary Table 3a). Unsterile soil added before birth also significantly increased the number of Bacteroidetes (P = 0.032). Instead, mice fed a high-fat diet and raised on bedding with unsterile soil added before birth or after weaning showed significantly more Actinobacteria (P < 0.001) (Figure 1B and Supplementary Table 3b).

Figure 1. Changes in the relative abundance of Actinobacteria, Bacteroidetes, and Firmicutes in fecal samples from mice treated with different living beddings and fed a normal or high-fat diet.

Values from all available fecal samples were averaged (n = 8, 9, or 10 per treatment). (A,B) Relative abundance of bacteria on the eighth week in fecal samples of mice raised on specific-pathogen-free grade (SPF) sterile bedding (control), sterile bedding with sterile soil added after weaning (SS), sterile bedding with dirty unsterile soil added after weaning (SD), and sterile bedding with unsterile soil added prior to birth (DE) fed (A) a normal diet (ND) or (B) a high-fat (HF) diet. (C–F) Relative bacterial abundance in samples collected on the fourth (W4), fifth (W5), seventh (W7), and eighth (W8) week from mice of the (C) SPF-ND group, (D) SPF-HF group, (E) SD-ND group, and (F) SD-HF group.

Microbes in the environment were reported to affect the colonization of the intestinal microflora in newborns³⁰. Here, we found that microbes and soil in the bedding could change the composition of gut microbiota after mice were weaned. When mice were fed a normal diet and lived on bedding with unsterile soil, they consistently showed more Bacteroidetes (P < 0.001) and fewer Firmicutes (P < 0.001) (Figure 1E and Supplementary Table 4a) during the first 3 weeks. After the third week, the abundance of these two phyla returned to the level from the previous week. When fed a high-fat diet and provided with bedding containing unsterile soil, mice showed an increased abundance of Actinobacteria on the third week of treatment (P = 0.030), becoming highest on the fourth week (P < 0.001) (Figure 1F and Supplementary Table 4b). In summary, we observed that the gut microbiome could respond rapidly to alterations in the living environment and diet, potentially facilitating adaptation to a variety of lifestyles.

Soil-associated changes in gut microbial diversity

We further discovered that unsterile soil could increase microbial diversity, particularly for mice on a normal diet, whether it was added before birth or after weaning (Figure 2, Supplementary Figure 3, and Supplementary Table 5, Supplementary Table 6). By comparison, microbial diversity was unaffected by sterile soil in mice fed a normal diet and decreased in those on a high-fat diet (Supplementary Figure 3 and Supplementary Table 5). Earlier, we found that merely adding environmental microbes to the bedding could increase diversity but had little effect on the composition of dominant bacteria in mice gut microbiota²⁸. Hence, microbes appear to contribute mainly to microbial diversity, whereas soil may alter microbial community structure.

Figure 2. Representation of α diversity in gut microbiota based on fecal samples collected from specific-pathogen-free (SPF) (blue curve) and sterile bedding with dirty unsterile soil added after weaning (SD) (red curve) groups over the fourth to the eighth week of age.

(A–C) SPF and SD groups fed a normal diet: (A) operational taxonomic units (OTUs), (B) Chao 1 estimates, and (C) Shannon index. (D–F) SPF and SD groups fed a high-fat diet: (D) OTUs, (E) Chao 1 estimates, and (F) Shannon index. Values from all available fecal samples were averaged (n = 8–10; *P < 0.05, **P < 0.01, based on a two-tailed least significant difference test).

Soil-associated changes in gut microbiota differ between male and female mice

An epidemiological survey has shown that immune system diseases, such as asthma and wheeze, display a skewed sex bias towards males³¹. Here, we found that abundance of the phyla Bacteroidetes and Firmicutes was related to gender, except in mice with sterile bedding with unsterile soil added prior to birth (DE group) exposed to unsterile soil bedding from birth (Supplementary Figure 4, Supplementary Figure 5 and Supplementary Table 7). Bacteroidetes were more abundant in males than in females of the mice raised on specific-pathogen-free bedding (SPF group) fed a normal diet (P = 0.019). By contrast, Firmicutes were more abundant in males than in females in the mice raised on sterile grade murine bedding with dirty unsterile soil (6:11 (w/w)) added after weaning (SD group) on a high-fat diet (P = 0.042) (Supplementary Figure 5 and Supplementary Table 7).

Soil-associated changes in gut microbiota at operational taxonomic unit (OTU) level

Random Forests is a supervised machine-learning technique that can classify a sample by estimating the importance of OTUs at species level according to their relative abundance and sample probability³². We found that distinct community signatures existed between any two treatments (Supplementary Table 8–Supplementary Table 10). For mice on a high-fat diet, there were 79 predictive species-level OTUs between the (SPF) and DE groups (baseline error = 0.44, cross-validation error = 0 ± 0), of which 63 were overrepresented in the DE group. Moreover, out of these 63, 25 and 20 were assigned to the classes Actinobacteria and Bacilli, respectively (Supplementary Table 8b). A comparison between SPF and SD groups gave similar results (Supplementary Table 8a); however, the decrease in beneficial bacteria was greater in the mice raised on sterile grade murine bedding with sterile soil (6:11 (w/w)) added after weaning (SS group) than in the SPF group (Supplementary Table 8c).

We further analyzed differences in the composition of gut bacterial communities between any two groups of mice on a normal diet (Supplementary Table 9). Compared to the SPF group, DE and SD groups presented more Actinobacteria and Bacilli, whereas the SS group showed more Bacteroidia (Supplementary Table 9). Notably, bacteria of the Bacteroidales S24–7 family were more prevalent among SPF mice fed a high-fat diet than in those fed a normal diet. Instead, the latter had more bacteria of the families Lachnospiraceae and Ruminococcaceae (Supplementary Table 10a). The DE group fed a high-fat diet showed more Actinobacteria and Bacilli than animals fed a normal diet. On the contrary, the latter had more Bacteroidales S24–7 and Lachnospiraceae (Supplementary Table 10c). Thus, unsterile/sterile soil bedding affected the structure of mice gut microbial communities in a diet-dependent manner. The effect varied depending on whether i) mice were raised on bedding with soil added before birth or after weaning, ii) the bedding contained sterile or unsterile soil, and iii) mice were fed a normal or high-fat diet.

Gut microbial community profiles identify soil as a key determinant of changes to the mouse gut microbiome

We used UniFrac distances to analyze the 16S rRNA datasets and measure similarities among microbial communities. The distance between mice exposed to two different living environments and fed a high-fat diet was similar to that between mice using the same bedding, and fed either a normal or a high-fat diet (Figure 3 and Supplementary Table 11). The distance between SPF groups fed a normal or a high-fat diet did not differ significantly from that between SPF and SS, and between SD and DE mice fed a high-fat diet. Indeed, these groups were among the least distant. Distances between mice maintained on different beddings and a high-fat diet were greater than those between SPF groups fed different diets. Thus, soil appears to influence the composition of the gut microbial community to the same extent as diet. In addition, a change in living environment also had an important effect on gut microbiota. When fed a normal diet, microbiota in the gut of SPF and DE mice were more similar than those of other groups (Figure 3 and Table 11).

Figure 3. Differences in gut microbial communities between treatments.

For mice fed a normal diet, the shortest UniFrac distance was between specific-pathogen-free (SPF) and sterile bedding with unsterile soil added prior to birth (DE) groups, and the longest one was between sterile bedding with dirty unsterile soil added after weaning (SD) and sterile bedding with sterile soil added after weaning (SS) groups; comparisons between SPF and SD, and between DE and SS showed no significant differences. For mice fed a high-fat diet, the shortest distances were SPF–SS and SD–DE, with no significant difference between them; the longest distances were SD–SS and DE–SS; SPF–SD was significantly longer than SPF-DE (n = 8–10; *P < 0.05, **P < 0.01, based on a two-tailed least significant difference test).

Unsupervised clustering using PCoA of UniFrac distance matrices confirmed that living environments and diets explained the variation in the data set (Figure 4 and Supplementary Figure 6, Supplementary Figure 7). When fed the same diet, the gut microbial communities of mice from one living environment clustered apart from those of mice subjected to other treatments (Figure 4A, B). For two different living environments, the two high-fat diet groups showed a similar transfer distance as mice maintained on the same bedding but fed two different diets (Figure 4C, D and Supplementary Figure 6). For the SPF group fed a normal diet, almost no differences were observed among fecal samples collected at four different time points (Supplementary Figure 7A). These results confirm that the living environment exerts a great influence on the composition of gut microbiota. It should be noted that we did not find significant clustering with respect to cage or gender.

Figure 4. Principal coordinates analysis (PCoA) of unweighted UniFrac distances for fecal microbiota sampled on the eighth week from mice on four types of bedding.

Analysis was based on the Illumina bacterial 16S rRNA gene data set (V4 region). Mice from the four groups were fed a (A) normal or (B) high-fat diet. (C) Mice in the sterile bedding with dirty unsterile soil added after weaning (SD) and specific-pathogen-free (SPF) groups were fed a normal or high-fat diet. (D) Mice of the sterile bedding with sterile soil added after weaning (SS) and sterile bedding with unsterile soil added prior to birth (DE) groups were fed a normal or high-fat diet.

Animals and groups

We purchased 6-week-old male (n=1) and female (n=3) C57 mice from B & K Universal (Shanghai, China) for breeding. One male was mated with three females. After the females became pregnant, two of them were shifted to a separate cage. The newborn mice were weaned and when they were 4 weeks old they were moved to separate cages.

We selected 8 males and 24 females from among the weaned mice to carry out further experiments. Next, two males and six females were selected from different cages and transferred to room number 2. The cages were furnished with sterile grade murine bedding and unsterile soil (6:11 w/w bedding to soil). The remaining 6 males and 18 females were raised in room number 1, on sterile grade bedding. The bedding was changed once a week. The soil used in the experiments was from the top 10 cm of a farm ground with abundant goats, hens, and ducks.

When mice were 6 weeks old, they were bred with 1 male per 3 females. In room number 1, 30 males and 30 females, aged 23–37 days (weight, 8.6–21.4 g) the fourth week group, W4), were selected for experiments. They were randomly separated into three groups: the SPF group, raised on specific-pathogen-free bedding; the SD group, raised on sterile bedding with unsterile soil (6:11 (w/w)) added after weaning; and the SS group, raised on sterile bedding with sterile soil (6:11 (w/w)) after weaning. The SPF group served as the control. The randomization procedure was as follows. Considering gender matching, more than 80 mice were used in this experiment, and it was very difficult to breed so many mice in the same time. When we had about 20 mice with an age difference within about 10 days, we divided them into two groups, one fed with normal diet and the other with a high-fat diet. These 20 mice had been treated with same bedding. The other groups of 20 mice were treated with the second or third of type bedding. The breeding mice were uniformly distributed and randomly grouped. The SD group was transferred to room number 2 after weaning. Next, 10 males and 10 females, who were born in room number 2, aged 23–33 days (W4) were selected for experiments and raised on the same kind of bedding as before weaning (group DE). For all experiments, two or three mice were kept in a cage and the bedding was changed once a week.

For each group, 10 mice (five males and five females) were fed the same sterile grade commercial normal pellet diet as before weaning, while 10 mice (five males and five females) were fed a sterile grade commercial high-fat pellet diet after weaning (Supplementary Table 1). All animals were fed the same sterile distilled water. They were allowed free access to water and food. All animals were housed in two specific-pathogen-free animal rooms with a 12-h light/12-h dark cycle at 24°C ± 2°C and humidity of 40% ± 5%.

For SPF and SD groups, the first fecal samples were collected before mice were shifted to separate cages, and then on the fifth, seventh, and eighth week (Supplementary Figure 2). For DE and SS groups, fecal samples were collected on the eighth week (Supplementary Figure 2). After collection, fecal samples were put immediately into an ice box and stored at −80°C within 8 h.

Care and use of animals

Animal experiments were performed in strict accordance with the guidelines of the Animal Research Ethics Board of Southeast University. All experiments were approved by the Animal Care Research Advisory Committee of Southeast University and the National Institute of Biological Sciences (approval number: 2014063009). All efforts were made to alleviate the suffering of animals.

Specifically, mouse health was monitored every other day and body weight measurements were performed every week. Mouse health was assessed by observing changes in body weight or fecal shape, and external physical appearance. The animals would be euthanized to minimize pain and distress if they became severely ill over the experimental period, as when they showed one or more symptoms, such as 15–20% weight loss, diarrhea, loss of hair quality, pain (arched back and curled up posture), or listlessness for more than 1 week.

Euthanasia was performed as follows. A 1% (w/v) dose of sodium pentobarbital (50 mg/kg) was administered through an intraperitoneal injection using a hypodermic needle. When mice lost consciousness, they were killed by neck dislocation and death was confirmed. Two individuals were required to perform injections, one to hold the animal and the other to perform the procedure. Animals were not left unattended during the procedure. The ARRIVE reporting guidelines were followed during this study (Supplementary File 1)⁴⁴.

DNA extraction

Genomic DNA was extracted from fecal samples according to the protocol proposed by Zoetendal et al.⁴⁵. Each sample (40–60 mg) was weighed and put into 15 ml 1 × PBS (pH 7.4). Samples were re-suspended completely by vigorous shaking and subsequently centrifuged at 700 × g at 4°C for 5 min. The supernatants were transferred to a 15-ml tube and centrifuged at 9000 × g at 4°C for 5 min. The pellets were transferred to a 2-ml Eppendorf tube and re-suspended in 300 μl 10× TE buffer (10 mM Tris-HCl pH 8.0, and 1 mM EDTA pH 8.0). Lysozyme (100 μl, 200 mg/ml) was added to each tube, and incubated for 1 h at 37°C. Subsequently, 50 μl 10% SDS (w/v) and 20 μl proteinase K (20 mg/ml) were added to the samples and incubated for 2 h at 50°C. Next, 100 μl 5 M NaCl was added, followed by incubation at 65°C for 10 min. The samples were mixed gently with an equal volume of Tris-Phenol (pH 8.0) for 5 min. The mixtures were centrifuged at 9000 × g for 5 min. The upper layer was transferred to a new tube. This step was repeated twice, after which chloroform was used instead of Tris-Phenol to clean the samples one more time. Finally, a 1/10 volume of 3 M sodium acetate (pH 5.2) and two volumes of 95% ethanol were added. Samples were mixed gently and centrifuged at 4°C and 9000 × g for 20 min. The precipitated genomic DNA was washed twice in 200 μl 70% ethanol. The dried DNA samples were re-suspended in 50 μl sterile double-distilled water. DNA quality was monitored by gel electrophoresis and spectrophotometry (260/280 nm). The samples were stored at −20°C.

MiSeq sequencing and data handling

The analysis of gut microbial communities was performed using MiSeq high-throughput sequencing of bacterial 16S ribosomal RNA (rRNA) genes. For each sample, the V4 hypervariable region was amplified using the 515F and 907R primer set, as described by Angenent et al.²⁹. PCR was performed with 300 ng microbial community DNA as a template, using 1 μl of Trans Start Fast Pfu Taq DNA Polymerase (TransGen Biotech, Beijing, China), 0.25 mM of each dNTP, 1× Trans Start Fast Pfu buffer, 0.2 μM forward primer, and 0.2 μM reverse primer, in a total reaction volume of 50 μl. The cycling conditions were: 94°C for 3 min followed by 27 cycles of 95°C for 45 s, 50°C for 45 s, and 72°C for 45 s, with a final extension at 72°C for 5 min. Each sample was amplified in duplicate, combined, purified, and then quantified using the QuantiFluor™-ST Handheld Fluorometer with UV/Blue Channels (Promega Corporation, Fitchburg, WI, USA). Sequencing of the PCR product libraries was performed on an Illumina MiSeq platform (Illumina Inc., San Diego, CA, USA) using 2 × 250 bp chemistry. We merged the raw paired-end reads using FLASh software version 1.2.7, set to a minimum overlap of 10 bp. Other parameters were set to default settings. PCR artifacts were removed by eliminating low-quality sequences using Trimmomatic version 0.30⁴⁶. Chimeric sequences were removed using Usearch version7.1⁴⁷. We clustered the quality-checked sequences into de novo OTUs at a 97% similarity threshold using the QIIME software version 1.8.0 software package⁴⁸. The generated OTUs were classified using the RDP classifier software version 11.5 at a 70% confidence threshold for sequences longer than 200 bp⁴⁹.

Data analysis and statistical tests

All samples were rarefied down to 45,816 sequences per sample to prevent bias due to sampling depth. We used QIIME to calculate α diversity indices of Chao1 estimator and Shannon diversity/richness for each sample⁴⁸. We calculated the metrics of unweighted UniFrac distances between any two treatments and performed principal coordinates analysis (PCoA) through QIIME to examine dissimilarities in community composition. PCoA was used to compare groups of samples based on unweighted UniFrac distance metrics by plotting n samples in (n − 1)-dimensional space.

Random Forests analysis in R version 3.2.3⁷ was performed with 500 trees and all default settings using 16S rRNA-based OTUs from the Illumina V4 data sets as described previously⁵⁰. We used out-of-box (OOB) error to estimate the generalization error for all 16S rRNA comparisons involving all treatments. For each comparison, 100 relevant subsets of samples were extracted from the table of OTUs, and the average OOB error estimates and OTU importance estimates were calculated from subset samples. For a direct evaluation of the predictive strength of the OTUs, we compared generalization errors at various sequencing depths: the lowest observed depth of 45,816 sequences and at sequencing depths of 100, 1000, 10,000, and 40,000 reads per sample. The mean and standard deviation of the OOB error were estimated for each classification task using 100 independent rarefactions of the data. The expected ‘baseline’ error was obtained by a classifier that simply predicted the most common class label.

T- test and analysis of variance with post hoc Tukey’s test, for comparison of two or more than two groups, respectively, were performed using SPSS software, version 18.0 (SPSS, Inc., Chicago, IL, USA).