Discussion
This study provides evidence that carrion flies are the superior method for landscape scale biodiversity surveillance compared to both camera traps and the other iDNA sampling methods (carrion fly iDNA identified 62 vertebrate species, sandfly iDNA detected 49 species, mosquito iDNA detected 16 species, and camera traps identified 28 species). Previous work has found that iDNA from carrion flies produced complementary results when compared to more traditional methods of biomonitoring such as camera traps (Gogarten et al., 2020; Rodgers et al., 2017). We found a similar pattern to Rodgers et al. (2017) in their comparison of carrion flies to existing camera trap data in that even with a small sample size and effort, carrion flies used as iDNA samplers were able to profile most of the diversity that had been found with long-term camera trapping while also revealing new species diversity. However, with our larger sampling effort, we were able to show that carrion flies were much more effective at describing diversity at the landscape scale and species richness likely would have increased for each iDNA sampler if we had increased our sampling effort, particularly at the site level, whereas species richness had fully saturated by 1,500 camera-days (Fig. 2 and Fig. 4).
While camera traps revealed lower gamma diversity than carrion flies and sandflies, they had the greatest median species richness at sites (alpha diversity) compared to each of the iDNA samplers. This points to camera traps reliably revealing the same suite of species from across sites resulting in higher alpha diversity but lower overall gamma diversity when compared to iDNA samplers. The missing species richness exhibited by camera traps at the landscape scale of our study is likely due to camera traps preferentially targeting larger-bodied mammals (Burton et al., 2015; O’Brien & Kinnaird, 2011). We found support for this as camera traps revealed the greatest diversity of carnivore species and ungulate species while missing most of the iDNA-described diversity of the arboreal species and the smaller-bodied mammals.
The improved biodiversity detection with carrion flies over other iDNA sources and camera traps comes with additional logistical advantages. Carrion flies can be trapped with inexpensive, homemade traps compared to the more expensive camera traps or the UV LED CDC light traps used to capture sandflies and mosquitoes. These homemade traps also did not require batteries compared to the UV LED CDC light traps which require 6V batteries that need to be changed approximately every 24 hours. Our trap design (Appendix S1: Fig. S2) was effective at attracting and trapping carrion flies while at the same time preventing individuals from accessing the bait. Using a blocking primer for the species used as bait is also an option when utilizing carrion flies as an iDNA source. Carrion flies are thus much more amenable to sampling in remote landscapes where transporting and shipping pre-made traps or bulky equipment can be difficult. We also identified the most species with carrion flies compared to the other iDNA samplers despite a much lower sampling effort of 3 traps per site compared to 27 traps per site for mosquitos and sandflies, which resulted in much fewer individual carrion flies (n = 1,759) compared to the total number of sandflies (n = 56,774) and mosquitos (n = 4,776). Carrion fly samples had the greatest amplification efficiency of non-human vertebrates (Table 1) despite only 5 individual flies coextracted compared with 15 mosquitos and 50 sandflies, although amplification efficiency reflected all captured sandflies (not only blood-fed females) and did not include human DNA which was highest in mosquito samples.
Although carrion flies were the most effective iDNA sampler, we wanted to examine whether the gamma and alpha diversity described varied substantially among iDNA sources to assess the effectiveness of sandflies and mosquitos as biodiversity samplers. We found that the overlap in vertebrate communities from each iDNA source was generally supported at the landscape scale, however, the different iDNA sources detected different communities of species at each site, and the same sites were typically distant in ordination space. Although carrion flies were the most effective sampler for detecting gamma diversity, we found carrion flies were less redundant with mosquitos and sandflies for detecting local biodiversity and carrion flies missed 85% and 67% of the species occurrences from across all sites in the sandfly data and in the mosquito data respectively, although most of these misses corresponded to the dominant species in either the sandfly or mosquito data as shown in Appendix S1: Fig. S4. This points to needing a substantially larger sampling effort at the site level for iDNA methods, particularly carrion flies, to saturate species detections.
Aside from the effect of sampling effort on diverging alpha diversity profiles, some of the mismatch is explained by the differing feeding ecologies of each insect group. We detected substantial taxonomic biases in vertebrate detections among iDNA sources that were likely driven by differences in the host preferences of the common mosquito and sandfly species. Aside from domestic dog, we found evidence of contrasting feeding preferences for vector taxa that were consistent with to their known relationships with host species. Sandflies overwhelmingly targeted armadillos (Dasypus novemcinctus and Dasypus kappleri ) which aligns with our understanding of both the relationship between armadillos and sandflies (Akhoundi et al., 2016; Alexander, 2000; Lainson et al., 1979) as well as armadillos as an important host species for the life cycle of Leishmania parasites (Lainson et al., 1979; Lainson & Shaw, 1989), which are transmitted by female sandfly vectors. Kocher et al. (2017) reported the first direct evidence of sandfly-armadillo interaction using metabarcoding of sandflies, and our findings overwhelmingly support this finding. Mosquitos also exhibited strong host preferences for humans, which even after rigorous quality control to eliminate contamination was the dominant iDNA data source representing 80% of the total reads. While many species of mosquito are opportunistic feeders, there are a number of species that target humans including Aedes aegypti which is the main vector for dengue, yellow fever, and chikungunya and is found throughout tropical and sub-tropical regions worldwide (McBride, 2016; Zwiebel & Takken, 2004). Interestingly, our metabarcoding data is corroborated by previous studies spanning the Amazon region that show upwards of 80% of A. aegypti bloodmeals contain human DNA (McBride, 2016). Compared to the targeted feeding behavior of sandflies and mosquitos (as well as leeches shown by Drinkwater et al. 2019), the feeding ecology of carrion flies may offer the least biased view of the diversity and relative biomass of species across a landscape because their feeding behavior is likely driven by both the presence of carrion and scat (Calvignac-Spencer et al., 2013; Rodgers et al., 2017). Given these findings, the feeding ecologies of each insect group should be an important consideration because choice of iDNA sampler can lead to mismatches in the vertebrate diversity profiles, especially with smaller sampling sizes.
By exploring the biases associated with each sampling type, we found evidence that carrion flies are significantly more effective in describing landscape level biodiversity when compared to diversity found with camera traps and other iDNA samplers, even with the least sampling effort and smallest sample size. We found that camera traps predictably were biased towards larger-bodied mammals and away from arboreal and smaller-bodied species, thus missing much of the biodiversity described with iDNA samplers, but also had the highest species richness at the site-level. Sandflies and mosquitos show feeding behavior targeted at armadillos and humans, respectively. Although sandflies and mosquitos were inferior samplers for large-scale biodiversity surveillance compared to carrion flies, they add value to existing host-vector-pathogen surveillance efforts. This is particularly relevant in forest frontiers where vectors maintain enzootic sylvatic cycles which can cause disease spillover into humans and/or domestic animals if they are bitten by these infected vectors (Figueiredo, 2007; Vasconcelos et al., 2001). Additionally, our findings support the known ecological relationships between these vector taxa and their hosts thus further validating metabarcoding results within the larger realm of ecological research. With the advances in high-throughput sequencing and the steadily decreasing expense of metabarcoding, iDNA should be used as an effective and efficient method for biomonitoring.