Specific gut bacterial responses to natural diets of tropical birds

Natural diets of tropical birds vary within species

We collected 62 regurgitated samples (using the tartar emetic method ²²) from multiple tropical bird species representing four bird orders (Columbiformes–Pigeons, Coraciiformes–Kingfishers, Psittaciformes–Parrots, and Passeriformes–Passerines). First, we characterized diet components visually and then through metabarcoding of 52 of these samples using universal primers targeting invertebrates (Cytochrome c oxidase subunit I: COI gene) and plants (Internal transcribed spacer 2: ITS2 gene) (Table S1 and Fig. 2). Through visual identification, we identified plant material in 26 samples. The most common visually identified invertebrate orders were Araneae (spiders—27 samples), and Coleoptera (beetles—27 samples) (Table S2). Metabarcoding sequences were analysed using the OBITools software²⁵. Overall, we found 47 plant operational taxonomic units (OTUs—97% sequence similarity threshold) and 180 invertebrate OTUs (Table S3). Plant items were dominated by the orders Rosales (27.7% OTUs), Fabales (8.5% OTUs), and Sapindales (8.5% OTUs). Except for four OTUs, all plants were identified to the genus level. Of the invertebrate OTUs, 54 belonged to feather mites (known feather symbionts), endoparasites, and rotifers (likely due to accidental consumption along with drinking water), and these OTUs were removed from further analyses, leaving 126 potential dietary invertebrate OTUs. Invertebrate samples were dominated by the classes Insecta (67.5% OTUs) and Arachnida (28.6% OTUs). At the order-level, dietary items were mainly represented by Araneae (spiders—28.6% OTUs), Hemiptera (true bugs—15.9% OTUs), Diptera (flies—14.3% OTUs), and Lepidoptera (moths and butterflies—10.3% OTUs). However, 77% of the invertebrate OTUs could not be identified to genus level, highlighting the limited research on genotyping invertebrate communities in Papua New Guinea.

Diet item identification differed markedly between visual and metabarcoding methods (Fig. 2, Tables S2 and S3). The diet components of individuals also varied notably within species (Figs. 2 and S1). Only diets of 12 out of 52 individuals were fully congruent between the two methods (Fig. 2). Of these 12 samples, eight had only plant material. Identification of invertebrate orders also differed between the two methods (Fig. 2, Table 1). Both methods identified the arthropod orders Hemiptera, Diptera, Orthoptera (crickets and locusts), and Araneae in the same samples (Fig. 2 and Table 1), while metabarcoding detected lower proportions of Coleoptera than the visual identification (Table 1).

Comparison of microbiomes and consumed diet items

For subsequent comparisons of diets and microbiomes, we utilised individual datasets from both visual identification (diet components identified at the order level) and metabarcoding (both OTU and order level), and a combination (order level) of both approaches (for details see “Methods” section on identifying prey items). Due to differences between the diet identification methods, a combination of the results was used to circumscribe the full diversity of consumed diets and to account for inherent biases associated with the two methods (i.e., the inability to identify plant material and smaller body parts of invertebrates visually, and extraction and sequencing biases associated with metabarcoding). We separated the microbiome dataset into three datasets due to sequencing limitations: dataset 1 included 12 birds with successfully sequenced crop microbiomes and diets identified using both methods, dataset 2 included 27 birds with successfully sequenced cloacal microbiomes and diets, and dataset 3 included 17 birds for which we obtained successfully sequenced crop and cloacal microbiomes (Table S1). Prior to subsequent analyses, each microbiome dataset was rarefied to even sequencing depths using the sample with the lowest number of sequences²⁶ (Fig. S2).

Crop microbiome similarity did not align with the consumed diet similarity (dataset 1)

Out of the collected crop samples (N = 62), samples from only 19 individuals were successfully sequenced for their microbiomes. Of these individuals, we acquired diet samples for 12 individuals. Bacterial 16S rRNA MiSeq sequences were analysed using the DADA2 pipeline²⁷ within QIIME2²⁸. There were 351,867 bacterial sequences (mean ± SD: 29,322 ± 33,009) in the crop microbiomes prior to rarefaction (Table S4). After rarefaction, bacterial sequences were identified to 615 amplicon sequence variants (ASVs—100% sequence similarity). Crop microbiomes were dominated by Proteobacteria (53.6%), Actinobacteria (18.9%), and Firmicutes (17.9%). Alpha diversities of individual microbiomes were calculated using the diversity function in the microbiome package²⁹ and they did not differ significantly between host orders [Chao1 richness: Kruskal Wallis (KW) χ² = 4.559, df = 3, p = 0.2271; Shannon’s diversity index: χ² = 2.853, df = 3, p = 0.4149], or literature-based dietary guilds (Chao1 richness: KW χ² = 4.317, df = 2, p = 0.1155; Shannon’s diversity index: KW χ² = 2.852, df = 2, p = 0.2403) (Fig. S3).

The compositional differences of crop microbiomes were investigated with the adonis2 function in the vegan package³⁰ using permutational multivariate analyses of variance tests (PERMANOVA). These analyses revealed that the bird host order did not influence the crop microbiome composition (PERMANOVA_{10,000 permutations}: Bray–Curtis: F = 1.251, R² = 0.0993, p = 0.1911; Jaccard: F = 1.154, R² = 0.0962, p = 0.2191) (Fig. S1). The effect of feeding guild was masked by host order as they are strongly correlated in this dataset. Furthermore, the lack of an effect of host taxa on crop microbiomes may be a result of the small sample sizes.

We further investigated whether alpha diversity of the crop microbiomes was influenced by the diet item diversity of individuals. The Chao1 richness estimates of the microbiomes and the richness of the consumed diet items (number of different diet items based on the combined results) of individuals were not significantly correlated (Table S5), suggesting that the diet richness does not impact crop microbiome richness. However, Shannon’s diversity index of crop microbiomes and diet diversity were marginally significantly negatively associated (Table S5). This suggests that despite the lack of an association between diet and microbiome richness, crop microbiome evenness could be influenced by diet diversity.

We then explored the association between the crop microbiome composition and the consumed diets, investigating correlations between Bray–Curtis and Jaccard dissimilarities of microbiomes, and Jaccard dissimilarity of diets using Mantel tests in the vegan package³⁰. The compositional similarity of the diets based on any of the methods (visual, metabarcoding—both OTU and order-level separately, and combined) did not correlate significantly with crop microbiome compositions (Table 2 and Fig. S4). We observed similar non-significant associations between diets and microbiomes when investigating host orders separately (Table S6). This suggests that overall crop microbiomes of individuals are not completely modelled by the composition of the consumed diets.

Host-taxon specific cloacal microbes are associated with different diet items (dataset 2)

We obtained 27 individuals from 15 bird species with successfully sequenced cloacal microbiomes and diet samples (based on both metabarcoding and visual identification). Prior to rarefying, we acquired 818,272 bacterial sequences from the cloacal swab samples (mean ± SD: 30,306 ± 20,903) (Table S7). After rarefaction, bacterial sequences were assigned to 1,324 ASVs that belonged to Actinobacteria (35.9%), Proteobacteria (32.6%), Firmicutes (21.2%) and Tenericutes (5.0%). Cloacal microbiome alpha diversity did not differ significantly between different bird orders (Chao1 richness: KW χ² = 2.624, df = 3, p = 0.4532; Shannon’s diversity: χ² = 6.595, df = 3, p = 0.0861) or literature-based dietary guilds (Chao1 richness: KW χ² = 1.128, df = 3, p = 0.7703; Shannon’s diversity: KW χ² = 1.673, df = 3, p = 0.6429) (Fig. S5).

However, cloacal microbiome beta diversity was significantly influenced by host bird order (PERMANOVA_{10,000 permutations}: Bray–Curtis: F = 2.159, R² = 0.2055, p < 0.0001; Jaccard: F = 1.749, R² = 0.1775, p < 0.0001) and literature-based dietary guilds (PERMANOVA_{10,000 permutations}: Bray–Curtis: F = 1.529, R² = 0.1456, p = 0.0008; Jaccard: F = 1.341, R² = 0.1361, p = 0.0023) (Fig. S1). The variation explained by dietary guilds was slightly secondary to the variation explained by the host taxon. We did not observe significant associations between alpha diversity of diet and cloacal microbiomes (Table S5) nor in compositional similarity of diets and cloacal microbiomes (Table 2 and Fig. 3). Similar to crop microbiomes, the compositional similarity of diets and cloacal microbiomes were not significantly associated when bird orders were analysed separately (Table S6). This suggests that cloacal microbiome composition does not align with the overall consumed diet similarities of hosts.

To explore whether specific gut bacterial symbionts of different host taxa are associated with different dietary items (Fig. 1b), we tested for correlations between the 30 most abundant bacterial genera and the proportion of order-level diet items in each individual using the taxa.env.correlation function in the microbiomeSeq package³¹. These analyses revealed that certain bacterial genera were positively correlated with certain dietary items (Fig. 4 and Table S8) and that the taxonomy of bacterial symbionts associated with the same diet item differed between host orders, suggesting that host-taxon specific microbes are affected by the same dietary items in different bird taxa. For example, the relative abundance of the plant order Rosales was significantly correlated with the bacterial genera Ureaplasma, and Helicobacter in pigeons, while Rosales was significantly associated with Helicobacter, Escherichia and Acinetobacter in passerine birds. These results indicate that the overall effect of diet on cloacal microbiomes results from a combination of associations between certain microbes and specific dietary items in different avian hosts.

A large portion of bacterial sequences are shared between the crop and the cloaca (dataset 3)

For 17 bird individuals, we successfully acquired both crop and cloacal microbiomes. Overall, prior to rarefaction, we acquired 571,488 (mean ± SD: 33,617 ± 17,188) bacterial sequences from cloacal swab samples and 562,557 sequences from crop samples (mean ± SD: 33,091 ± 35,586). After rarefaction, sequences aligned to 1,176 bacterial ASVs (Table S9). The microbiome alpha diversity did not differ significantly between the two regions (Chao1 richness: KW χ² = 0.3633, df = 1, p = 0.5466; Shannon’s diversity index: KW χ² = 1.759, df = 1, p = 0.1848) (Fig. 5a,b). Overall, the phylum Proteobacteria (43.9%) dominated the crop microbiomes, followed by Firmicutes (20%) and Actinobacteria (19%). In the cloaca, microbiomes were dominated by Actinobacteria (40.2%), followed by Proteobacteria (25.1%) and Firmicutes (21.7%) (Fig. 6). The relative abundance of the bacterial phyla in both regions of the digestive tract differed markedly between bird species but we observed comparable microbial compositions within species (Fig. 6). Overall, bacterial community compositions did not differ significantly between the two regions of the gut (PERMANOVA_{10,000 permutations}: Bray–Curtis: F = 0.9188, R² = 0.0279, p = 0.5985; Jaccard: F = 0.8995, R² = 0.0273, p = 0.6825) (Fig. 5c), while both the crop (PERMANOVA_{10,000 permutations}: Bray–Curtis: F = 1.661, R² = 0.2771, p = 0.0026; Jaccard: F = 1.393, R² = 0.2432, p = 0.0051) and the cloacal (PERMANOVA_{10,000 permutations}: Bray–Curtis: F = 1.721, R² = 0.2841, p < 0.0001; Jaccard: F = 1.521, R² = 0.2597, p = 0.0006) microbiomes were significantly affected by host order (Fig. 5c). The similarity between the crop and the cloacal microbiomes within individuals indicates that microbiomes are likely to be influenced by the same factors, e.g., host taxon.

Only a small number of ASVs (richness) was shared between the crop and the cloacal microbiomes (mean ± SD: 17.1% ± 12.2%). However, they accounted for a large proportion of the total number of bacterial sequences (61.5% ± 32.3%) (Fig. 6). This suggests that the most abundant ASVs are shared between the crop and the cloacal microbiomes. We further explored the association of richness of these shared ASVs and their relative abundances with the gut length of hosts. We did not find a significant association between host body mass (a proxy for gut length¹⁰) and the number of shared ASVs between the two regions (lm: R² = 0.0312, F = 0.4835, p = 0.4975), indicating that body size did not influence ASV sharing between the two regions. We also did not find a significant association between the combined relative abundance (the proportion of bacterial sequences) of the shared ASVs and body mass (lm: R² = 0.1631, F = 2.921, p = 0.1081; Fig. S6). However, the relationship between host body mass and relative abundance of shared ASVs tended to be negatively associated, suggesting that larger birds (with longer digestive tracts) share fewer bacterial sequences between the crop and the cloaca.

Source: Ecology - nature.com