Fig. 3.

Animal-to-animal taxonomic and network differences. a Principal component analysis of the OTU distributions across the 16 animals. The first two principal components (PCs) are shown, comprising 92% of the total variance. FORG animals are shown in green and CONC in red. Visually, it seems clear that the diet difference explains most of the variation in OTU distributions. An animated 3-dimensional version of this plot that includes PC#3 (3% of variance) is presented as Additional file 2. Each point is labeled with the breed of the animal in question: Su: Suffolk, Ra: Rambouillet, Ha: Hampshire. b Principal component analysis of the distribution of reads mapped to metabolic network nodes. The first two principal components (PCs) are shown, comprising 98% of the total variance. However, diet is no longer the main source of variation. Instead, principal component 1 separates three CONC animals (numbers 1220, 1239 and 1348; high RFI) from the rest of the dataset. Inspection of the node-level data suggests that these three animals are unusual in that they have higher than usual node-to-node variation in the number of mapped reads (namely a few nodes with a large number of mapped reads) and are also highly correlated with each other, unlike some of the other CONC animals with rather different profiles. An animated 3-dimensional version of this plot that includes PC#3 (1% of variance) is presented as Additional file 3. c Minimum and maximum pairwise node distances seen when reads were randomly and proportionally reassigned to each animal. On the y-axis is the same distance scale as y in panel D, on the x-axis is the proportion of simulations with a given minimum/maximum (Materials and Methods). The color scheme is as for (d) Dashed lines give the minimums and maximums seen in the real data of d. d Pairwise differences in distribution of reads mapped to OTUs (x-axis) and nodes (y-axis). FORG to FORG comparisons are shown in green, CONC to CONC in red, and FORG to CONC in blue. For each animal, a vector representing all mapped reads was normalized to unit length and then standard Euclidian distances computed between it and all other animals (Materials and Methods). For the FORG to FORG and CONC to CONC pairs, we computed the Pearson’s correlation of OTU and node distance and compared that value to that seen from randomized datasets (Materials and Methods). e As for c, except with the OTU distances. The x-axis gives OTU distances and the y-axis simulation frequencies