Avian Malaria Parasites Modulate Gut Microbiome Assembly in Canaries

Abstract

Rodent and human malaria parasites cause dysbiosis in the host gut microbiome, but whether Plasmodium species affecting birds cause dysbiosis in their hosts is currently unknown. Here we used a model of avian malaria infection to test whether parasite infection modulates the bird microbiome. To this aim, bird fecal microbiomes were characterized at different time points after infection of canaries with the avian malaria parasite Plasmodium homocircumflexum. Avian malaria caused no significant changes in the alpha and beta diversity of the microbiome in infected birds. In contrast, we discovered changes in the composition and abundance of several taxa. Co-occurrence networks were used to characterize the assembly of the microbiome and trajectories of microbiome structural states progression were found to be different between infected and uninfected birds. Prediction of functional profiles in bacterial communities using PICRUSt2 showed infection by P. homocircumflexum to be associated with the presence of specific degradation and biosynthesis metabolic pathways, which were not found in healthy birds. Some of the metabolic pathways with decreased abundance in the infected group had significant increase in the later stage of infection. The results showed that avian malaria parasites affect bacterial community assembly in the host gut microbiome. Microbiome modulation by malaria parasites could have deleterious consequences for the host bird. Knowing the intricacies of bird-malaria-microbiota interactions may prove helpful in determining key microbial players and informing interventions to improve animal health.

Keywords: avian malaria; microbiome; parasite-microbiota interactions.

Figure 1

Experimental design. Canaries were inoculated with P. homocircumflexum-infected (n = 8) or uninfected (n = 8) donor blood. Blood and fecal samples were collected at different time points as indicated. Created with BioRender.com.

Figure 2

Temporal dynamics of P. homocircumflexum parasitemia. Individual parasitemia values (% of infected erythrocytes) of P. homocircumflexum based on microscopy are presented. * DPI selected for fecal sample collection and microbiome analysis.

Figure 3

Effect of P. homocircumflexum infection on host microbial diversity and taxonomic profiles of bird microbiome. (A) Shannon entropy was used in the longitudinal analysis to compare the differences in alpha diversity between infected and uninfected birds at different time intervals (i.e., from 0 to 8 DPI, from 0 to 16 DPI, from 0 to 24 DPI, from 0 to 36 DPI, and from 0 to 85 DPI). Kruskal–Wallis was used to compare the differences (alpha = 0.05). (B) Similar longitudinal analysis was performed for beta diversity, based on Bray–Curtis distance. Between-groups comparisons were performed using Mann–Whitney U test (alpha = 0.05). (C) Longitudinal feature-volatility analysis of bacterial genera among infected and uninfected birds. The top 20 important features (i.e., from 100 taxa detected by random forest modeling exhibiting important temporal variations in abundance) are shown, including their temporal signal (importance) and net average change. (D) Volcano plot showing the differential bacterial abundance in bird microbiome between control and infected groups throughout the duration of experiment. Taxa with significant differences between the groups are represented with blue (Wald test, p < 0.05) and red (Wald test, p (corrected with Benjamin and Hochberg method) < 0.05) dots. The gray dots represent taxa with no significant differences between groups. Taxa with significant differences in their abundance were identified using DeSeq2 algorithm.

Figure 4

Co-occurrence networks of bird microbiome in the different experimental groups at 8, 16, 24, 36 and 85 DPI. Bacterial co-occurrence networks were inferred from the microbiome of P. homocircumflexum-infected birds and control. Nodes represent bacterial taxa and connecting edges stand for a co-occurrence correlation (SparCC > 0.75). Node sizes are proportional to the eigenvector centrality value. Edges representing positive or negative correlations were colored in lilac and red, respectively. Only nodes with at least one connection are displayed.

Figure 5

Changes in microbiome structural states of P. homocircumflexum-infected and control groups. Scatter plot showing the mean of observed features versus number of (A) connected nodes and (B) edges found in the microbial co-occurrence networks of infected (left) and control (right) birds throughout the course of experiment, from state 1 (8 DPI) to 5 (85 DPI) connected by arrows.

Figure 6

Network tolerance to directed attack. Values of connectivity loss in P. homocircumflexum-infected (red squares) and control (green circles) birds at different days of experiment were compared. At each sampling day a threshold (dashed lines) for connectivity loss was determined where the difference of removed nodes between the infected and control group was the largest.

Figure 7

Impact of P. homocircumflexum infection on the predicted functional profiles of bird gut microbiome. (A) Venn diagram showing the common and different predicted bacterial pathways found in the microbiome of birds infected with P. homocircumflexum and uninfected birds at different sampling days. (B) Volcano plot showing differential pathway abundance in Plasmodium-infected and uninfected birds detected by DESeq2 analysis at 8, 16, 24, 35 and 85 DPI. The blued dots indicate all statistically significant (p(un-adjusted) < 0.05) pathways, red dots indicate pathways with statistically significant (p(adjusted) < 0.05) log2 fold changes in the absolute value (cut-off of 1); the black dots are not significant (p > 0.05). Detailed information on pathway identity is presented in Table S6. (C) Venn diagram. Comparison of unique and shared pathways with significant changes in abundance (p(adjusted) < 0.05) between the Plasmodium-infected and uninfected groups at 24 and 36 DPI. Only pathways with statistically significant log2 fold changes in the absolute value (cut-off of 1) were considered.