We obtained two litters of C57BL/6 inbred mice from a breeding colony at the University of Michigan. The first litter included 2 males and 2 females and the second litter included 4 males and 4 females. The two litters were born to different mothers 8 d apart. At 21 d after birth the mice from the first litter were weaned and separated by sex into two cages so that the 2 males and 2 females were each co-housed. In the second litter, 2 males and 2 females were also separated by sex into two cages. The remaining 4 mice from the second litter were also weaned and placed into four separate cages at 21 d after birth. At weaning weights and fresh feces were collected from each animal daily; this continued for the rest of the experiment. Immediately after weaning, all animals experienced rapid weight gain until approximately 25 d post weaning (dpw) when their growth rate slowed but continued to increase for the remainder of the experiment (). During the 150 dpw, the weight of the average female increased by 1.7-fold and the average male increased 2.3-fold. Among the three animals that were sampled until 364 dpw, the female gained an additional 20% and the two males gained 23%. Among the mice in the second litter, the median ratio of pairwise differences in weight between co-housed and single-housed animals was 0.86 for the females and 1.00 for the males. Co-housed males gained 1.9 g more than their single-housed littermates; however the co-housed females gained 1.9 g less than their single-housed littermates. Although co-housed mice from the first litter gained more weight and had a greater difference in weight than the co-housed mice from the second litter, Spearman correlation coefficients between weights of animals from the same litter were within the range of coefficients calculated between animals of different litters. Taken together, these results indicate that after controlling for age, genetics, diet, environment, sex, litter, and housing there was still a significant amount of variation in the weights of the mice from these litters.
Figure 1. Weights of mice over the first 364 d post weaning (dpw). For each litter two male mice were co-housed and two female mice were co-housed. In addition, two males and two females from the second litter were each individually housed. The (more ...)
To determine whether there was a change in the gut microbiome in the immature and mature animals, we used 454 FLX Titanium to sequence the V35 region of the bacterial 16S rRNA gene. We focused our microbial community analysis on the feces sampled at 0–9 dpw (“early”) and 141–150 dpw (“late”). Sequences were assigned to operational taxonomic units (OTUs) where the average distance between sequences was not greater than 0.03 and used to recreate their phylogeny. There were no statistically significant effects of sampling period (early vs. late) or sex on α diversity measures including the observed community richness (Sobs = 98.3 OTUs), inverse Simpson diversity index (1/D = 16.4 OTUs), Shannon diversity index (H’ = 3.26), and the phylogenetic diversity (PD = 5.54). We characterized the structure of the microbiome by calculating the pairwise distance between community structures using θYC (). Non-metric dimensional scaling plots demonstrated that the community structures of all 12 mice could be partitioned between the early and late time periods (AMOVA; all p < 10−4) and that there was not a significant effect of litter, single vs. co-housing, or sex on community structure. Similar results were observed when we used the weighted or unweighted UniFrac metrics (data not shown).
Figure 2. Temporal variation of the community structure of the murine gut microbiome for all 12 mice together as well as those from two female and one male animal over the first 364 dpw as represented in NMDS plots using the ΘYC distance (more ...)
Observing that the overall community had a significantly different structure during the late period compared with the early period, we attempted to identify the bacterial populations responsible for the shift. At the phylum level, there were no significant differences between the early and late periods (p > 0.05). Among the early and late samples, 70.54% of the sequences affiliated with members of the Bacteroidetes and 29.21% affiliated with the Firmicutes. Minor members of the community affiliated with the Actinobacteria (0.09%), Proteobacteria (0.02%), Tenericutes (0.10%), TM7 (0.02%), and the Verrucomicrobia (0.02%). To obtain a higher resolution characterization of the community, we identified those OTUs that had an average relative abundance greater than 0.5% across all samples (i.e., minimum average of 9.5 sequences observed per sample; n = 32 OTUs). We used a repeated measures paired group analysis of variance to identify those OTUs that were differentially represented between the early and late periods. We identified 11 OTUs whose relative abundance increased and 16 whose relative abundance decreased between the early and late periods (). Interestingly, some closely related OTUs had different profiles in their relative abundance during the early and late periods. For example, two members of the genus Lactobacillus (OTUs 12 and 23) decreased in abundance and one (OTU 9) increased in abundance. Similar results were observed for members of the family Porphyromonadaceae, which showed significant decreases (OTUs 5, 7 and 8) and increases (OTUs 6 and 13) in relative abundance between the early and late periods; three OTUs affiliated with the Porphyromonadaceae did not exhibit a significant difference in relative abundance between the early and late periods (OTUs 1, 4 and 14).
Table 1. The difference in average relative abundance of OTUs between the early and late periods
Considering the clear shift in community structure between the early and late periods, we were curious whether the bacterial populations we observed early were the same as those found late or whether they were the same populations but in different abundance. Traditional metrics of membership (e.g., Jaccard β-diversity distance) are adversely affected by under sampling because rare populations are unevenly represented. To circumvent this difficulty, we assumed that the ten days within each period represented the core-microbiome for the entire period and for that mouse and calculated the fraction of sequences from the early and late periods that affiliated with a shared OTU for each mouse. Across the 12 mice, between 94.1 and 98.5% of the sequences from the early or late periods belonged to OTUs that were found in both periods and there was not a significant difference between the early and late periods (paired T-test; p = 0.67). When we looked for OTUs with an average relative abundance greater than 0.2% that were unique to either periods, we were unable to detect any that were unique to a period across all of the animals. These results suggest that although the membership of the early and late periods was largely the same, it was changes in the relative abundance of these populations that explained the observed differences in community structure.
To measure the difference in stability among the early and late communities, we measured the community-level variation within each period for each animal by calculating the mean squared distance between all points within a period for each animal. Across the 12 animals the mean squared distances for each animal decreased by between 1.5 and 7.6-fold (p = 0.0001). To investigate this reduced variation further, we measured the association between varying sized time intervals for each mouse in the early and late periods (). The late samples had a lower level of variation between time points within the same animal and between different animals compared with the early samples. In addition, we observed that as the time interval between samples increased, the dissimilarity between samples increased for the early time points, while it remained constant for the late time points. These data suggest that the intra-animal variation was random within the early and late time periods. Further support for random temporal variation within a period was provided by non-significant co-occurence statistics, which indicated that the microbial populations were not associating in a preferential manner (C-score, Checkerboard, and V-ratio, all p > 0.05). We calculated the coefficient of variation (i.e., the ratio of the standard deviation to the mean relative abundance) within a period for each OTU and animal to identify those OTUs that exhibited differential stability between the two periods (). Six OTUs had a greater than 2-fold reduction in their coefficient of variation including three members of the order Bacteroidales (OTUs 10, 11 and 27), one member of the genus Turicibacter (OTU 20), one member of the genus Barnesiella (OTU 15), and one member of the family Porphyromonadaceae (OTU 13); these were also the 6 OTUs that had the most dramatic increases in relative abundance (). None of the OTUs had a significant increase in its coefficient of variation between the early and late periods.
Figure 3. Average ΘYC distances between days for the same animal (self-comparison) and between different animals (other-comparison) for 0 to 9 dpw (blue) and 141 and 151 (red) dpw. Error bars represent the 95% confidence interval.
Table 2. The difference between the coefficient of variation for the early and late periods for each OTU
We performed additional sequencing using samples collected between 11 and 25 dpw and at 45, 65, and 125 dpw to better define the shift between the early and late periods. We calculated the root mean squared θYC distance (RMSD) between all of the samples for each animal to their early and late points for each mouse to determine when the community structure changed. The difference between the RMSD for each sample to the early and late time points indicated the affiliation of each sample to the early or late period. These data indicate that the communities all shifted to the late community structure by 11 to 17 dpw (). We also obtained sequence data at 364 dpw for three of the mice to quantify the long-term stability I the community. The community structures observed at 364 dpw clearly clustered among the late samples indicating that the late community structure exhibited constancy over long periods of time (Figs. 2 and 4).
Figure 4. Relative difference in root mean squared θYC distances to the early (0–9 dpw) and late (141–150 dpw) samples for each animal. Rectangles with “NC” indicate that the samples were not characterized (more ...)
Prior to weaning the mice consumed solid chow and milk from their mother, which contained nutrients and high levels of IgA. We reasoned that although the behavioral and dietary effects of weaning would affect the microbiome, the dynamics of IgA in the weaned animals could explain a portion of the changes we observed in the microbiome. We quantified the amount of secretory IgA (sIgA) as a fraction of total protein using the same feces that were characterized by 16S rRNA gene sequencing. Interestingly, the sIgA levels increased an average of 25-fold between the early and late periods and were elevated by 6–11 dpw (). Although there was only a qualitative association between shifts in the community structure and sIgA levels, it appeared that the shift in sIgA preceded the shift in the community structure providing evidence to support the hypothesis that maturation of the immune system affected the observed changes in the microbiome.
Figure 5. Relative percentage of secretory IgA as total protein in the feces of each animal over the first 150 dpw. Rectangles with “NC” indicate that the samples were not characterized and the sample with “ND” (more ...)