In total we sequenced 621 fecal samples of 31 zoo-housed carnivore and herbivore species. After quality filtering and read merging, the dataset consists of 29,777,361 sequences with an average of 47,796 sequences per sample. Following the DADA2 pipeline in QIIME 2, we identified 21,058 different features across all samples (2,315 to 134,414 features per sample). The most common feature is represented 686,398 times in 292 samples.
Composition of fecal microbiota of major mammalian groups
We found significant differences between herbivores and carnivores in the microbial composition (ANOSIM statistic: R = 0.50, p < 0.001, number of permutations: 999, distance="bray") as shown in Fig. 1B. The four major bacterial families across all herbivore species are Spirochaetaceae, Lachnospiraceae, Rikenellaceae and Oscillospiraceae. Spirochaetaceae represent on average 15.9% of the microbiota of herbivore species and within these, they are twice as common in Perissodactyla than in ruminants. While this family is equally distributed across perissodactylan species, within the ruminants it only occurs in larger proportions in giraffes and okapis. In contrast, we found on average 20.2% of Lachnospiraceae in Perissodactyla and only 11.2% in ruminants, where larger proportions were observed in reindeer. Rikenellaceae, the third most-common family in herbivorous species, constitutes on average to 16.1% of the fecal microbiota of ruminants and to 12.6% that of Perissodactyla. With respect to the Oscillospiraceae, we found notable differences between both herbivore groups. While this family is equally spread across nearly all ruminants, it only appears in tapirs and black rhinoceros in greater proportions of all Perissodactyla. Besides those four major families, we identified Bacteroidaceae in many ruminants and an uncultured bacterium p-251-o5 of the Bacteroidales order in Perissodactyla, especially in the grevy’s zebras. Other bacterial families such as Tanerellaceae, Erysipelotrichaceae, Clostridiaceae, Fusobacteriaceae and Enterobacteriaceae constitute on average less than 5% to the microbiota across all herbivore species.
The most dominant bacterial family in carnivore species is Fusobacteriaceae, occurring on average in 23.2% of all Felidae and in 23.3% of all Canidae. However, within the Canidae, this family is low-abundant in red pandas and brown bears as it constitutes to less than 5% of both fecal microbiotas. The distribution of Clostridiaceae, the second dominant family in carnivores, is on average similar for Felidae and Canidae. Clostridiaceae form a large proportion of the microbiota, accounting for more than 30%, in both bears and red pandas. Those species also differ from other Canidae with regard to Bacteroidaceae. Whereas this family is frequently found in most carnivores, it is low-abundant (< 2%) in the red pandas, brown bears, polar bears and fossas. Additionally, we found on average 16.0% Peptostreptococcaceae in Felidae and only 9.9% of this family in Canidae, but the value calculated for Felidae is mostly influenced by its high abundance of 33.0% in fossas. Beside these major bacterial families in carnivores, some others are largely represented in both bear species and red pandas. For example, we found that Enterobacteriaceae contribute on average 25.3% to the fecal microbial composition in red pandas, to 22.7% in polar bears and to 20.4% in brown bears. Furthermore, Erysipelotrichaceae are more dominant in brown bears and red pandas than in other Canidae. With regard to the Felidae, Lachnospiraceae are another dominant family being equally distributed across all sampled felid species. Other bacterial families such as Spirochaetaceae, Lachnospiraceae, Rikenellaceae and Oscillospiraceae, which were dominant in herbivorous species, accounted for less than 5% of the carnivore microbiota.
Microbial diversity within and between herbivores and carnivores
The microbial diversity measured by effective number of species differs highly significantly between carnivores and herbivores (Fig. 1C) (Kruskal-Wallis: p < 0.001, df = 3, Dunn Test with Bonferroni correction p < 0.001). Also, these differences are highly significant regarding the Shannon index and species richness (Fig. 2). Consequently, carnivorous species show a strongly reduced microbial diversity compared to herbivorous species. In contrast, alpha diversity within the two groups is quite similar across all measurements. For example, within the Caniformia the mean ENS per species varies between 51 in polar bears and 229.2 in arctic foxes, which is comparable to the variation within the Feliformia (94.5–217.6). In contrast ENS values in herbivores are much higher, ranging between 679 in black rhinoceros and 2,869.9 in the plains zebras and between 1,036.2 in common elands and 2107.6 in bongos for ruminants respectively.
Regarding the beta diversity, the principal coordinate analysis (PCoA) of the weighted UniFrac distance matrix explains a total of 63.3% of data variability within the first three main axes. The permutation test for homogeneity of multivariate dispersions shows homogeneity within the four animal groups (F = 0.670, p = 0.570, permutations = 999) as well as within carnivores and herbivores respectively (F = 1.345, p = 0.250, permutations = 999). The following ADONIS test for differences between each group was significant for the four species groups (R²=0.020, p < 0.001, permutations = 999) but not for the diet type (R²=0.004, p = 0.065, permutations = 999), which illustrates the differences between carnivore and herbivore species. This is also confirmed by the PCoA of the unweighted UniFrac measurement (Fig. 3). Similar to the weighted UniFrac, the homogeneity of dispersion is given for both, the animal groups (F = 0.670, p = 0.570, permutations = 999) and diet types (F = 1.345, p = 0.260, permutations = 999). For this metric, we found significant differences between diet types (ADONIS: R²=0.004, p < 0.05, permutations = 999) and animal groups (R²=0.020, p < 0.001, permutations = 999). Even if slightly less data variability (46.3%) is explained by this distance matrix, it shows a recognizable pattern within the Carnivora. At a group-specific level, the Carnivora are divided into three clusters (Fig. 3B). The first cluster, closest to the Perissodactyla, consists of the polar and brown bear as well as the red panda samples. A little distant from these lies the center of the second cluster, made of the big and small cats as well as the South American Cerdocyonina represented by the bush dog and maned wolf samples. Finally, the third cluster, which is most distant from the herbivorous species is composed of the Vulpini group (fennec fox, arctic fox, bat-eared fox) and the African wild dog samples. Overall, it can be observed that the samples of the Perissodactyla and Ruminantia are more similar to each other than those of the carnivorous species.
For a more detailed analysis of this high variation within the Carnivora, which is in contrast to the herbivore species, we compared one representative species for each of the four groups with respect to the taxonomic assignment. Figure 4 and Fig. 5 show all samples of lions, brown bears, wildebeests and plains zebras. The results clearly show that the variability within the zebra and wildebeest samples is lower compared to the lion and brown bear samples. More detailed, we identified 19 microbial families that consist to more than 5% of the microbiota within the brown bear samples across four zoos. Those families vary largely in their occurrence, which is measured by the coefficient of variation (CV). Thus, families occurring in at least five samples which show the greatest fluctuations are Staphylococcaceae (2.0% ± 6.6%, CV = 3.2) which are represented in seven brown bear samples across half of the sampled zoos. In addition to this family, Moraxellaceae (3.1% ± 7.9%, CV = 2.5), Streptococcaceae (3.0% ± 6.7%, CV = 2.2) and Lachnospiraceae (2.7% ± 5.8%, CV = 2.1) also show large deviations within the bear samples. Bacterial families which make up greater proportion of the microbiota and are more common across all samples appear to be more stable. Some of those are Erysipelotrichaceae (10.2% ± 9.8%, CV = 1.0), Peptostreptococcaceae (15.9% ± 11.2%, CV = 0.7) and Clostridiaceae (25.0% ± 14.3%, CV = 0.6). The lion samples show a similar pattern, for which we identified 21 bacterial families (> 5%). Again, Peptostreptococcaceae (11.0% ± 9.6%, CV = 0.9) as a common family in all samples, as well as Lachnospiraceae (7.5% ± 6.2%, CV = 0.8) and Fusobacteriaceae (17.8% ± 12.1%, CV = 0.7) are among the most stable bacterial families in the lion microbiota. In contrast Coriobacteriaceae (1.8% ± 3.5%, CV = 1.9), Erysipelotrichaceae (5.0% ± 7.8%, CV = 1.6) and Enterobacteriaceae (5.7% ± 10.7%, CV = 1.9) are largely responsible for the high variability within the samples.
In contrast, the zebra samples show less variance in their taxonomic composition. Here, the greatest variability occurs among the Bacillaceae (2.0% ± 4.9%, CV = 2.4), Oscillospiraceae (1.8% ± 3.0%, CV = 1.7) and Fibrobacteraceae (1.6% ± 3.1%, CV = 1.9). In contrast, the most equally spread bacterial families are Rikenellaceae (7.2% ± 3.7%, CV = 0.5), Spirochaetaceae (15.6% ± 6.8%, CV = 0.4) and Lachnospiraceae (11.3% ± 3.6%, CV = 0.3). Similar to the plains zebra, the wildebeest samples also reveal less variability compared to carnivores. Again, the greatest variation is found in least abundant bacterial families. Within those are e.g. Micrococcaceae (3.0% ± 5.4%, CV = 1.8), Clostridiaceae (1.2% ± 2.3%, CV = 1.9) or Moraxellaceae (1.8% ± 3.7%, CV = 2.0). Compared to that, lesser variation between samples is found in families that are high-abundant as Oscillospiraceae (8.1% ± 2.9%, CV = 0.4), Lachnospiraceae (5.6% ± 2.3%, CV = 0.4) or Prevotellaceae (10.8% ± 5.5%, CV = 0.5). Further results are shown in detail in an additional file (see Additional file 2).
To further investigate the variability of the most abundant bacterial families in herbivores and carnivores, Fig. 6 shows the CV plotted against the number of samples and against the total percentage of occurrence. On the one hand, this underlines the fact that the CV of the most abundant bacterial families in herbivorous animals is in general lower than that of carnivorous animals. On the other hand, low- abundant bacterial families that are present in almost all fecal samples of an animal group, show greater deviations between samples (e.g. Enterobacteriaceae). In contrast, the CV for high-abundant families (e.g. Clostridiaceae and Fusobacteriaceae) is much lower. Considering the number of samples analyzed, it is noticeable that the CV does not necessarily decrease with regard to a larger number of samples being analyzed. To examine whether this effect is possibly due to species-specific differences, we created randomized subsets of bacterial families that occur in more than 7% of all herbivore or carnivore species, because low-abundant families seem to seem to have a higher variability per se as shown before. Within all species, this results in a decreased coefficient of variation as the number of samples increases (Fig. 7). In addition, species-specific differences become visible. For example, giraffes show a constantly low variability in both bacterial families, even when only a few samples are considered. In contrast, wildebeests and plains zebras are more variable when only a small number of samples are taken into account and first stabilize at a sample number of 15 in both analyzed bacterial families. Within carnivores, the tiger samples show a constant CV for all bacterial families from a sample number of n = 10. Even if the variability within the lion samples is higher compared to the tiger ones, they also become stable from a sample number of 10 onwards. Besides species-specific differences we also found differences in the variability between bacterial families in the brown bear. While the pattern for Peptostreptococcaceae and Clostridiaceae is the same as in tigers and lions, the high CV values of the Fusobacteriaceae is not noticeably declining with an increased sample size. Detailed results are shown in the additional file 3 (see Additional file 3).
Microbial indicators for herbivore and carnivore animals
Indicator families were analyzed for each of the four groups and each group combination using the IndVal.g function. We identified a total of 276 indicator families, most of them for herbivores, especially for Perissodactyla (Table 1). With 18 indicator families, Canoidea and Feloidea share less indicators than Perissodactyla and Ruminantia and only minor proportions of indicator families were found in combinations of herbivore and carnivore species. The complete results are presented in the additional file (see Additional file 5).
Table 1
Microbial indicators for different animal groups and their combination. Indicators were assigned at microbial family level.
Group
|
Number of indicator species
|
Canoidea
|
10
|
Feloidea
|
6
|
Perissodactyla
|
43
|
Ruminantia
|
16
|
Canoidea + Feloidea
|
18
|
Perissodactyla + Ruminantia
|
42
|
Canoidea + Perissodactyla
|
1
|
Canoidea + Ruminantia
|
3
|
Feloidea + Perissodactyla
|
2
|
Canoidea + Feloidea + Perissodactyla
|
3
|
Canoidea + Feloidea + Ruminantia
|
6
|
Canoidea + Perissodactyla + Ruminantia
|
4
|
Feloidea + Perissodactyla + Ruminantia
|
2
|
Almost all predicted indicator families show high A values, meaning that this indicator only occurs in the tested group, but is not necessarily spread across all of its members. In contrast, the B values, showing the distribution of an indicator across all group members are much more variable. Indicator families restricted to Canoidea are Gemellaceae (A = 1.00, B = 0.03) and Xiphinematobacteraceae (A = 1.00, B = 0.02), but they do not occur in all of the samples. Regarding the Feloidea, no exclusive indicators were found. However, Coriobacteriaceae (A = 0.88, B = 0.88) are strongly related to this group and distributed among nearly all members. In general, all indicator families associated to the carnivore groups show low B values which might be a further indication of greater diversity within the two groups as seen in the PCoA analysis. However, this view changes when one considers the indicator families that occur in both the Feloidea and the Canoidea. In particular, Enterobacteriaceae (A = 0.98, B = 0.94), Clostridiaceae (A = 0.96, B = 0.95) and Fusobacteriaceae (A = 0.99, B = 0.83) occur in almost all carnivore species and appear to be clear indicator families for those in general. Additionally, these families are also the most dominant ones in the fecal microbiota composition of carnivores (Fig. 1b).
In contrast, more indicator families were found in herbivores. Fibrobacteraceae (A = 0.81, B = 0.97), Synergistaceae (A = 1.00, B = 0.75), Defluviitaleaceae (A = 0.88, B = 0.80) and Methanocorpusculaceae (A = 0.79, B = 0.88) occur almost exclusively in Perissodactyla and are present in almost all species. For ruminant species, one of the most prominent indicators are Barnesiellaceae (A = 0.89, B = 0.72) and Atopobiaceae (A = 0.73, B = 0.46), which occur in many members of this group. Looking at the combined indicators of Perissodactyla and ruminants, many microbial families are found almost exclusively in those two groups and are present in all of their members. Again, those indicator families are among the most dominant ones in the taxonomy plot (Fig. 1b) i.e. Spirochaetaceae (A = 0.99, B = 1.00), Rikenellaceae (A = 0.96, B = 0.99) and Oscillospiraceae (A = 0.87, B = 0.90).