Island biogeography theory applied to the microbiomes of parasites
Habitat fragment area and protection status, but not degree of isolation or distance to source, have a measurable but lesser effect on the microbiome of bat flies compared to parasite species (Tables 2 and 3). As bat flies travel with their host bat, it is not surprising that isolation and distance to source do not explain bat fly microbiome variation in these narrowly-separated habitat patches. These stretches of inhospitable agricultural landscape between forest patches may be more of a barrier for parasites of other vertebrates that cannot fly. Mean and median bacterial ASV richness does not decrease with decreasing habitat fragment area, increasing isolation, and increasing distance from a source, contradicting expectations of the island biogeography model (Figure 5). Given the host-specificity of the bat fly microbiome, it is unrealistic to expect ASV richness to change with environment, especially if bacterial ASVs in a bat fly are selected for or maternally transmitted. Instead, variation of the microbiome in response to environment may be better reflected by variation in relative abundance of ASVs. Examining both relative abundance and diversity of bacteria using ecological interaction networks provided a clear statistical signal that habitat fragment area impacts microbiome composition (Figures 6 and 7).
Habitat fragmentation resulting in decreasing habitat patch area was associated with the loss of nodes that act as bridges between modules in bacterial interaction networks (Figures 6, 7A). There were fewer ASVs with high betweenness centrality in small fragments and more ASVs with small betweenness centrality, consistent with the loss of connectivity between disparate portions of the network (Figure 6). Large fragments had lower raw modularity scores than small fragments, indicating more connectivity between neighborhoods than in small fragments (Figure 7A). ASVs that acted as bridge nodes in large sites, are present in smaller sites (Figure 5), but these ASVs no longer perform the function of connecting modules in the network in smaller fragments.
Site F4 is an outlier to this pattern in that it is intermediate in area, isolation, and distance to a source, but has similar modularity and betweenness centrality to the sites within REGUA and the largest site (F10). This is likely driven by the high sample size at F4 compared to other fragments of similar area, isolation, and distance to a source. While we standardized the field collecting method across sampling sites, we sequenced bat fly microbiomes only from the most prevalent bat fly species, corresponding to the most prevalent bat species, across all of the sampling sites, which may have introduced sampling bias into measures of network size. Generally, sites with higher parasite microbiome sample sizes are also the sites with a higher number of captured bat fly species. Considering only the bat fly species that were selected for sequencing, fragments F10 (the largest fragment outside of REGUA, 9 bat fly species) and F4 (8 bat fly species) have the greatest parasite richness outside of REGUA. The three REGUA sites, F10, and F4 are the only sampling sites where the bat species Artibeus lituratus and Desmodus rotundus and their respectively associated bat flies Paratrichobius longicrus and Strebla wiedemanni were collected; all other sites are missing one or more of these species. Bat species and bat fly species responses to habitat disturbance are highly species specific, and may not always be negative (Hiller et al., 2020; Pilosof et al., 2012; Saldaña-Vázquez et al., 2013). Anecdotally, site F4 had many large trees, suitable for roosts, compared to other patches of similar size. It is possible that factors of the habitat at F4 which we did not evaluate, such as availability of permanent tree roosts (Patterson et al. 2007) or access to particular foraging resources (Pilosof et al. 2012), may mediate the effects of habitat fragment size in some cases, allowing relatively smaller fragments to support more bat species, and hence more parasite species, which in turn may affect structure and connectivity of parasite microbiomes.
The loss of bridge nodes as sampling sites decrease in area and parasite species richness is driven by network size (i.e., number of interactions), which corresponds to sampling size at each site (Figures 6 and 7). When we attempted to control for variation in sample size by using subsampling, SPIEC-EASI networks built from subsampling sites with large sample sizes never reached stability. Stability is assessed by subsampling, so that when we provided SPIEC-EASI with a reduced dataset, it was not able to infer the more complex networks from the REGUA sites, F10, and F4. This may indicate that higher network size in these sites is not an artifact of sample size. PCoA between correlations of orbit incidences were used as a size-independent examination of the topology of each network. There is no evidence that the environment alters the underlying topology of the network, but it may drive network size through increasing parasite richness (Figures 7). Regardless, sample size cannot be controlled for when examining network size, so the correlation may be spurious.
The finding that large fragments have higher null-adjusted modularity scores than small fragments may appear contradictory to our conclusion that bridge nodes are absent from fragmented habitats (Figure 7B). However, this pattern is consistent with our finding that habitat fragmentation causes changes in network size and not topology. Randomizations of edges to create null distributions preserves the degree distribution of the measured network, so measured networks with many nodes of high degree will have null networks that have low modularity. Large fragments have many ASVs with high degree (i.e., many connections) and these are mostly distinct from ASVs with high betweenness centrality (i.e., more shortest paths between nodes traverse through these ASVs). The ASVs with high degree form dense subgraphs rather than nodes connecting many modules in large, near to source fragments, leading to greater disparity in modularity between the null and measured networks than in small, distant fragments. More samples in a site lead to more detected connections (i.e., larger network size), but do not change the shape of the network by detecting more connections outside of densely connected neighborhoods.